In a first, Google has released data on how much energy an AI prompt uses

Google has just released a technical report detailing how much energy its Gemini apps use for each query. In total, the median prompt—one that falls in the middle of the range of energy demand—consumes 0.24 watt-hours of electricity, the equivalent of running a standard microwave for about one second. The company also provided average estimates for the water consumption and carbon emissions associated with a text prompt to Gemini.

It’s the most transparent estimate yet from a Big Tech company with a popular AI product, and the report includes detailed information about how the company calculated its final estimate. As AI has become more widely adopted, there’s been a growing effort to understand its energy use. But public efforts attempting to directly measure the energy used by AI have been hampered by a lack of full access to the operations of a major tech company. 

Earlier this year, MIT Technology Review published a comprehensive series on AI and energy, at which time none of the major AI companies would reveal their per-prompt energy usage. Google’s new publication, at last, allows for a peek behind the curtain that researchers and analysts have long hoped for.

The study focuses on a broad look at energy demand, including not only the power used by the AI chips that run models but also by all the other infrastructure needed to support that hardware. 

“We wanted to be quite comprehensive in all the things we included,” said Jeff Dean, Google’s chief scientist, in an exclusive interview with MIT Technology Review about the new report.

That’s significant, because in this measurement, the AI chips—in this case, Google’s custom TPUs, the company’s proprietary equivalent of GPUs—account for just 58% of the total electricity demand of 0.24 watt-hours. 

Another large portion of the energy is used by equipment needed to support AI-specific hardware: The host machine’s CPU and memory account for another 25% of the total energy used. There’s also backup equipment needed in case something fails—these idle machines account for 10% of the total. The final 8% is from overhead associated with running a data center, including cooling and power conversion. 

This sort of report shows the value of industry input to energy and AI research, says Mosharaf Chowdhury, a professor at the University of Michigan and one of the heads of the ML.Energy leaderboard, which tracks energy consumption of AI models. 

Estimates like Google’s are generally something that only companies can produce, because they run at a larger scale than researchers are able to and have access to behind-the-scenes information. “I think this will be a keystone piece in the AI energy field,” says Jae-Won Chung, a PhD candidate at the University of Michigan and another leader of the ML.Energy effort. “It’s the most comprehensive analysis so far.”

Google’s figure, however, is not representative of all queries submitted to Gemini: The company handles a huge variety of requests, and this estimate is calculated from a median energy demand, one that falls in the middle of the range of possible queries.

So some Gemini prompts use much more energy than this: Dean gives the example of feeding dozens of books into Gemini and asking it to produce a detailed synopsis of their content. “That’s the kind of thing that will probably take more energy than the median prompt,” Dean says. Using a reasoning model could also have a higher associated energy demand because these models take more steps before producing an answer.

This report was also strictly limited to text prompts, so it doesn’t represent what’s needed to generate an image or a video. (Other analyses, including one in MIT Technology Review’s Power Hungry series earlier this year, show that these tasks can require much more energy.)

The report also finds that the total energy used to field a Gemini query has fallen dramatically over time. The median Gemini prompt used 33 times more energy in May 2024 than it did in May 2025, according to Google. The company points to advancements in its models and other software optimizations for the improvements.  

Google also estimates the greenhouse gas emissions associated with the median prompt, which they put at 0.03 grams of carbon dioxide. To get to this number, the company multiplied the total energy used to respond to a prompt by the average emissions per unit of electricity.

Rather than using an emissions estimate based on the US grid average, or the average of the grids where Google operates, the company instead uses a market-based estimate, which takes into account electricity purchases that the company makes from clean energy projects. The company has signed agreements to buy over 22 gigawatts of power from sources including solar, wind, geothermal, and advanced nuclear projects since 2010. Because of those purchases, Google’s emissions per unit of electricity on paper are roughly one-third of those on the average grid where it operates.

AI data centers also consume water for cooling, and Google estimates that each prompt consumes 0.26 milliliters of water, or about five drops. 

The goal of this work was to provide users a window into the energy use of their interactions with AI, Dean says. 

“People are using [AI tools] for all kinds of things, and they shouldn’t have major concerns about the energy usage or the water usage of Gemini models, because in our actual measurements, what we were able to show was that it’s actually equivalent to things you do without even thinking about it on a daily basis,” he says, “like watching a few seconds of TV or consuming five drops of water.”

The publication greatly expands what’s known about AI’s resource usage. It follows recent increasing pressure on companies to release more information about the energy toll of the technology. “I’m really happy that they put this out,” says Sasha Luccioni, an AI and climate researcher at Hugging Face. “People want to know what the cost is.”

This estimate and the supporting report contain more public information than has been available before, and it’s helpful to get more information about AI use in real life, at scale, by a major company, Luccioni adds. However, there are still details that the company isn’t sharing in this report. One major question mark is the total number of queries that Gemini gets each day, which would allow estimates of the AI tool’s total energy demand. 

And ultimately, it’s still the company deciding what details to share, and when and how. “We’ve been trying to push for a standardized AI energy score,” Luccioni says, a standard for AI similar to the Energy Star rating for appliances. “This is not a replacement or proxy for standardized comparisons.”

I gave the police access to my DNA—and maybe some of yours

Last year, I added my DNA profile to a private genealogical database, FamilyTreeDNA, and clicked “Yes” to allow the police to search my genes.

In 2018, police in California announced they’d caught the Golden State Killer, a man who had eluded capture for decades. They did it by uploading crime-scene DNA to websites like the one I’d joined, where genealogy hobbyists share genetic profiles to find relatives and explore ancestry. Once the police had “matches” to a few relatives of the killer, they built a large family tree from which they plucked the likely suspect.

This process, called forensic investigative genetic genealogy, or FIGG, has since helped solve hundreds of murders and sexual assaults. Still, while the technology is potent, it’s incompletely realized. It operates via a mishmash of private labs and unregulated websites, like FamilyTree, which give users a choice to opt into or out of police searches. The number of profiles available for search by police hovers around 1.5 million, not yet enough to find matches in all cases.

To do my bit to increase those numbers, I traveled to Springfield, Massachusetts.

The staff of the local district attorney, Anthony D. Gulluni, was giving away free FamilyTree tests at a minor-league hockey game in an effort to widen its DNA net and help solve several cold-case murders. After glancing over a consent form, I spit into a tube and handed it back. According to the promotional material from Gulluni’s office, I’d “become a hero.”

But I wasn’t really driven by some urge to capture distantly related serial killers. Rather, my spit had a less gallant and more quarrelsome motive: to troll privacy advocates whose fears around DNA I think are overblown and unhelpful. By giving up my saliva for inspection, I was going against the view that a person’s DNA is the individualized, sacred text that privacy advocates sometimes claim.

Indeed, the only reason FIGG works is that relatives share DNA: You share about 50% with a parent, 25% with a grandparent, about 12.5% with a first cousin, and so on. When I got my FamilyTree report back, my DNA had “matched” with 3,309 people.

Some people are frightened by FIGG or reject its punitive aims. One European genealogist I know says her DNA is kept private because she opposes the death penalty and doesn’t want to risk aiding US authorities in cases where lethal injection might be applied. But if enough people share their DNA, conscientious objectors won’t matter. Scientists estimate that a database including 2% of the US population, or 6 million people, could identify the source of nearly any crime-scene DNA, given how many distant relatives each of us has.

Scholars of big data have termed this phenomenon “tyranny of the minority.” One person’s voluntary disclosure can end up exposing the same information about many others. And that tyranny can be abused.

DNA information held in private genealogy websites like FamilyTree is lightly guarded by terms of service. These agreements have flip-flopped over time; at one point all users were included in law enforcement searches by default. Rules are easily ignored, too. Recent court filings indicate that the FBI, in its zeal to solve crimes, sometimes barges past restrictions to look for matches in databases whose policies exclude police.

“Noble aims; no rules” is how one genetic genealogist described the overall situation in her field.

My uncertainty grew the more questions I asked. Who even controls my DNA file? That’s not easy to find out. FamilyTree is a brand operated by another company, Gene by Gene, which in 2021 was sold to a third company, MyDNA—ultimately owned by an Australian mogul whose name appears nowhere on its website. When I reached FamilyTree’s general manager, the genealogist Dave Vance, he told me that three-quarters of the profiles on the site were “opted in” to law enforcement searches.

One solution holds that the federal government should organize its own national DNA database for FIGG. But that would require new laws, new technical standards, and a debate about how our society wants to employ this type of big data—not just getting individual consent like mine. No such national project—or consensus—exists.

I’m still ready to join a national crime-fighting database, but I regret doing it the way I did—spitting in a tube on the sidelines of a hockey game and signing a consent form that affects not just me but all my thousands of genetic relatives. To them, I say: Whoops. Your DNA; my bad.

This article first appeared in The Checkup, MIT Technology Review’s weekly biotech newsletter. To receive it in your inbox every Thursday, and read articles like this first, sign up here.

The case against humans in space

Elon Musk and Jeff Bezos are bitter rivals in the commercial space race, but they agree on one thing: Settling space is an existential imperative. Space is the place. The final frontier. It is our human destiny to transcend our home world and expand our civilization to extraterrestrial vistas.

This belief has been mainstream for decades, but its rise has been positively meteoric in this new gilded age of astropreneurs. Expanding humanity beyond Earth is both our birthright and our duty to the future, they insist. Failing to do so would consign our species to certain extinction—either by our own hand, perhaps through nuclear war or climate change, or in some cosmic disaster, like a massive asteroid impact.

But as visions of giant orbital stations and Martian cities dance in our heads, a case against human space colonization has found its footing in a number of recent books. The argument grows from many grounds: Doubts about the practical feasibility of off-Earth communities. Concerns about the exorbitant costs, including who would bear them and who would profit. Realism about the harsh environment of space and the enormous tax it would exact on the human body. Suspicion of the underlying ideologies and mythologies that animate the race to settle space.

And, more bluntly, a recognition that “space sucks” and a lot of people have “underestimated the scale of suckitude,” as Kelly and Zach Weinersmith put it in their book A City on Mars: Can We Settle Space, Should We Settle Space, and Have We Really Thought This Through?, which was released in paperback earlier this year.

cover of A City on Mars
A City on Mars: Can We Settle Space, Should
We Settle Space, and Have We Really Thought This Through?

Kelly and Zach Weinersmith
PENGUIN RANDOM HOUSE, 2023 (PAPERBACK RELEASE 2025)

The Weinersmiths, a husband-wife team, spent years thinking it through—in delightfully pragmatic detail. A City on Mars provides ground truth for our lofty celestial dreams by gaming out the medical, technical, legal, ethical, and existential consequences of space settlements. 

Much to the authors’ own dismay, the result is a grotesquery of possible outcomes including (but not limited to) Martian eugenics, interplanetary war, and—­memorably—“space cannibalism.” 

The Weinersmiths puncture the gauzy fantasy of space cities by asking pretty basic questions, like how to populate them. Astronauts experience all kinds of medical challenges in space, such as radiation exposure and bone loss, which would increase risks to both parents and babies. Nobody wants their pregnant “glow” to be a by-product of cosmic radiation.

Trying to bring forth babies in space “is going to be tricky business, not just in terms of science, but from the perspective of scientific ethics,” they write. “Adults can consent to being in experiments. Babies can’t.”

You don’t even have to contemplate going to Mars to make some version of this case. In Ground Control: An Argument for the End of Human Space Exploration, Savannah Mandel chronicles how past and present generations have regarded human spaceflight as an affront to vulnerable children right here on Earth.

cover of Ground Control
Ground Control: An Argument for the End of Human Space Exploration
Savannah Mandel
CHICAGO REVIEW PRESS, 2024

“Hungry Kids Can’t Eat Moon Rocks,” read signs at a protest outside Kennedy Space Center on the eve of the Apollo 11 launch in July 1969. Gil Scott-Heron’s 1970 poem “Whitey on the Moon” rose to become the de facto anthem of this movement, which insists, to this day, that until humans get our earthly house in order, we have no business building new ones in outer space.

Ground Control, part memoir and part manifesto, channels this lament: How can we justify the enormous cost of sending people beyond our planet when there is so much suffering here at home? 

Advocates for human space exploration reject the zero-sum framing and point to the many downstream benefits of human spaceflight. Space exploration has catalyzed inventions from the CAT scan to baby formula. There is also inherent value in our shared adventure of learning about the vast cosmos.

Those upsides are real, but they are not remotely well distributed. Mandel predicts that the commercial space sector in its current form will only exacerbate inequalities on Earth, as profits from space ventures flow into the coffers of the already obscenely rich. 

In her book, Mandel, a space anthropologist and scholar at Virginia Tech, describes a personal transformation from spacey dreamer to grounded critic. It began during fieldwork at Spaceport America, a commercial launch facility in New Mexico, where she began to see cracks in the dazzling future imagined by space billionaires. As her career took her from street protests in London to extravagant space industry banquets in Washington, DC, she writes, “crystal clear glasses” replaced “the rose-colored ones.”

Mandel remains enchanted by space but is skeptical that humans are the optimal trailblazers. Robots, rovers, probes, and other artificial space ambassadors could do the job for a fraction of the price and without risk to life, limb, and other corporeal vulnerabilities.  

“A decentralization of self needs to occur,” she writes. “A dissolution of anthropocentrism, so to speak. And a recognition that future space explorers may not be man, even if man moves through them.” 

In other words, giant leaps for mankind no longer necessitate a man’s small steps; the wheels of a rover or the rotors of a copter offer a much better bang for our buck than boots on the ground.

In contrast to the Weinersmiths, Mandel devotes little attention to the physical dangers and limitations that space imposes on humans. She is more interested in a kind of psychic sickness that drives the impulse to abandon our planet and rush into new territories.

Mary-Jane Rubenstein, a scholar of religion at Wesleyan University, presents a thorough diagnosis of this exact pathology in her 2022 book Astrotopia: The Dangerous Religion of the Corporate Space Race, which came out in paperback last year. It all begins, appropriately enough, with the book of Genesis, where God creates Earth for the dominion of man. Over the years, this biblical brain worm has offered divine justification for the brutal colonization and environmental exploitation of our planet. Now it serves as the religious rocket fuel propelling humans into the next frontier, Rubenstein argues.

cover of Astrotopia
Astrotopia: The Dangerous Religion of the Corporate Space Race
Mary-Jane Rubenstein
UNIVERSITY OF CHICAGO PRESS, 2022  (PAPERBACK RELEASE 2024)

“The intensifying ‘NewSpace race’ is as much a mythological project as it is a political, economic, or scientific one,” she writes. “It’s a mythology, in fact, that holds all these other efforts together, giving them an aura of duty, grandeur, and benevolence.”

Rubenstein makes a forceful case that malignant outgrowths of Christian ideas scaffold the dreams of space settlements championed by Musk, Bezos, and like-minded enthusiasts—even if these same people might never describe themselves as religious. If Earth is man’s dominion, space is the next logical step. Earth is just a temporary staging ground for a greater destiny; we will find our deliverance in the heavens.   

“Fuck Earth,” Elon Musk said in 2014. “Who cares about Earth? If we can establish a Mars colony, we can almost certainly colonize the whole solar system.”

Jeff Bezos, for one, claims to care about Earth; that’s among his best arguments for why humans should move beyond it. If heavy industries and large civilian populations cast off into the orbital expanse, our home world can be, in his words, “zoned residential and light industry,” allowing it to recover from anthropogenic pressures.

Bezos also believes that space settlements are essential for the betterment of humanity, in part on the grounds that they will uncork our population growth. He envisions an orbital archipelago of stations, sprawled across the solar system, that could support a collective population of a trillion people. “That’s a thousand Mozarts. A thousand Einsteins,” Bezos has mused. “What a cool civilization that would be.”

It does sound cool. But it’s an easy layup for Rubenstein: This “numbers game” approach would also produce a thousand Hitlers and Stalins, she writes. 

And that is the real crux of the argument against pushing hard torapidly expand human civilization into space: We will still be humans when we get there. We won’t escape our vices and frailties by leaving Earth—in fact, we may exacerbate them. 

While all three books push back on the existential argument for space settlements, the Weinersmiths take the rebuttal one step further by proposing that space colonization might actually increase the risk of self-annihilation rather than neutralizing it.

“Going to space will not end war because war isn’t caused by anything that space travel is apt to change, even in the most optimistic scenarios,” they write. “Humanity going to space en masse probably won’t reduce the likelihood of war, but we should consider that it might increase the chance of war being horrific.” 

The pair imagine rival space nations exchanging asteroid fire or poisoning whole biospheres. Proponents of space settlements often point to the fate of the dinosaurs as motivational grist, but what if a doomsday asteroid were deliberately flung between human cultures as a weapon? It may sound outlandish, but it’s no more speculative than a floating civilization with a thousand Mozarts. It follows the same logic of extrapolating our human future in space from our behavior on Earth in the past.

So should we just sit around and wait for our inevitable extinction? The three books have more or less the same response: What’s the rush? It is far more likely that humanity will be wiped out by our own activity in the near term than by any kind of cosmic threat. Worrying about the expansion of the sun in billions of years, as Musk has openly done, is frankly hysterical. 

In the meantime, we have some growing up to do. Mandel and Rubenstein both argue that any worthy human future in space must adopt a decolonizing approach that emphasizes caretaking and stewardship of this planet and its inhabitants before we set off for others. They draw inspiration from science fiction, popular culture, and Indigenous knowledge, among other sources, to sketch out these alternative visions of an off-Earth future. 

Mandel sees hope for this future in post-scarcity political theories. She cites various attempts to anticipate the needs of future generations—ideas found in the work of the social theorist Aaron Benanav, or in the values expressed by the Green New Deal, or in the fictional Ministry for the Future imagined by Kim Stanley Robinson in his 2020 novel of the same name. Whatever you think of the controversial 2025 book Abundance, by Ezra Klein and Derek Thompson, it is also appealing to the same demand for a post-scarcity road map.  

To that end, Mandel envisions “the creation of a governing body that would require that techno-scientific plans, especially those with a global reach, take into consideration multigenerational impacts and multigenerational voices.”  

For Rubenstein, religion is the poison, but it may also offer the cure. She sees potential in a revival of pantheism, which is the belief that all the contents of the universe—from rocks to humans to galaxies—are divine and perhaps alive on some level. She hasn’t fully converted herself to this movement, let alone become an evangelist, but she says it’s a spiritual direction that could be an effective counterweight to dominionist views of the universe.

“It doesn’t matter whether … any sort of pantheism is ‘true,’” she writes. “What matters is the way any given mythology prompts us to interact with the world we’re a part of—the world each of our actions helps to make and unmake. And frankly, some mythologies prompt us to act better than others.”

All these authors ultimately conclude that it would be great if humans lived in space—someday, if and when we’ve matured. But the three books all express concerns about efforts by commercial space companies, with the help of the US government, to bypass established space laws and norms—concerns that have been thoroughly validated in 2025.  

The combustible relationship between Elon Musk and Donald Trump has raised eyebrows about cronyism—and retribution—between governments and space companies. Space is rapidly becoming weaponized. And recent events have reminded us of the immense challenges of human spaceflight. SpaceX’s next-­generation Starship vehicle has suffered catastrophic failures in several test flights, while Boeing’s Starliner capsule experienced malfunctions that kept two astronauts on the International Space Station for months longer than expected. Even space tourism is developing a bad rap: In April, a star-studded all-woman crew on a Blue Origin suborbital flight was met with widespread backlash as a symbol of out-of-touch wealth and privilege.

It is at this point that we must loop back to the issue of “suckitude,” which Mandel also channels in her book through the killer opening of M.T. Anderson’s novel Feed: “We went to the moon to have fun, but the moon turned out to completely suck.”

The dreams of space settlements put forward by Musk and Bezos are insanely fun. The reality may well suck. But it’s doubtful that any degree of suckitude will slow down the commercial space race, and the authors do at times seem to be yelling into the cosmic void. 

Still, the books challenge space enthusiasts of all stripes to imagine new ways of relating to space that aren’t so tactile and exploitative. Along those lines, Rubenstein shares a compelling anecdote in Astrotopia about an anthropologist who lived with an Inuit community in the early 1970s. When she told them about the Apollo moon landings, her hosts burst out in laughter. 

“We didn’t know this was the first time you white people had been to the moon,” they said. “Our shamans go all the time … The issue is not whether we go to visit our relatives, but how we treat them and their homeland when we go.” 

Becky Ferreira is a science reporter based in upstate New York, and author of First Contact, a book about the search for alien life, which will be published in September. 

Meet the researcher hosting a scientific conference by and for AI

In October, a new academic conference will debut that’s unlike any other. Agents4Science is a one-day online event that will encompass all areas of science, from physics to medicine. All of the work shared will have been researched, written, and reviewed primarily by AI, and will be presented using text-to-speech technology. 

The conference is the brainchild of Stanford computer scientist James Zou, who studies how humans and AI can best work together. Artificial intelligence has already provided many useful tools for scientists, like DeepMind’s AlphaFold, which helps simulate proteins that are difficult to make physically. More recently, though, progress in large language models and reasoning-enabled AI has advanced the idea that AI can work more or less as autonomously as scientists themselves—proposing hypotheses, running simulations, and designing experiments on their own. 

James Zou
James Zou’s Agents4Science conference will use text-to-speech to present the work of the AI researchers.
COURTESY OF JAMES ZOU

That idea is not without its detractors. Among other issues, many feel AI is not capable of the creative thought needed in research, makes too many mistakes and hallucinations, and may limit opportunities for young researchers. 

Nevertheless, a number of scientists and policymakers are very keen on the promise of AI scientists. The US government’s AI Action Plan describes the need to “invest in automated cloud-enabled labs for a range of scientific fields.” Some researchers think AI scientists could unlock scientific discoveries that humans could never find alone. For Zou, the proposition is simple: “AI agents are not limited in time. They could actually meet with us and work with us 24/7.” 

Last month, Zou published an article in Nature with results obtained from his own group of autonomous AI workers. Spurred on by his success, he now wants to see what other AI scientists (that is, scientists that are AI) can accomplish. He describes what a successful paper at Agents4Science will look like: “The AI should be the first author and do most of the work. Humans can be advisors.”

A virtual lab staffed by AI

As a PhD student at Harvard in the early 2010s, Zou was so interested in AI’s potential for science that he took a year off from his computing research to work in a genomics lab, in a field that has greatly benefited from technology to map entire genomes. His time in so-called wet labs taught him how difficult it can be to work with experts in other fields. “They often have different languages,” he says. 

Large language models, he believes, are better than people at deciphering and translating between subject-specific jargon. “They’ve read so broadly,” Zou says, that they can translate and generalize ideas across science very well. This idea inspired Zou to dream up what he calls the “Virtual Lab.”

At a high level, the Virtual Lab would be a team of AI agents designed to mimic an actual university lab group. These agents would have various fields of expertise and could interact with different programs, like AlphaFold. Researchers could give one or more of these agents an agenda to work on, then open up the model to play back how the agents communicated to each other and determine which experiments people should pursue in a real-world trial. 

Zou needed a (human) collaborator to help put this idea into action and tackle an actual research problem. Last year, he met John E. Pak, a research scientist at the Chan Zuckerberg Biohub. Pak, who shares Zou’s interest in using AI for science, agreed to make the Virtual Lab with him. 

Pak would help set the topic, but both he and Zou wanted to see what approaches the Virtual Lab could come up with on its own. As a first project, they decided to focus on designing therapies for new covid-19 strains. With this goal in mind, Zou set off training five AI scientists (including ones trained to act like an immunologist, a computational biologist, and a principal investigator) with different objectives and programs at their disposal. 

Building these models took a few months, but Pak says they were very quick at designing candidates for therapies once the setup was complete: “I think it was a day or half a day, something like that.”

Zou says the agents decided to study anti-covid nanobodies, a cousin of antibodies that are much smaller in size and less common in the wild. Zou was shocked, though, at the reason. He claims the models landed on nanobodies after making the connection that these smaller molecules would be well-suited to the limited computational resources the models were given. “It actually turned out to be a good decision, because the agents were able to design these nanobodies efficiently,” he says. 

The nanobodies the models designed were genuinely new advances in science, and most were able to bind to the original covid-19 variant, according to the study. But Pak and Zou both admit that the main contribution of their article is really the Virtual Lab as a tool. Yi Shi, a pharmacologist at the University of Pennsylvania who was not involved in the work but made some of the underlying nanobodies the Virtual Lab modified, agrees. He says he loves the Virtual Lab demonstration and that “the major novelty is the automation.” 

Nature accepted the article and fast-tracked it for publication preview—Zou knew leveraging AI agents for science was a hot area, and he wanted to be one of the first to test it. 

The AI scientists host a conference

When he was submitting his paper, Zou was dismayed to see that he couldn’t properly credit AI for its role in the research. Most conferences and journals don’t allow AI to be listed as coauthors on papers, and many explicitly prohibit researchers from using AI to write papers or reviews. Nature, for instance, cites uncertainties over accountability, copyright, and inaccuracies among its reasons for banning the practice. “I think that’s limiting,” says Zou. “These kinds of policies are essentially incentivizing researchers to either hide or minimize their usage of AI.”

Zou wanted to flip the script by creating the Agents4Science conference, which requires the primary author on all submissions to be an AI. Other bots then will attempt to evaluate the work and determine its scientific merits. But people won’t be left out of the loop entirely: A team of human experts, including a Nobel laureate in economics, will review the top papers. 

Zou isn’t sure what will come of the conference, but he hopes there will be some gems among the hundreds of submissions he expects to receive across all domains. “There could be AI submissions that make interesting discoveries,” he says. “There could also be AI submissions that have a lot of interesting mistakes.”

While Zou says the response to the conference has been positive, some scientists are less than impressed.

“How do you get leaps of insight?”

Lisa Messeri

Lisa Messeri, an anthropologist of science at Yale University, has loads of questions about AI’s ability to review science: “How do you get leaps of insight? And what happens if a leap of insight comes onto the reviewer’s desk?” She doubts the conference will be able to give satisfying answers.

Last year, Messeri and her collaborator Molly Crockett investigated obstacles to using AI for science in another Nature article. They remain unconvinced of its ability to produce novel results, including those shared in Zou’s nanobodies paper. 

“I’m the kind of scientist who is the target audience for these kinds of tools because I’m not a computer scientist … but I am doing computationally oriented work,” says Crockett, a cognitive scientist at Princeton University. “But I am at the same time very skeptical of the broader claims, especially with regard to how [AI scientists] might be able to simulate certain aspects of human thinking.” 

And they’re both skeptical of the value of using AI to do science if automation prevents human scientists from building up the expertise they need to oversee the bots. Instead, they advocate for involving experts from a wider range of disciplines to design more thoughtful experiments before trusting AI to perform and review science. 

“We need to be talking to epistemologists, philosophers of science, anthropologists of science, scholars who are thinking really hard about what knowledge is,” says Crockett. 

But Zou sees his conference as exactly the kind of experiment that could help push the field forward. When it comes to AI-generated science, he says, “there’s a lot of hype and a lot of anecdotes, but there’s really no systematic data.” Whether Agents4Science can provide that kind of data is an open question, but in October, the bots will at least try to show the world what they’ve got. 

The Download: Google’s AI energy expenditure, and handing over DNA data to the police

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology.

In a first, Google has released data on how much energy an AI prompt uses

Google has just released a report detailing how much energy its Gemini apps use for each query. In total, the median prompt—one that falls in the middle of the range of energy demand—consumes 0.24 watt-hours of electricity, the equivalent of running a standard microwave for about one second. The company also provided average estimates for the water consumption (five drops per query) and carbon emissions associated with a text prompt to Gemini.

It’s the most transparent estimate yet from a Big Tech company with a popular AI product, and the report includes detailed information about how the company calculated its final estimate.

Earlier this year, MIT Technology Review published a comprehensive series on AI and energy, at which time none of the major AI companies would reveal their per-prompt energy usage. Google’s new publication, at last, allows for a peek behind the curtain that researchers and analysts have long hoped for. Read the full story.

—Casey Crownhart

I gave the police access to my DNA—and maybe some of yours

Last year, I added my DNA profile to a private genealogical database, FamilyTreeDNA, and clicked “Yes” to allow the police to search my genes.

In 2018, police in California announced they’d caught the Golden State Killer, a man who had eluded capture for decades. Once the police had “matches” to a few relatives of the killer, they built a large family tree from which they plucked the likely suspect.

This process, called forensic investigative genetic genealogy, or FIGG, has since helped solve hundreds of murders and sexual assaults.

But I wasn’t really driven by some urge to capture distantly related serial killers. Rather, my spit had a less gallant and more quarrelsome motive: to troll privacy advocates whose fears around DNA I think are overblown and unhelpful. By giving up my saliva for inspection, I was going against the view that a person’s DNA is the individualized, sacred text that privacy advocates sometimes claim. Read the full story.

—Antonio Regalado

This article appeared in The Checkup, MIT Technology Review’s weekly biotech newsletter. To receive it in your inbox every Thursday, and read articles like this first, sign up here.

Meet the researcher hosting a scientific conference by and for AI

In October, a new academic conference will debut that’s unlike any other. All of the work shared at Agents4Science will have been researched, written, and reviewed primarily by AI, and will be presented using text-to-speech technology. 

That idea is not without its detractors. Among other issues, many feel AI is not capable of the creative thought needed in research, makes too many mistakes and hallucinations, and may limit opportunities for young researchers. 

Nevertheless, a number of scientists and policymakers are very keen on the promise of AI scientists—and some even think they could unlock scientific discoveries that humans could never find alone. Read the full story.

—Peter Hall

The must-reads

I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.

1 Elon Musk tried to persuade Mark Zuckerberg to buy OpenAI
But the bid was rejected earlier this year. (Insider $)
+ OpenAI is asking Meta for evidence of any coordinated plans. (TechCrunch)
+ I’m guessing the cage fight is still off then. (FT $)

2 AI giants are seeking real-world data that can’t be scraped from the internet
It’s a bid to make their models more accurate and to find new use cases. (Rest of World)

3 Russia’s state-backed messenger app will be preinstalled on all phones
Critics say the MAX app is essentially a government spy tool. (Reuters)
+ Around 18 million people have registered to use it so far. (CNN)
+ How Russia killed its tech industry. (MIT Technology Review)

4 The Trump administration is refusing to fully fund a major HIV program
It’s ignoring a directive from Congress to withhold around $3 billion. (NYT $)
+ HIV could infect 1,400 infants every day because of US aid disruptions. (MIT Technology Review)

5 How Trump decides which chip companies may have to give up equity
Increasing your investments in the US? You’re off the hook. (WSJ $)
+ America-first chipmaking remains a fantasy, though. (Economist $)
+ Experts think Trump’s unconventional Intel deal may backfire. (Wired $)
+ DeepSeek’s new AI model is compatible with Chinese-made chips. (FT $)

6 The EU is speeding up its plans for a digital euro 💶
It’s considering running it on a public blockchain, to experts’ concern. (FT $)
+ Is the digital dollar dead? (MIT Technology Review)

7 We don’t have to open new mines to obtain minerals for clean energy
Although we have to get better at using the material we do mine. (New Scientist $)
+ How one mine could unlock billions in EV subsidies. (MIT Technology Review)

8 This newly-discovered gene could usher in new chronic pain treatments
One day, cutting out certain foods could lessen discomfort. (Economist $)
+ The pain is real. The painkillers are virtual reality. (MIT Technology Review)

9 Why Africa is buying so many solar panels
It’s not just its more affluent nations snapping them up, either. (Wired $)
+ The race to get next-generation solar technology on the market. (MIT Technology Review)

10 How families are using AI to run their households
No more quibbling over meal planning. (WP $)

Quote of the day

“If AGI doesn’t come to pass sometime soon, I wouldn’t be surprised if this whole thing pops.”

—Bhavya Kashyap, an angel investor, tells Insider why investors are fuelling a risky bubble by rushing to buy stocks in the hottest AI companies.

One more thing

How AI is changing gymnastics judging

The 2023 World Championships last October marked the first time an AI judging system was used on every apparatus in a gymnastics competition. There are obvious upsides to using this kind of technology: AI could help take the guesswork out of the judging technicalities. It could even help to eliminate biases, making the sport both more fair and more transparent.

At the same time, others fear AI judging will take away something that makes gymnastics special. Gymnastics is a subjective sport, like diving or dressage, and technology could eliminate the judges’ role in crafting a narrative.

For better or worse, AI has officially infiltrated the world of gymnastics. The question now is whether it really makes it fairer. Read the full story.

—Jessica Taylor Price

We can still have nice things

A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or skeet ’em at me.)

+ Finally, some good news—a sweet little Australian marsupial called an ampurta is no longer endangered (thanks Glen!)
+ What would a GTA set in London look like?
+ Why glass houses aren’t all they’re cracked up to be (geddit?)
+ Over in Denmark, there’s a national competition encouraging cities to get rid of their gray concrete tiles and replace them with peaceful green spaces (thanks Alice!)

Triple Whale’s Moby AI Gets Things Done

We’ve all heard the buzz surrounding agentic AI agents. What’s missing for many of us is how they can help our business. What is an AI agent? Can it really perform tasks and get things done?

I asked those questions and more to Anthony DelPizzo. He’s with Triple Whale, the Shopify-backed ecommerce analytics platform that has launched its own AI agent called Moby. It responds to ChatGPT-like prompts, suggests marketing channels, and even composes emails.

The entire audio of our conversation is embedded below. The transcript is condensed and edited for clarity.

Eric Bandholz: Who are you, and what do you do?

Anthony DelPizzo: I lead product marketing at Triple Whale and have been here for about nine months. Before that, I spent nearly four years at Klaviyo.

Triple Whale is an analytics platform for ecommerce brands. We merge fragmented data across marketing and sales into a single system and dashboard to help merchants make strategic decisions. To date, we’ve processed over $65 billion in gross merchandise volume.

We launched Moby, an agentic AI agent, about a month ago after a long testing phase. Moby is a set of AI tools that interact directly with merchants’ data. Think of it as ChatGPT focused on the platforms you already work with. Merchants can ask Moby both simple and complex questions and get answers tailored to their own data.

Moby Agents take it a step further. They’re akin to autonomous teammates that can analyze information, generate insights, and even take actions across ad platforms, marketing channels, operations, and more. The result could be much higher conversions or lower overhead.

Moby is built on Triple Whale’s massive data warehouse. It draws on those benchmarks and works natively with metrics such as CAC and ROAS. By using the data, Moby can connect cleanly with large language models such as Anthropic and OpenAI for each type of query.

Moby is embedded within the Triple Whale platform. It doesn’t just analyze; it can also perform tasks such as activating ads or drafting emails.

Bandholz: Do you share customer data with those LLMs?

DelPizzo: We have privacy agreements with all LLM  partners. Data stays within Triple Whale’s private environment. We’re not sending entire datasets to Anthropic, OpenAI, or any other company. Instead, Moby provides context to the LLMs based on the prompt, allowing our customers to use the LLMs securely.

For example, a prompt could be, “How should I prepare for BFCM to grow revenue 30%?” Moby’s Deep Dive feature breaks requests like this into multiple steps, with each acting as an agent examining a different aspect of the business. The result is a structured plan merchants can use to prepare for Black Friday and Cyber Monday.

Merchants use Moby for general prompts and analysis, not just seasonal planning. We provide a prompting guide to help start with effective questions and then refine the queries through follow-ups.

Bandholz: Say I prompt Moby to analyze my sales, margins, and ads, for guidance. What then?

DelPizzo: Moby would connect to your data as a Triple Whale client — product margins, SKUs, ad performance, Klaviyo, Attentive, logistics, and more. By analyzing these inputs, it can identify growth levers, such as which products or channels drove profit last year and which ones are trending now. For instance, if a brand has started performing well on AppLovin, the mobile ad platform, Moby might suggest scaling there for BFCM.

Triple Whale’s platform includes eight attribution models, along with post-purchase surveys, to track what’s driving results. We’ve also added marketing mix modeling to measure the impact from click and non-click channels, including Amazon. Moby can run correlations at a statistically significant level, which gives merchants confidence in the conclusions.

Based on that, it forecasts likely outcomes tied to business goals. If a brand wants to grow revenue by 30%, Moby highlights which levers — spending, channels, creative — are likely to help reach that target. Merchants can even see Moby’s reasoning step by step, like watching strategists think through a plan.

Moby’s analysis isn’t limited to numbers. Using AI vision, it can review ad creative, such as color choices, hooks, and copy. It also analyzes email performance by scanning HTML, subject lines, and preview text. It can draft email copy informed by this analysis, giving merchants ideas to test.

Bandholz: Can you cite anonymous customer wins from Moby?

DelPizzo: We rolled out early access to Moby and Moby Agents in February, five months ago. In April, a $100 million global brand used it during a four-day giveaway. On the final day, the team asked Moby, “What should we adjust in our plan?”

Moby responded with a detailed budget allocation by channel and predicted the revenue impact. They followed it exactly and ended up having their highest revenue day ever — 35% above their previous record, more than $200,000 higher.

Another example is LSKD, a fitness apparel brand in Australia with more than 50 stores. They used Moby to analyze the performance of their marketing channels. One agent uncovered over $100,000 in fraudulent spend from an influencer’s self-bidding, which saved the company that money. Since adopting Moby Agents, LSKD’s ROAS has grown about 40%.

Bandholz: How can merchants go wrong with Moby?

DelPizzo: The most common challenge is trying to adopt too much at once. Success usually comes from starting small. We provide a library of 70 pre-built agents, but using all of them right away can feel overwhelming.

The best outcomes are from teams that begin with a single agent, adapt it to their business, and build confidence with steady results. From there, they expand to other areas — maybe they start with the conversion rate optimization team, then retention, then other steps in the funnel. That gradual approach tends to be more sustainable.

Bandholz: Why use Moby instead of building a custom data tool with an LLM such as DeepSeek?

DelPizzo: One factor is the dataset it draws from. Moby is trained on $65 billion in GMV and has access to broad ecommerce benchmarks. It’s not about sharing brand-specific data but rather using aggregated insights to provide context — like knowing typical CAC or ROAS levels in different industries, or, say, margins for apparel versus skincare.

Another piece is the infrastructure. Building from scratch requires a unified schema for orders, events, and performance data. At Triple Whale, our large team of engineers has worked on this for years, and it’s still evolving. Without that groundwork, it’s hard to achieve the same level of ecommerce-specific intelligence.

Custom setups are possible, but Moby combines benchmarks, context, and infrastructure in a way that’s difficult to replicate.

Bandholz: Where can people support you, follow you, reach out?

DelPizzo: Our site is TripleWhale.com. Our socials include X and LinkedIn. I’m on LinkedIn.

Tips For Running Competitor Campaigns In Paid Search via @sejournal, @timothyjjensen

Paid search professionals constantly debate the merits of running paid search campaigns bidding on competitor brand names. Questions such as the following may arise:

  • Is bidding on your competitors ethical?
  • Are the high costs-per-click (CPCs) worth spending the budget on?
  • Are you actually reaching people with buying intent?

In this article, I’ll talk through answers to these questions and more to help you understand if a competitor search campaign might be right for your brand.

Competitor Bidding Ethics

Google and Microsoft allow you to bid on your competitor’s name within keywords (and this right has even been tested in the courts here and here.), but you cannot directly mention a trademarked brand name (that you don’t have the rights to use) in ad copy.

In addition, even if you don’t include their name, you should not write your ad copy in a way that a user thinks they may be going to your competitor’s site instead of yours.

For instance, you might use the headline “Official Site” (without mentioning whose official site you’re pointing to). When a user sees that in conjunction with having searched for the competitor’s name, they may naturally think they’re going to that company’s site.

Finally, the landing page should also clearly feature your brand’s name and logo in order to avoid deception.

Cost-Benefit Analysis Of Competitor Bidding

Let’s face it: competitor keywords can have expensive CPCs. High competition around these keywords in many industries drives up cost.

You’ll also generally struggle to achieve a decent quality score due to other companies’ brand keywords naturally being deemed less relevant to your ads and landing pages, which can also impact cost.

Because of the high potential cost, competitor bidding does not make sense for all industries or brands.

For instance, if you’re selling products with a low profit margin, bidding on these pricy keywords may not work. Generally, this tactic works best for higher cost, higher margin products and services, as it’s easier to still yield a return on investment (ROI) after higher costs-per-acquisition (CPAs) and lower conversion rates.

Be careful also about entering competitor bidding “wars” for the sole reason that other brands are bidding on your name. This action can quickly lead to rising CPCs for all with little payoff.

One scenario where I’ve seen competitor bidding work best is when a company offers a very specific, complex service that’s difficult to sum up in a search query but has established brands that the right prospects would be familiar with.

For instance, if you’re promoting software for a particular type of industrial machine, niche buyers may be aware of companies that already provide that software.

Once you’ve established a use case for competitor bidding, you should establish a list of brands to use.

Determining Competitors To Bid On

When figuring out which competitor brands to bid on, you should rely on a combination of both internal company data as well as ad platform data.

First of all, talk with key stakeholders in marketing and sales to determine who the brand considers to be top competitors.

Who has similar products and services? Which brands target similar prospects (whether by location, demographic, or company traits)?

Note that this list may not and likely will not contain all potential competitors.

If you have established paid search campaigns already, use auction insights to see the top brands showing up for the same queries as yours. Of course, these may not all be completely relevant and will require some vetting through.

Once you’ve compiled a list, it’s time to think through the keywords you’ll bid on.

Who Is (And Isn’t) Your Audience

Be careful about going unnecessarily broad in the keywords you’re using in competitor campaigns.

Generally, if you’re just bidding on the brand name alone, you’re likely reaching a lot of existing customers looking to log in, place online orders, or find a nearby location without giving a second thought to anything else.

For instance, Apple isn’t going to sell many MacBooks by bidding on the word “Microsoft.”

Ideally, you want to reach people who are in a research phase, indicated by wording in their search query:

  • [Brand name] + cost/pricing
  • [Brand name] + compare/vs
  • [Brand name] + reviews
  • [Brand name] + pros/cons
  • [Brand name] + alternatives
  • [Brand name] + features

While a potentially riskier strategy, as people may be in a heated moment, you could also test targeting people experiencing issues and potentially in the market to switch:

  • [Brand name] + support
  • [Brand name] + troubleshoot
  • [Brand name] + cancel

Create Your Ads

Now, think through the ad copy you’ll put in front of prospects searching for competitors. Take some time to review competitor ads and offers, considering how your calls-to-action (CTAs) will stack up.

Think through areas where you “win” against certain competitors and highlight those. Remember that these may vary based on the brand you’re bidding against.

For instance, you may have lower costs than a certain competitor and highlight pricing for those searches, while you may have higher costs than another competitor but have unique features to highlight.

Also, look at how your offers compare. If one competitor offers a seven-day demo and you offer a 30-day demo, feature that in your ad.

This also should be an area you regularly monitor and adjust CTAs based on how competitors tweak their ads and offers.

What Happens After The Ad?

One maxim applicable to any paid search campaign is that what happens on the search engine results page up to the ad click is only one portion of the user experience.

A significant portion of the decision process happens after reaching the landing page, beyond what you can control in keywords and ad copy.

Think through what your prospect is seeing based on the context that they were researching a competitor. Your homepage probably isn’t the best place to land them, and the same sales landing page you use for more general keywords may not be ideal either.

Assuming a user is comparison shopping, placing some content on your landing page positioning your brand against others will likely help.

For instance, you could create a table showing how your features and pricing stack up vs. competitors (either mentioning specific names or providing industry averages).

You could also hone in on trust signals that set your brand apart. Highlight industry awards you’ve won. Mention the number of accounts serviced. Talk about how many integrations you have with commonly used products.

If you need to establish a baseline for comparing against other companies, prompt a large language model (LLM) to put together a list of features for your brand and a list of top competitors.

Provide the URLs for pages that would contain products/services to flesh this out.

Launch And Monitor Results

Once you have your competitor campaigns fleshed out, it’s time to get them off the ground and see what performance looks like.

In addition to ensuring proper conversion tracking and watching for lead/sale quality, you’ll also want to keep an eye out for both how current competitors change up their offers and new competitors entering the space that may be worth targeting.

With a carefully thought-out setup and proper monitoring, you may find that competitor search campaigns allow you to capture leads or sales from queries you were not previously reaching.

On the other hand, you may discover that for your industry, the CPAs and conversion rates aren’t worthwhile, but as with anything in PPC, you ran a test and learned the results.

At the very least, take stock of potential competitors in your field and consider testing if you are looking to expand your reach in paid search.

More Resources:


Featured Image: SvetaZi/Shutterstock

New Ecommerce Tools: August 21, 2025

Every week we handpick and publish a list of new products and services from vendors of ecommerce merchants. This installment includes updates on cross-border transactions, marketing, social commerce, AI shopping agents, AI-powered subscriptions, and order management platforms.

Got an ecommerce product release? Email releases@practicalecommerce.com.

New Tools for Merchants

dLocal and Tiendamia partner on cross-border ecommerce in Latin America. dLocal, a cross-border payment platform connecting global merchants to emerging markets, has partnered with Tiendamia, a Latin America-based marketplace. Through the integration, Tiendamia can accept cross-border payments and offer a range of local payment methods, from cards and cash-based options, across Ecuador, Costa Rica, Peru, and Argentina, where it also supports digital wallets, and enable domestic transactions in Uruguay. Tiendamia can pay local providers in Ecuador, Costa Rica, Peru, and Uruguay.

Home page of dLocal

dLocal

Privy debuts marketing automation features for ecommerce brands. Privy, an ecommerce marketing platform, has launched features to help merchants simplify workflow and personalize the customer journey. Privy Flows is a new visual marketing automation tool to trigger emails and texts based on customer actions. The SMS and MMS campaign composer enables sellers to design and send high-performing mobile campaigns quickly. Smart Triggers offers combining conditions, such as exit intent and scroll depth, to deliver a message at the right moment.

Payoneer and Stripe enhance online checkout experience for SMBs. Payoneer, a global payments and funding platform, has partnered with Stripe, the payment processor, expanding Payoneer’s Online Checkout offering for cross-border direct-to-consumer merchants. Launching in the Asia Pacific region first, the upgraded Payoneer Checkout capabilities, powered by Stripe, will empower SMBs to accept a broader range of payments via online webstore checkout, including buy-now pay-later options such as Affirm and Klarna, and digital wallets such as Apple Pay and Google Pay.

BlueSnap partners with Commerce for B2B payments and accounts receivable automation. BlueSnap, a payment orchestration platform for B2B and B2C businesses, has announced its integration with Commerce, the parent company of BigCommerce, to deliver automations for B2B payments and accounts receivable. BigCommerce merchants can sync customer and invoice data in real time with back-office systems. Buyers can view and pay inventory orders and vendor invoices in one branded portal. Enable autopay, early pay discounts, invoice reminders, and real-time updates.

Riskified partners with Human on AI shopping agent commerce. Riskified, a provider of ecommerce fraud prevention and risk intelligence, has partnered with Human Security, a cybersecurity company, to advance a unified security framework for merchants. By aligning Human Security’s recently launched Sightline featuring AgenticTrust with Riskified’s ecommerce risk management expertise in fraud prevention, chargeback protection, and policy abuse prevention, merchants can apply consistent trust policies and transaction decisions across both human and AI-driven interactions, according to Riskified.

Home page of Riskified

Riskified

Subotiz launches AI-powered subscription platform. Subotiz is a new platform combining subscription management, global payments, and intelligent automation. Flexible subscription billing supports recurring plans, tiered pricing, trials, promotions, and dynamic billing cycles. AI-powered automation monitors user behavior and payment patterns to reduce churn. Subotiz provides built-in tools for customizable cross-border tax configurations and compliance. Also, Subotiz connects to over 200 payment methods across multiple gateways, currencies, and regions.

eBay launches AI-powered seller tools. At its Open25 seller event in April, eBay unveiled a suite of features and updates to its marketplace, including embedded text offer-making, AI-powered messaging support, and new seller protections. eBay has released an AI assistant for messaging on the eBay mobile app and web in the U.S. and U.K. With Offers in Messaging, buyers and sellers can negotiate directly in the message thread — sending, receiving, countering, and accepting offers without switching screens.

Warp launches SMB suite to simplify logistics. Warp, an enterprise freight transportation service, has launched the SMB Suite, a bundled multichannel logistics platform tailored for the apparel, retail, and consumer goods sectors. Warp’s offerings include less-than-truckload, pool distribution, big and bulky final mile delivery, inbound vendor consolidation, and zone skipping, integrating its nationwide network of tech-enabled cross-docks, flexible routing systems, and unified technology stack.

WebSell integrates with Microsoft Dynamics 365 Business Central. WebSell, an ecommerce platform for retailers and wholesalers, has integrated with Microsoft Dynamics 365 Business Central, allowing companies to connect back-office and ecommerce operations in a streamlined platform. The integration automatically syncs data between Microsoft Dynamics 365 Business Central and an online store, including products, inventory, prices, customer information, and orders. WebSell supports both B2B and B2C models and includes tools such as customer-specific pricing, multi-store management, search engine optimization, and marketing services.

Home page of WebSell

WebSell

Mayple and Emirates Courier Express partner for cross-border delivery. Mayple Global, an ecommerce logistics platform, has partnered with Emirates Courier Express to expand cross-border delivery capabilities for U.S. merchants. The collaboration utilizes Mayple’s centralized logistics hub in Dubai and Emirates Courier Express’s network to deliver packages to eight international markets, shortening transit times, reaching challenging markets, simplifying customs handling, and accessing competitive shipping rates. Mayple’s model centralizes inventory in Dubai, so that brands can ship from a single hub.

CollAble launches Find.ly storefront tool for influencers. CollAble, a digital influencer network, has launched Find.ly, a storefront and link aggregator tool turning influencers’ social media posts into shoppable storefronts. Influencers can (i) curate and organize products in a visually appealing storefront, (ii) synchronize social media posts with shoppable product links, enabling audiences to purchase directly from the post, and (iii) integrate with affiliate networks to ensure real-time tracking and commission attribution.

THG Commerce expands social commerce features as TikTok Shop Partner. THG Commerce, an ecommerce platform from THG Ingenuity, a sales acceleration provider, has announced the expansion of its social commerce capabilities and official TikTok Shop Partner status. The expanded services include social strategy development, live commerce execution, content creation, influencer and affiliate marketing, community management, and social analysis and reporting.

Deck Commerce launches modular order management solution. Deck Commerce, an order management system for D2C brands, has launched Commerce Centers, a modular order management platform that helps brands retain shoppers through fulfillment, delivery, and returns. According to Deck Commerce, each Center helps brands improve key steps in the process by making inventory more available, orders easier to manage, fulfillment faster and more accurate, and service more reliable.

Home page of Deck Commerce

Deck Commerce

Google: Why Lazy Loading Can Delay Largest Contentful Paint (LCP) via @sejournal, @MattGSouthern

In a recent episode of Google’s Search Off the Record podcast, Martin Splitt and John Mueller discussed when lazy loading helps and when it can slow pages.

Splitt used a real-world example on developers.google.com to illustrate a common pattern: making every image lazy by default can delay Largest Contentful Paint (LCP) if it includes above-the-fold visuals.

Splitt said:

“The content management system that we are using for developers.google.com … defaults all images to lazy loading, which is not great.”

Splitt used the example to explain why lazy-loading hero images is risky: you tell the browser to wait on the most visible element, which can push back LCP and cause layout shifts if dimensions aren’t set.

Splitt said:

“If you are using lazy loading on an image that is immediately visible, that is most likely going to have an impact on your largest contentful paint. It’s like almost guaranteed.”

How Lazy Loading Delays LCP

LCP measures the moment the largest text or image in the initial viewport is painted.

Normally, the browser’s preload scanner finds that hero image early and fetches it with high priority so it can paint fast.

When you add loading="lazy" to that same hero, you change the browser’s scheduling:

  • The image is treated as lower priority, so other resources start first.
  • The browser waits until layout and other work progress before it requests the hero image.
  • The hero then competes for bandwidth after scripts, styles, and other assets have already queued.

That delay shifts the paint time of the largest element later, which increases your LCP.

On slow networks or CPU-limited devices, the effect is more noticeable. If width and height are missing, the late image can also nudge layout and feel “jarring.”

SEO Risk With Some Libraries

Browsers now support a built-in loading attribute for images and iframes, which removes the need for heavy JavaScript in standard scenarios. WordPress adopted native lazy loading by default, helping it spread.

Splitt said:

“Browsers got a native attribute for images and iframes, the loading attribute … which makes the browser take care of the lazy loading for you.”

Older or custom lazy-loading libraries can hide image URLs in nonstandard attributes. If the real URL never lands in src or srcset in the HTML Google renders, images may not get picked up for indexing.

Splitt said:

“We’ve seen multiple lazy loading libraries … that use some sort of data-source attribute rather than the source attribute… If it’s not in the source attribute, we won’t pick it up if it’s in some custom attribute.”

How To Check Your Pages

Use Search Console’s URL Inspection to review the rendered HTML and confirm that above-the-fold images and lazy-loaded modules resolve to standard attributes. Avoid relying on the screenshot.

Splitt advised:

“If the rendered HTML looks like it contains all the image URLs in the source attribute of an image tag … then you will be fine.”

Ranking Impact

Splitt framed ranking effects as modest. Core Web Vitals contribute to ranking, but he called it “a tiny minute factor in most cases.”

What You Should Do Next

  • Keep hero and other above-the-fold images eager with width and height set.
  • Use native loading="lazy" for below-the-fold images and iframes.
  • If you rely on a library for previews, videos, or dynamic sections, make sure the final markup exposes real URLs in standard attributes, and confirm in rendered HTML.

Looking Ahead

Lazy loading is useful when applied selectively. Treat it as an opt-in for noncritical content.

Verify your implementation with rendered HTML, and watch how your LCP trends over time.


Featured Image: Screenshot from YouTube.com/GoogleSearchCentral, August 2025. 

Google Confirms New Google Verified Badge for Local Services Ads via @sejournal, @brookeosmundson

Google just announced a new unifying identity for its Local Services Ads (LSAs) verification badges.

Called Google Verified, the badge will replace several different trust signals that advertisers and consumers have been seeing over the years.

This includes the Google Guaranteed, Google Screened, License Verified by Google, and the Money Back Guarantee program.

Starting in October 2025, eligible LSAs that pass the necessary screenings will display this streamlined mark: a single badge designed to communicate credibility in a more consistent way.

Why is Google Consolidating Badges?

In the past, Google’s verification system was fragmented.

Different types of businesses had different badges, and consumers were left guessing what each one actually meant. Was a “Screened” provider more trustworthy than a “Guaranteed” one? Did a license verification carry more weight than a money-back promise?

The lack of consistency made it harder for advertisers to explain their value and for consumers to make decisions.

By rolling everything into one identity, Google Verified aims to simplify the process for everyone involved.

The badge will not only appear across Local Service Ads but will also include transparency for consumers. When a user taps or hovers over the badge, they can see the specific checks a business has passed.

How Does This Change Impact Advertisers?

For marketers and business owners, the simplified badge system removes some of the confusion around what signals matter.

Instead of juggling multiple programs, the message is now clear: your business is either Google Verified, or it’s not.

That said, the bar for participation may feel higher. Businesses that don’t keep their documentation, licensing, and other requirements up to date risk losing the badge.

Since Google has indicated it may only show the badge when it predicts it will help users make decisions, credibility and visibility could become even more closely linked.

In short, advertisers who maintain verification stand to benefit from increased trust, while those who lag behind could see their ads appear less competitive.

This update doesn’t require marketers to overhaul their entire strategy by any means. However, there are a few practical steps you can take to ensure a smooth transition by October.

  • Review eligibility now. Make sure your licenses, insurance, and background checks are up-to-date before October.
  • Build in reminders. Treat verification like an ongoing compliance process, not a one-time task.
  • Educate clients or internal teams. If you manage LSA campaigns for others, help them understand that the badge isn’t just a cosmetic update. It reflects ongoing credibility.
  • Monitor performance post-launch. Once the new badge rolls out, watch for shifts in click-thru rate (CTR) and conversion rates. If verification gives a measurable lift, you’ll want to highlight that value in your reporting.

A Shift Toward Ongoing Trust

Google Verified may look like a rebrand on the surface, but it’s also a signal that trust in digital advertising is moving toward continuous validation.

For businesses, this means credibility is not something you earn once; it’s something you prove over and over again.

For advertisers, the key takeaway is simple: don’t treat this as a one-time update. Verification will become an expectation, not a nice-to-have, and it could influence not just how consumers view your ads but how often those ads are shown.