On March 20, 2026, Google quietly added a new entry to its official list of web fetchers. Not a crawler. Not a training bot. An agent.
Google-Agent is the user agent string for AI systems running on Google infrastructure that browse websites on behalf of users. When someone asks an AI assistant to research a product, fill out a form, or compare options across websites, Google-Agent is the thing that actually visits the page. Project Mariner, Google’s experimental AI browsing tool, is the first product using it.
This is not Googlebot. Googlebot crawls the web continuously, indexing pages for search. Google-Agent only shows up when a human asks it to. That distinction changes everything about how it operates.
Robots.txt Does Not Apply
Google classifies Google-Agent as a user-triggered fetcher. The category includes tools like Google Read Aloud (text-to-speech), NotebookLM (document analysis), and Feedfetcher (RSS). All of them share one property: a human initiated the request. Google’s position is that user-triggered fetchers “generally ignore robots.txt rules” because the fetch was requested by a person.
The logic: If you type a URL into Chrome, the browser fetches the page regardless of what robots.txt says. Google-Agent operates on the same principle. The agent is the user’s proxy, not an autonomous crawler.
This is a meaningful departure from how OpenAI and Anthropic handle similar traffic. ChatGPT-User and Claude-User both function as user-triggered fetchers, but they respect robots.txt directives. If you block ChatGPT-User in robots.txt, ChatGPT won’t fetch your page when a user asks it to browse. Google made a different call.
Website owners who relied on robots.txt as a universal access control mechanism now have a gap. If you need to restrict access from Google-Agent, you’ll need server-side authentication or access controls. The same tools you’d use to block a human visitor.
Cryptographic Identity: Web Bot Auth
The more significant development is buried in a single line of Google’s documentation: Google-Agent is experimenting with the web-bot-auth protocol using the identity https://agent.bot.goog.
Web Bot Auth is an IETF draft standard that works like a digital passport for bots. Each agent holds a private key, publishes its public key in a directory, and cryptographically signs every HTTP request. The website verifies the signature and knows, with cryptographic certainty, that the visitor is who it claims to be.
User agent strings can be spoofed by anyone. Web Bot Auth cannot. Google adopting this protocol, even experimentally, signals where agent identity is heading. Akamai, Cloudflare, and Amazon (AgentCore Browser) already support it. Google brings the critical mass.
This matters because the web is about to have an identity problem. As agent traffic increases, websites need to distinguish between legitimate AI agents acting on behalf of real users and scrapers pretending to be agents. IP verification helps, but cryptographic signatures scale better and are harder to fake.
What This Means For Your Website
Google-Agent creates a three-tier visitor model for the web:
Human visitors browsing directly.
Crawlers indexing content for search and training (Googlebot, GPTBot, Google-Extended).
Agents acting on behalf of specific humans in real time (Google-Agent, ChatGPT-User, Claude-User).
Each tier has different access rules, different intentions, and different expectations. A crawler wants to index your content. An agent wants to complete a task. It might be reading a product page, comparing prices, filling out a contact form, or booking an appointment.
Here’s what to do now:
Monitor your logs. Google-Agent identifies itself with a user agent string containing compatible; Google-Agent. Google publishes IP ranges for verification. Start tracking how often agents visit, which pages they hit, and what they attempt to do.
Check your CDN and firewall rules. If your security tools aggressively block non-browser traffic, Google-Agent may be getting rejected before it reaches your server. Verify that Google’s published IP ranges are permitted.
Test your forms and flows. Google-Agent can submit forms and navigate multi-step processes. If your checkout, booking, or contact forms rely on JavaScript patterns that confuse automated systems, agent visitors will fail silently. Semantic HTML and clear labels remain the foundation.
Accept that robots.txt is no longer a complete access control tool. For content you genuinely need to restrict, use authentication. robots.txt was designed for crawlers. The agent era needs different boundaries.
The Hybrid Web Isn’t Coming. It’s Logged
A year ago, the idea that AI agents would browse websites alongside humans was a conference talk prediction. Today, it has a user agent string, published IP ranges, a cryptographic identity protocol, and an entry in Google’s official documentation.
The web didn’t split into human and machine. It merged. Every page you publish now serves both audiences simultaneously, and Google just made it possible to see exactly when the non-human audience shows up.
Two sections are specifically worth highlighting. Google directly names popular optimization tactics it says aren’t necessary, and it redefines the AEO/GEO conversation as part of standard SEO.
Google Says AEO And GEO Are ‘Still SEO’
Google opens by confirming that foundational SEO best practices remain relevant for generative AI search. Its AI features are “rooted in our core Search ranking and quality systems” and rely on retrieval-augmented generation (RAG) and query fan-out to surface content from the Search index.
On the terminology debate, Google is direct. It defines “AEO” as “answer engine optimization” and “GEO” as “generative engine optimization,” then states:
“From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.”
This echoes positions Google employees have taken at conferences. Gary Illyes and Cherry Prommawin told Search Central Live attendees that GEO and AEO don’t require separate frameworks. The position now appears in Google’s published documentation, providing an official reference to cite.
What Google Says You Don’t Need To Do
The guide includes a “Mythbusting generative AI search” section listing tactics it calls unnecessary for Google Search. The guide is more explicit than Google’s prior AI features page, particularly in naming llms.txt, chunking, inauthentic mentions, and AEO/GEO directly.
The guide says site owners can ignore the following for Google Search.
On llms.txt files and other “special” markup, Google says you don’t need to create machine-readable files, AI text files, markup, or Markdown to appear in generative AI search. Google may discover and index many file types beyond HTML, but that doesn’t mean those files receive special treatment.
On “chunking” content, the guide says there’s no requirement to break content into small pieces for AI systems. Google’s systems “are able to understand the nuance of multiple topics on a page and show the relevant piece to users.” Danny Sullivan made similar comments in January 2026, saying he’d spoken with Google engineers who recommended against chunking.
On rewriting content for AI systems, Google says AI systems can understand synonyms and general meanings. Site owners don’t need to capture every long-tail keyword variation or write in a specific way for generative AI search.
On seeking inauthentic “mentions,” the guide acknowledges that AI features can surface what’s said about products and services across blogs, videos, and forums. But it says seeking inauthentic mentions “isn’t as helpful as it might seem” because core ranking systems focus on quality while other systems block spam.
On structured data, the guide says it isn’t required for generative AI search and there’s no special schema.org markup to add. It recommends continuing to use structured data as part of an overall SEO strategy for rich results eligibility.
Several recommendations run counter to advice that appears in some AI search optimization guides. Multiple GEO resources have promoted chunking and structured data as priorities for AI search visibility.
What Google Says To Focus On
The optimization advice follows familiar SEO territory, though Google contextualizes it for AI features.
Google puts particular emphasis on “non-commodity content.” It contrasts commodity content (“7 Tips for First-Time Homebuyers”) with a non-commodity alternative (“Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line”). The distinction is whether content provides unique insight beyond common knowledge.
On the technical side, pages must be indexed and eligible for snippets to appear in generative AI features. Google recommends following crawling best practices, using semantic HTML where possible, following JavaScript SEO best practices, providing good page experience, and reducing duplicate content.
Local and ecommerce optimization gets its own section. Google recommends Merchant Center feeds and Google Business Profiles for product and local business visibility in AI responses. It also mentions Business Agent, a conversational experience that lets customers chat with brands on Google Search.
Agentic Experiences Get Initial Guidance
A new section on agentic experiences describes AI agents as “autonomous systems that can perform tasks on behalf of people, such as booking a reservation or comparing product specifications.”
Google notes that browser agents may access websites by analyzing screenshots, inspecting the DOM, and interpreting the accessibility tree. The guide links to web.dev’s guide to agent-friendly website best practices and references the Universal Commerce Protocol (UCP) as an emerging protocol that “will allow Search agents to do more.”
Google announced UCP earlier this year, and Vidhya Srinivasan’s annual letter said it was co-developed with Shopify with more than 20 companies endorsing it.
Why This Matters
This guide gives Google’s most explicit guidance yet on what you should and shouldn’t do for generative AI features in Search. It consolidates positions that were previously scattered across conference talks, podcast appearances, and blog posts into a single reference.
The mythbusting section carries the most weight. Google is now telling you in its own documentation to skip tactics that a growing industry of AEO/GEO services has been promoting. That doesn’t settle the debate for non-Google AI platforms like ChatGPT or Perplexity, which may weight signals differently. But for Google’s own AI features, the guidance is now on record.
The agentic experiences section puts browser agents and UCP into Google’s official documentation for site owners. The guidance is early, and Google frames it as optional for businesses where agent access is relevant.
Looking Ahead
Google’s closing section says you don’t need to accomplish everything in the document to succeed. It notes that “plenty of content thrives in Google Search (including generative AI experiences) without any overt SEO at all.”
The agentic experiences guidance is labeled as something to explore “if this is something that’s relevant to your business and you have extra time.” That suggests Google sees agent optimization as forward-looking rather than urgent.
Welcome to this week’s Pulse. The updates affect how you measure AI assistant traffic, what structured data does for visibility, and how a major publisher is planning for life after search.
Here’s what matters for you and your work.
Google Analytics Adds Native AI Assistant Channel
Google Analytics now assigns traffic from recognized AI chatbots to a dedicated “AI Assistant” default channel group. Custom channel groups with regex patterns are no longer the only way to separate AI assistant visits from referrals.
Key Facts
Sessions from recognized AI assistants now receive “ai-assistant” as the medium, route to a new “AI Assistant” default channel, and get a reserved “(ai-assistant)” campaign label. Google named ChatGPT, Gemini, and Claude as examples, but hasn’t published the full list of recognized referrers. All three changes happen automatically.
Why This Matters
Anyone running a custom channel group to isolate AI chatbot traffic can now compare their setup against Google’s native version. The custom regex patterns Google recommended last August still cover platforms outside the recognized referrer list. Both can run side by side.
The bigger question is what you do with the data once it’s visible. AI assistant traffic is now a distinct line item in acquisition, user, and channel reports. That makes it easier to compare conversion behavior, session quality, and volume against organic search without filtering or manual workarounds.
Google hasn’t said how quickly the recognized referrer list will expand as new platforms launch. If you track AI assistants beyond the three named examples, keep your custom groups running.
Google deprecated FAQ rich results, completing a removal that started a few years ago. The company added a notice to its FAQ structured data documentation without a blog post or separate explanation.
Key Facts
FAQ rich results stopped appearing in search results. Google will remove the FAQ search appearance filter in Search Console, the rich result report, and support for Rich Results Test in June. API support ends in August.
Why This Matters
If your reporting pipelines pull FAQ-specific data from the API, those API calls need to be updated before the August cutoff.
Leaving the markup in place shouldn’t create problems, but it no longer produces that visible result. Whether FAQ schema aids AI search is a separate question, and the deprecation doesn’t answer it.
Ahrefs Report: Adding Schema Didn’t Increase AI Citations
An Ahrefs report tracked 1,885 pages that added JSON-LD schema and found no meaningful increase in AI citations across Google AI Overviews, AI Mode, or ChatGPT.
Key Facts
Ahrefs matched each treated page against controls that never added schema and measured changes over 30-day windows. AI Overviews showed a 4.6% decline relative to controls, while AI Mode (+2.4%) and ChatGPT (+2.2%) showed changes too small to distinguish from noise.
Why This Matters
The correlation between schema and AI citations has been widely cited as evidence that structured data improves AI visibility. Ahrefs tested whether the relationship appeared causal and found no evidence of a meaningful lift, at least for pages already being cited. Sites with schema tend to also invest in better content, stronger authority, and more links. Those factors may explain the correlation better than the markup itself.
The report can’t say whether schema helps pages that aren’t yet visible to AI systems. That’s a different population that needs its own test. For pages already earning citations, though, adding JSON-LD is unlikely to be the unlock.
Condé Nast CEO: Plan As If Search Traffic Will Be Zero
Condé Nast CEO Roger Lynch said he told company teams to plan their businesses as if search traffic were zero. Lynch made the comments in an interview on TBPN, a tech talk show OpenAI acquired in April.
Key Facts
Lynch described three consecutive years in which internal forecasts underestimated the actual declines in search traffic. He expects search to settle at a single-digit percentage of total traffic, not literally zero.
Lynch pointed to a “barbell effect” in which large, authoritative brands and small, niche publications are performing well, while brands in the middle are most exposed. Condé Nast’s digital subscriptions grew 29% in revenue last year.
Why This Matters
Lynch is describing what the third-party data has been showing for months. Chartbeat reported a 60% decline in search referrals for small publishers over two years. The Reuters Institute found that media leaders expect search traffic to fall by more than 40% over three years. The difference is that a CEO running Vogue, The New Yorker, and GQ is now building budgets around those numbers.
The barbell observation is worth testing against your own client portfolio or publishing operation. Lynch’s argument is that brands without deep category authority or a strong niche focus lack a clear path forward. AI Overviews, commerce links, and sponsored results fill the page before organic listings appear.
Theme Of The Week: The Measurement Is Catching Up To The Problem
The tools and signals that defined search visibility for years are being deprecated, questioned, or abandoned by the publishers who depended on them.
FAQ rich results are gone. Schema’s role in AI citations is weaker than the correlation suggested. A major publisher is planning as if search traffic won’t recover. Each story involves an environment where the old measurement infrastructure no longer matches the landscape.
The GA4 update is the other side of that coin. Google is building native tracking for the traffic source that’s growing while the traditional one contracts.
AI assistant traffic is a fraction of what search delivers. But it’s now visible by default, in the same reports, next to the channels it’s measured against.
Top Stories Of The Week:
More Resources:
Featured Image: PeopleImages/Shutterstock; Paulo Bobita/Search Engine Journal
When a brand stops appearing in ChatGPT, or when its share of voice in Perplexity drops by half over a quarter, the typical response from the marketing org is to write more content. Sometimes a lot more. The thinking goes that if AI systems aren’t surfacing the brand, the fix is to feed them more material to work with. That instinct is a misdiagnosis. It’s a retrieval-layer fix being applied to what is increasingly a different kind of problem entirely, and the cost shows up as wasted budget, missed quarters, and a creeping sense that the work isn’t connecting to the outcomes anymore.
The mistake is treating AI visibility as a single problem when it isn’t. There are three structurally different layers between your brand and the answer a user receives, each with its own failure modes, its own fixes, and increasingly its own organizational owner. Diagnose the wrong layer, and the fix doesn’t land.
Where Most Of The Conversation Has Been Living
The first layer is retrieval. This is where the AI search optimization conversation has spent most of the last two years. The mechanics are familiar in shape if not in detail. When a model needs to answer a question grounded in real-world content, it pulls relevant material from external sources and uses that material to construct the response. The technical name is retrieval-augmented generation, or RAG, and the layer it operates on is the gateway between your content and the model’s output.
This is where crawlability, parseability, and chunk-friendliness do their work. If your content can’t be retrieved cleanly, nothing downstream matters. The visibility tracking platforms most marketing teams have evaluated this year measure outcomes that depend on this layer functioning, which is why they tend to reward the same disciplines that produced good results in classical search: structured content, schema markup, self-contained answers, clean technical implementation.
But retrieval has a structural limit, and Microsoft Research has been unusually direct about it. Plain RAG, in their words, struggles to connect the dots. It retrieves chunks of text that look relevant to the question, but it cannot reason about how those chunks relate to each other. When the answer requires synthesizing information across multiple sources, or when the question is broad enough that the right answer depends on understanding patterns across an entire dataset, retrieval alone breaks down. The model gets the chunks and has to guess at the relationships, and guessing is where hallucinations enter.
The discipline question this layer asks is straightforward. Can the model retrieve our content at all, and is it retrieving the right content for the right query? Most marketing teams have some version of this work in flight already, even if the specific tactics have shifted from classical SEO. But retrieval is only the gateway. Even when a model retrieves your content correctly, what it does with it depends on whether you exist as a recognized thing in the layer above.
Where Entity Recognition Does The Real Work
The second layer is the relationship layer, and the dominant structure on it is the knowledge graph. The major search infrastructures all maintain one. Google’s Knowledge Graph, Microsoft’s Satori, and the open knowledge graph built on Wikidata and schema.org collectively define how your brand is represented as an entity, what category you sit in, and which other entities you’re connected to.
This is the layer that decides whether AI Overviews and large language model responses treat you as a recognized member of your category, or as one fuzzy candidate string among many. Brands that exist as clean, well-defined entities get cited consistently. Brands that exist as undifferentiated tokens scattered across the open web get pattern-matched against fifty other candidates and lose more often than they win.
Knowledge graphs have been around long enough that the discipline is reasonably mature. Schema markup on owned properties, consistent naming and identifiers across the open web, structured presence on the high-trust nodes like Wikidata entries and review platforms, and the slow accumulation of brand mentions in contexts that the graph treats as authoritative. This is where the unlinked brand mentions conversation lives, because consistent contextual mentions strengthen the entity even without a hyperlink attached. The fix at this layer is structural rather than volume-based. Writing more content does almost nothing if the entity definition underneath it is fuzzy.
The discipline question here is harder than the retrieval-layer question. Are we a clean, defensible entity in our category, or are we still being pattern-matched against fifty other candidate strings? A brand that can’t answer that question affirmatively is going to lose ground in AI search, regardless of how much content it produces, because the second layer is where the model decides what your content is actually about.
The knowledge graph tells the model what your brand is. But increasingly, your brand has to function inside a third layer that most marketing teams haven’t met yet, where the model isn’t just understanding you, it’s being asked to reason about you on behalf of someone making a decision.
The Layer Enterprise Companies Are Quietly Building Right Now
The third layer is the context graph, and this one needs a careful introduction because most of the marketing conversation hasn’t reached it yet.
A context graph has the same structural shape as a knowledge graph, with entities, relationships, and typed connections, but it’s grounded differently. A knowledge graph models the world. It tells you what things are and how they relate in general. A context graph models a specific organization’s data, decisions, policies, and operational reality. The cleanest framing I’ve seen calls a knowledge graph the library and a context graph the operating manual written by the people who actually run the place. The library tells you what exists. The operating manual tells you what’s relevant, what’s authorized, and what to do about it right now. The library is read-only semantic infrastructure. The operating manual is a living operational layer that grows every time a business process executes.
What separates a context graph from anything that came before it is that governance lives inside the graph rather than alongside it. Policies, permissions, validity windows, and authorization rules are nodes the graph itself queries, not external documentation applied at the edges. When an agent retrieves something from a context graph, the result has already been filtered through what’s currently authorized, currently valid, and currently applicable. The graph is also continuously evolving, so what it knows about you this week is not necessarily what it knew last quarter. That’s where the word “governed” comes from when people in this space talk about governed retrieval. It isn’t a frame, but rather the architecture.
That architecture used to be invisible to anyone outside the organization that built it, which is why marketers haven’t had to think about it. That changed at Google Cloud Next ’26, when Google introduced the Knowledge Catalog inside its new Agentic Data Cloud. Google’s own description of the product, written in their own first-party blog content, says the Knowledge Catalog constructs a unified, dynamic context graph of your entire business, enabling you to ground agents in all of your business data and semantics. That sentence is the moment the term left the data-engineering blogs and entered enterprise procurement vocabulary.
The reason this matters for marketing is that context graphs are what’s going to power the next generation of agents inside your enterprise customers. Gartner projects that 40% of enterprise applications will be integrated with task-specific AI agents by the end of 2026, up from less than 5% in 2025. Procurement agents, competitive intelligence agents, content strategy agents, vendor evaluation agents. These agents won’t be reasoning about your brand from the open web. They’ll be reasoning about your brand from inside their company’s context graph, and what that graph says about you depends on what got ingested into it.
That ingestion is where the work for marketing lives. The brand that arrives at the context graph fragmented arrives weak. If your category positioning is inconsistent across owned and earned media, the graph picks up the contradictions and represents you ambiguously. If your entity data is fuzzy on the second layer, it stays fuzzy when it gets pulled into the third. If your third-party signal is thin or contradictory, the graph has nothing solid to anchor to. The work is upstream of the graph, but the consequences land downstream of it, inside an agent’s reasoning process that you’ll never see directly.
I think of this discipline as governed visibility. The practice of making sure your brand arrives at the context graph in a state that holds up under governed retrieval. Clean entity definition, consistent third-party representation, reliable structured data, and a category position that doesn’t fall apart when an agent traverses the relationships around it. Governed visibility isn’t a new tactic stack. It’s the result of doing the second-layer work well enough that the third layer has something solid to ingest.
The discipline question at this layer is the one most marketing teams haven’t started asking yet. When an agent inside our customer’s company is reasoning about us, what does it find, and is the version of us it finds the version we’d want it to act on?
Three layers, three different problems, three different fixes. But also three different responsibility zones, and that’s where most teams are quietly losing ground.
The Reason Most Teams Will Lose This Even Though They’re Working Hard
Each layer maps to a different organizational responsibility, and most marketing teams only own one of the three cleanly.
The retrieval layer is shared with web, dev, and sometimes IT. Marketing influences what gets published, but the infrastructure that makes content retrievable sits in someone else’s domain.
The knowledge graph layer is genuinely marketing’s territory. Schema discipline, entity definition, third-party signal, brand consistency, the slow structural work that compounds over years.
The context graph layer is where IT owns the infrastructure inside the customer’s organization, but marketing has to influence what gets ingested. The work is upstream, and the consequences land downstream, often invisibly.
The teams that win in 2026 are the ones that figured out how to operate across all three responsibility zones rather than perfecting their work on just one. Most teams I see are still optimizing their owned content, which is the retrieval layer, while losing ground on entity definition, which is the knowledge graph layer, and remaining completely absent from the context graph conversation, which is the layer where some enterprise businesses are quietly standing up right now.
The work isn’t writing more content. The work is figuring out which layer the problem actually lives on, and building the disciplines to operate on all three. Governed visibility is the third-layer discipline that marketing is going to have to develop, whether or not the term sticks. The brands that build it now will look prepared in eighteen months. The brands that don’t will be wondering why their content investments stopped producing the visibility they used to.
If any of this lands or contradicts what you’re seeing inside your own teams, I want to hear about it. Drop a comment about which layer your work has been concentrated on, where you’re seeing the gaps, or where the responsibility zones break down inside your organization. The patterns are still forming, and the conversations in the comments tend to be fresher than anything else.
A lot of the measurement frameworks for this kind of work sit in The Machine Layer, which expands the original 12 KPIs for the GenAI era into something teams can actually run against.
Last week, Cyrus Shepard published an AI citation ranking factors study, and it created a lot of noise on X, LinkedIn, and a number of private WhatsApp groups I’m in. Not only the distinction between what is a factor, and what is a correlation, especially given a lot of studies in SEO and AI are multifarious and have high levels of imponderable complexity. To be clear, this isn’t a criticism of Cyrus’s work; the study is excellent, and the correlation/causation caveat is one he makes himself explicitly.
This led me to think about the parallels with other ranking factor studies done previously, which have implied direct traffic is a considerable traditional SEO ranking factor. At the time, these studies received a lot of negative feedback, and this was again discussed by many online after the documentation in Google’s DOJ trial revealed a “popularity” signal.
It makes sense for direct traffic to be a component of how popularity is measured through Chrome. Google uses Chrome data to find new websites. It also judges a page’s “quality” based on how users interact with it after clicking, but the atomic levels of how this is done, and how much weight the variables here carry, are not public knowledge.
Direct Traffic x Popularity Correlation
Direct traffic is widely considered a symptom of good performance, not a primary driver of search rankings.
Treating direct traffic as a ranking factor leads to a misinformation loop, which encourages superficial, low-effort tactics, such as purchasing bot traffic, in a misguided attempt to boost popularity, as it’s very possible to have high levels of direct traffic and poor SEO performance.
A wider view suggests that high direct traffic is typically an indicator of a strong brand, correlating with genuine ranking factors like numerous brand searches, high-quality backlinks, and strong social engagement.
These elements are the true causes of high ranking; the direct traffic merely serves as a quantifiable measure of the brand’s overall health and success, an “all ships rise in high tides” effect.
Spikes in direct traffic, don’t correlate with Organic Search traffic. (Image from author, May 2026)
If Chrome data were a direct factor, a sudden spike in browser activity on a specific URL would immediately push it up the SERPs, and this would be a gameable exploit.
This would also be something Google would pick up as it looks to stamp out obvious manipulations of search ranking, and this would have happened many years ago.
Other Insights From The DOJ Files
NavBoost and Glue are specialized systems within Google’s infrastructure that focus on user interaction signals rather than the raw volume of direct traffic.
NavBoost looks at historical clickstream data and user behavior on search results to identify which pages are most relevant for specific queries, effectively acting as a memory of what users have found helpful.
While NavBoost focuses on traditional organic results, Glue extends those same user interaction principles to all other SERP features: knowledge panels, video carousels, image packs, and featured snippets.
They allow Google to gauge a site’s authority based on how users interact with it in the search ecosystem, independent of the user’s traffic source.
Based on what we know from various official (and unofficial) sources, research, and the general SEO hive mind, we can define popularity as a sign of brand strength characterized by user behaviors such as autocompletes and bookmarks.
It functions as a correlation to high rankings because it naturally aligns with the various signals that make a page rank.
Google may avoid using Chrome data directly as a ranking factor, choosing instead to use it as a dataset to train or validate its AI models. This we don’t know, and we likely won’t be able to prove or disprove through research.
Thank you to Ryan Jones, Mark Williams-Cook, Chris Green, Gerry White, Kristine Schachinger, Charlie Whitworth, Emina Demiri Watson, (and anyone else I’ve missed) for the fun weekend discussions on this topic.
If your organic traffic is down but your pipeline looks fine, you’re not imagining it. AI-generated answers are intercepting the journey earlier, meaning users are getting what they need from a citation or a recommendation before they ever hit your site. The click never happens. But the influence did.
That’s the measurement problem most marketing teams haven’t solved yet, and the KPIs they’re reporting on weren’t designed to catch it.
Your Brand Can Appear In 1,000 AI Responses & GA4 Shows Nothing
Citations, brand mentions, and AI recommendations don’t pass through your tag manager. They don’t fire an event in GA4 or register a session in your CRM. They happen in the interface of the AI tool, and by the time a user reaches your site, or doesn’t, the influence has already occurred.
Tracking these signals requires monitoring AI outputs directly: which queries surface your brand, in which tools, and with what frequency and context.
That’s a different data collection layer entirely from what most teams have in place.
Ways To Connect AI Signals To Business Outcomes Across Every Channel
Once you’re capturing AI visibility signals, the next problem is connecting them to outcomes.
Last-click and even multi-touch attribution models weren’t designed for journeys where the most influential touchpoint leaves no clickstream trace.
Learn: Incrementality testing, which lets you isolate the lift that AI visibility is actually driving by comparing performance across exposed and unexposed segments.
Learn: Media mix modeling, which takes a broader view, quantifying AI’s contribution alongside paid, organic, and direct channels in a single revenue model.
Used together, they give you a defensible number to bring into a budget conversation.
The Three-Layer Stack That Makes AI Search Defensible in a Budget Review
The stack works in sequence.
At the top, you’re monitoring AI visibility: citation rate, share of voice in AI responses, and brand mention frequency across tools like ChatGPT, Gemini, and Perplexity.
In the middle, incrementality and MMM translate that visibility into estimated conversion impact.
At the bottom, you’re tying those estimates to pipeline and revenue data so the whole chain holds up under scrutiny. The teams getting this right aren’t using one new metric. They’re connecting three existing disciplines, SEO, media measurement, and analytics, around a shared data model.
DAC’s Felicia Delvecchio, VP of Media, Vincent DeLuca, Director of SEO, and Gavin Bowick, Lead Web Analytics are running through exactly how that model is built in a free live session.
Condé Nast CEO Roger Lynch says he told company teams to plan their businesses as if search traffic were zero.
Lynch made the comments in an interview on TBPN, a tech talk show OpenAI acquired in April. He described three consecutive years in which internal budget forecasts underestimated actual declines in search traffic.
Lynch said:
“Each of the last three years, we would do our budgets, and we’d put forecasts in of search traffic declining… Because we’d seen the pattern of algorithm changes. And generally those algorithm changes were negative.”
“Every year, our search traffic was down more than we had forecast. So last year I told our teams, ‘Assume there’s no search.’ You have to have your businesses planned as if search is zero.”
Lynch told TBPN that Condé Nast doesn’t expect search traffic to literally reach zero. He expects it to settle at a single-digit percentage of total traffic.
What Changed
Lynch described how the search results page has changed, based on a comparison his team prepared for a recent board meeting. Lynch recalled:
“We took a snapshot of search results from seven or eight years ago. And what you saw were a few sponsored links, then the ten blue links.”
“Do the same search today, you get an AI overview, then you get rows and rows and rows of commerce links, then you get sponsored stuff.”
He noted that someone had recently asked him how search revenue could be up. “Have you done a search recently?” Lynch replied. “I basically have to go to the second page to get an organic result.”
Lynch acknowledged that changes in search traffic have affected Condé Nast’s business. The company has continued to grow revenue and profitability despite the decline, which he called a “headwind” rather than a crisis.
The Barbell Effect
Lynch described what he called a barbell effect across the Condé Nast portfolio. In his telling, large, authoritative brands and small niche publications with loyal audiences are performing well. Brands caught in the middle are the most exposed.
“Vogue has grown every year I’ve been at the company. It grows revenue, grows profitability every year,” Lynch said.
The New Yorker had its most successful year ever, he added. On the other end, Lynch pointed to Pitchfork, which represents about 1% of Condé Nast’s revenue but has a loyal audience in its category.
Lynch explained:
“If you try to be too broad, too large of an audience, this is not the era for that… You either need to be large and authoritative in a big category… or you need to be really nailing a specific niche where you have a loyal audience that’s willing to pay.”
Lynch added that brands in the middle of that barbell, those without deep authority in a category or strong enough niche focus, don’t have a clear path forward.
He added:
“If you don’t have really strong authoritative brands, or brands that have very strong niche in certain areas, or direct audiences, then you’re just going to be fighting that all the way down.”
Subscriptions As The Replacement
Condé Nast’s digital subscriptions grew 29% in revenue last year, according to Lynch. The company reported double-digit growth, which is continuing this year.
Lynch noted the company has raised subscription prices “fairly materially” over the past couple of years. He expected retention to decline with each increase. Instead, retention improved every year.
The company is also expanding subscriptions to smaller brands. Pitchfork and Tatler both launched paid digital subscriptions recently.
Why This Matters
Lynch’s comments are consistent with third-party measurements indicating that publisher search referrals are under pressure. Chartbeat data reported in March showed search referral traffic fell 60% for small publishers over two years. A Reuters Institute survey found media leaders expect search traffic to decline by more than 40% over three years.
Google’s VP of Search, Liz Reid, has reframed those losses as reductions in low-quality “bounce clicks.” Google hasn’t shared publisher-facing data to support that claim.
Lynch’s directive carries weight because of the portfolio behind it. Condé Nast operates Vogue, The New Yorker, GQ, Vanity Fair, Architectural Digest, Condé Nast Traveler, Wired, and Pitchfork, among others. When the CEO of a portfolio that includes those brands says teams should budget for zero search traffic, it gives industry data a concrete example from a major publisher.
The barbell observation matters for anyone managing a publisher caught between the two extremes. Lynch is describing a version of the pressure Chartbeat’s size-segmented data has tracked. Small and mid-tier publishers without deep category authority or direct audience relationships face the steepest declines.
Looking Ahead
Lynch told TBPN the company has started evaluating each brand’s plan for a low-search future. The company is prioritizing brands that can show a path forward without search traffic.
Lynch’s comments may put pressure on other large publishers to formalize similar planning. The trend data has been consistent enough that budgeting for search decline is already common. Budgeting for zero is a different level of preparation.
I recently spoke with an SEO who, along with his entire team, had just been laid off. The company was rapidly losing organic traffic, leadership was frustrated, and from their perspective, nothing was being done to fix it. The SEO saw it very differently. They had submitted more than 1,400 tickets over the previous 18 months, each documenting an issue and outlining the importance of what needed to be done. The backlog was extensive, detailed, and, in their mind, proof that the SEO team was working hard to reverse the decline. The problem was that none of the requested actions had been implemented. Engineering time had been consistently redirected to CEO initiatives, product launches, and other internal priorities that always seemed to matter more. From the SEO’s point of view, the work existed. From the business’s point of view, nothing had changed. Traffic declined, visibility dropped, and eventually a decision was made to eliminate this underperforming team.
A backlog is not progress. It is an unimplemented intent.
This is the uncomfortable reality many practitioners struggle to accept. Submitting tickets is not the job. Getting them implemented is. If your recommendations never make it into production, they do not exist in any meaningful way. They do not drive traffic, they do not improve visibility, and they do not protect the business as Google continues to evolve. And right now, that evolution is accelerating, which makes the gap between activity and impact even more dangerous.
Align With What Already Matters
You can see how organizations are frantically responding to the pressure to perform in AI Search, albeit subtly. Work that sat untouched for months as “SEO improvements” suddenly gets prioritized when it is reframed as AI readiness, Generative Engine Optimization, or content structuring for AI discovery. Nothing about the underlying work changes, but the framing does, because it aligns with what leadership believes matters in that moment. It may feel frustrating, even cynical, but it reveals a deeper truth.
At IBM, we struggled to get many SEO initiatives prioritized. A report later flagged our site search experience as poor and negatively impacting sales of our own search product. The required improvements were largely the same as those we had been recommending for external SEO. By relabeling them as “site search fixes” under this new mandate, we were able to accelerate implementation and improve both internal and external search performance. Work is not prioritized because it is the right thing to do. It is prioritized because it aligns with the current narrative of impact and executive priorities. To understand why so much SEO work fails to cross that threshold, you have to look at where decisions are actually made.
The Line You Don’t See Until It Stops You
After selling my agency, I took on a project for a company that was already performing well in organic search. Then Google launched paid search, and everything shifted. Large advertisers began reallocating their budgets because buying search traffic directly from Google suddenly looked more efficient than advertising on websites that simply arbitrage organic traffic to generate the ad impressions they had purchased. The board’s response was immediate and direct. They wanted to dominate every aspect of their category and be in the top three across the board, and they were willing to provide me with whatever resources were necessary to make that happen.
So I went to engineering with my plan and list of activities for total domination, expecting complete alignment and momentum. Instead, the CTO walked me to a whiteboard and pointed to a faint dotted line. Anything above that line, he explained, might get implemented this fiscal year. Anything below it would not. There was no debate or negotiation. Every idea, no matter how strategically sound, had to either fit above that line or displace something already there. It was a simple constraint of available resources, and it made one thing clear: what was already there mattered. He told me that those initiatives were also blessed by the same executives who greenlit mine. These existing initiatives were tied directly to revenue, others to compliance or security, and some were simply protected by stakeholders with enough influence to keep them in place.
That was the moment the reality became clear. This line, invisible in every audit and absent from every SEO tool, determines what actually gets built. I call it the “IT line of death.” Your mission, as an SEO or GEO manager, is to find creative ways to get your activities into or to replace one of those above-the-line projects.
From Tasks To Contribution Value
Most SEO recommendations do not fail because they are wrong. They fail because they are not competitive within that resource allocation system. This means everything is a trade-off. Engineering does not evaluate your recommendation in isolation; they evaluate it against everything else competing for their time and resources. Revenue-driving features, compliance requirements, infrastructure improvements, and existing commitments all carry weight. And so does the requester. When SEO shows up as a collection of disconnected fixes, it struggles to compete because it lacks a clearly articulated cost, ownership, and relative impact.
That realization forces a shift in how SEO needs to be approached. It is no longer enough to identify issues. You have to justify why they deserve to exist above the line and are as important as or more important than another activity. That means translating work into effort, impact, and trade-offs. It means moving from tasks to contribution value. Audits, tickets, and backlogs describe activity, but engineering teams do not fund activity. They fund outcomes. If you cannot explain why your recommendation is worth more than another team’s request, it will not get done.
This is where many SEO programs stall. They are rich in insight but weak in prioritization, and that gap becomes even more visible when you look at how work actually gets implemented. It is often difficult to tie SEO activities directly to revenue or basket size, but that does not remove the responsibility to try.
Fix The Systems, Not The Symptoms
Once you understand your organization’s IT line of death, the question becomes practical. How do you get work implemented in an environment where everything is competing? The answer is not to push harder, but to work differently within the system. In most organizations, the fastest path to implementation is not to create new work but to align with work already in motion. Engineering teams are constantly updating templates, redesigning page structures, migrating platforms, or refactoring components. Those initiatives already sit above the line. They already have a budget, attention, and momentum. When SEO is introduced as a separate request, it has to fight for priority. When it is embedded into an existing initiative, it inherits that priority. Some of the most impactful SEO changes are implemented this way, folded into broader projects rather than introduced as standalone efforts.
This becomes even more effective when you focus on scale. Isolated fixes rarely justify prioritization, but changes that act as force multipliers do. Updating a template rather than a single page can affect thousands of URLs. Adjusting CMS logic can eliminate entire categories of issues. Fixing navigation or internal linking can reshape how the entire site is understood and crawled. These are the types of changes that connect relatively small effort to large-scale impact, which makes them far more competitive at the line.
Even then, success depends on understanding the problem at its source. One of the most common failure points in SEO is diagnosing symptoms instead of causes. Large numbers create urgency, but they can also mislead. Thousands of redirects, tens of thousands of 404 errors, and duplicate pages across a site often trigger large remediation efforts, yet they are frequently just the visible output of a much smaller issue.
I worked with a company that generated pages from a product feed daily, with URLs based on the product name and its first attribute. It seemed logical, but the attribute was not stable. Every time it changed, the URL changed with it. That single design decision created a cascade of problems. New pages were constantly being created, old URLs turned into 404s, and the site effectively churned its own index. The Search Console error log reflected this chaos, filled with tens of thousands of issues that needed fixing. But none of those issues was the real problem. The solution was not to clean up the errors; it was to stop creating them. By realigning the URL structure to a stable identifier such as a SKU, the entire system stabilized. The errors disappeared because the mechanism producing them was removed. One change replaced thousands of remediation tasks.
This is the difference between work that stays below the line and work that crosses it. The former treats symptoms, the latter resolves the system that generates them. This dynamic is not unique to a single company or a single moment in time. It shows up consistently across organizations, industries, and levels of Search maturity. Whether the constraint is engineering bandwidth, compliance requirements, or competing product priorities, the outcome is the same. Work that cannot justify itself at the line does not happen. We explored this further in a podcast episode, breaking down how this pattern repeats and why so many well-intentioned initiatives stall before they ever reach production. The conclusion was consistent. Most SEO work does not fail because it is wrong; it fails because it is not framed in a way the organization can act on.
Once you understand that, the role of SEO changes. You are no longer just identifying issues; you are shaping decisions. You are defining what is worth doing, why it matters now, and what impact it will have relative to everything else competing for attention. That is what moves work from backlog to implementation.
In the end, nothing gets done because it is best practice. It gets done because it is worth doing.
Google Research published a paper that studies how to make generative AI systems produce answers that do more than sound plausible. The researchers say that their ALDRIFT framework “opens exciting avenues” for moving beyond answers that merely have a high probability.
The paper, titled “Sample-Efficient Optimization over Generative Priors via Coarse Learnability,” examines a problem in which generated answers must remain likely under a model while also moving toward a separate goal. The research points toward new avenues for addressing the AI plausibility trap.
Google ALDRIFT
The evidence in the paper centers on a framework called ALDRIFT (Algorithm Driven Iterated Fitting of Targets). The method repeatedly refines a generative model toward lower-cost answers and uses a correction step to reduce accumulated error during the process.
The paper also introduces “coarse learnability.” The term means the learned model does not need to perfectly match the ideal target. It needs to keep enough coverage over important parts of the answer space so useful possibilities are not lost too early. Under that assumption, the authors prove that ALDRIFT can approximate the target distribution with a polynomial number of samples.
ALDRIFT Operates On A Two-Part Setup
ALDRIFT operates on a two-part setup:
The generative model represents what kinds of answers remain likely under the model.
The outside scoring process measures whether a candidate answer performs well against the target goal.
The authors describe that score as a “cost.” The word “cost” refers to the measured penalty assigned to a candidate answer. A lower cost means the candidate did better according to the requirement being checked. ALDRIFT does not simply search for any low-cost answer. It searches for answers that score well while still remaining likely under the generative model.
Some AI Answers Need To Work As A Whole
The researchers are focused on AI answers for problems where the response has to function in the real world such as their examples of route planning and conference planning.
Route planning: The paper explains that an LLM may evaluate whether individual route segments are scenic, but may struggle to ensure that those segments connect into a valid path.
Conference planning: An LLM may group sessions by topic, while a classical algorithm may be needed to schedule those sessions into a timetable without conflicts.
These examples show why the paper treats plausible answers as only part of the problem. The harder issue is producing answers that remain coherent when separate parts have to work together as one complete solution.
The Coarse Learnability Assumption
The paper treats this as a problem of guiding a generative model toward answers that hold together across all their parts. The authors connect the problem to inference-time alignment, where a model is adjusted during use based on whether a specific answer works as a complete solution. That connection gives the research practical relevance, although the paper’s contribution remains theoretical and depends on the coarse learnability assumption.
The phrase “coarse learnability assumption” means the paper’s theory depends on an assumption that the model can keep enough useful possibilities available while it is being pushed toward better answers.
It does not mean the model has to learn the target perfectly. It means the model has to preserve enough coverage of the answer space so the process does not get stuck too early or lose possible better answers.
The paper identifies several gaps in how existing optimization methods are understood:
Limitation of existing methods: Classical model-based optimization methods rely on “asymptotic convergence arguments.” This means they are theoretically understood after very large amounts of sampling, but not necessarily in practical settings with limited samples.
Failure with expressive models: The paper says these classical assumptions “break down” when using expressive generative models like neural networks.
Gap in understanding: The authors say the “finite-sample behavior” of optimization in this setting is “theoretically uncharacterized.” That means the theory does not fully explain how these methods behave when only limited samples are available.
The paper’s solution is to introduce “coarse learnability” to explain how a generative model can be pushed toward better answers while keeping enough useful possibilities available along the way.
The LLM Evidence Is Limited
The paper’s main proof applies to analytic generative models, which are easier to analyze mathematically than modern LLMs. The LLM evidence is narrower: the authors use GPT-2 in simple scheduling and graph-related problems, showing behavior that supports the idea without proving that the same assumptions hold for modern LLMs.
The Research Points To A Foundation For Future Research
The paper offers a theoretical foundation for studying how generative models could be combined with external checking processes.
The research shows that Google researchers are exploring a framework for addressing the “plausible answer” problem, and the authors write that the “framework opens exciting avenues for future research.” They conclude that this research points “toward a principled foundation for adaptive generative models.”
Takeaways
The “Coverage” Requirement: Coarse learnability means the model does not have to learn the target perfectly. It needs to avoid losing useful areas of the answer space where better solutions might exist.
The Correction Step Matters: ALDRIFT uses a correction step to keep the search closer to the intended target as the model is pushed toward better answers.
Two-Part Approach: The framework uses a division of labor. The generative model handles qualitative or semantic preferences, while a separate process checks whether the answer works as a complete solution.
Limited LLM Evidence: Tests with GPT-2 showed behavior that supports the idea in simple scheduling and graph-related examples, but not proof that the same assumptions hold for modern LLMs.
Real-World Use Is The Larger Goal: The research matters to SEOs and businesses because AI answers are increasingly expected to do more than summarize information. They need to support decisions, plans, and actions that hold together outside the chat interface. While the framework is likely not being used in production, it does show Google is making progress on providing answers that are more than plausible.
The sign on AI-referred traffic conversion flipped. I’m not sure if enough of us have noticed.
Twelve months ago, visitors arriving at U.S. retailers from AI assistants converted at roughly half the rate of visitors from other channels. In March 2026, they converted 42% better. Same channel. Same stores. Different year.
Adobe Analytics published the 2026 Q2 AI Traffic Report on April 16 (Adobe’s fiscal Q2 covers calendar Q1 2026). The growth numbers land first: AI-referred traffic to U.S. retailers grew 393% year-over-year in Q1 2026, peaking at 1,151% YoY in December. Engagement up 12%, time spent up 48%, pages per visit up 13%, revenue per visit up 37%. All measured against non-AI traffic in March 2026, using Adobe’s own analytics data from retailers running on the Adobe platform.
The real story is the conversion sign flip. The channel went from worst-performing in U.S. retail to best-performing. In 12 months.
If you run or optimize a website, this changes which number actually matters to you.
One caveat worth naming up front. Adobe publishes this report alongside Adobe LLM Optimizer, a product they sell for making websites more visible to AI assistants. The research and the product roll out together, and the link sits inside the report itself. The underlying numbers are Adobe’s own, self-reported from their analytics platform, and the kind of data that would be hard to fake and easy to challenge if it weren’t accurate. But the framing should be read knowing the vendor also sells the tool that addresses the problem the report describes. Thanks to Els Aerts for flagging this.
2026 Adobe Report Suggests AI Traffic Converts Better Than Non-AI Traffic
This is not something slowly getting better. This is something that’s gone from pretty much broken to kind of working.
Maturation would look like half the non-AI rate to 25% worse to 10% worse to break-even to slight edge. Three, four years of grind. Slow curve. Predictable report cycles. That’s what maturation normally looks like for a new channel. Paid search did that. Mobile did that. Social did that. AI-referred traffic is not doing that. Two measurement checkpoints twelve months apart, sign flipped. Different kind of event.
The playbooks calibrated to “AI traffic is early, optimize gradually, the channel isn’t mature yet” are calibrated to the wrong curve. Any agency, consultant, or vendor still saying “early stage” or “not ready” about AI retail traffic hasn’t read this month’s numbers. The tell is in the timeline they propose. If the pitch is “let’s learn what works over the next year,” they missed the flip.
They’re working from a brief that’s twelve months out of date.
Why AI Agents Fail To Parse Non-Readable Retail Websites
Adobe’s report dedicates an entire section to what they call Citation Readability: how well a page can be understood, parsed, and surfaced by AI systems. The gap between top and bottom performers is brutal. Homepages from top-AI-visit-share retailers score 62% higher than the bottom. Search results pages, 32% higher. Blog and editorial content, 30% higher.
Read that as an operator’s diagnostic. Adobe is telling you why the growth is uneven.
The 393% aggregate is what’s getting through despite readability gaps. Retailers whose pages AI models can actually parse and cite are pulling the average up. Retailers whose pages AI can’t read reliably are dragging it down.
Most website owners don’t even know their website isn’t entirely readable by machines.
Not “we know we’re behind on AI.” Not “we’re testing.” Website owners who run their analytics every morning, review conversion rates every week, argue about CRO every quarter, have no visibility into what a GPTBot, ClaudeBot, or PerplexityBot sees when it crawls their product page. Their dashboards don’t show when an AI indexer fetched a shell. Their session recordings don’t capture bots. Their attribution rarely tags AI referrals cleanly.
The real conversion lift on websites that are actually machine-readable is higher than the aggregate suggests. The average is being held down by everyone else.
Comparing Dell’s Internal Data Vs. Adobe’s AI Traffic Trends
Eight days before Adobe published this data, Dell’s head of global consumer revenue programs told Digital Commerce 360 that agentic shopping is delivering “nothing to the point that is earth-shaking” yet.
Both things are true at the same time.
There’s a chance Dell’s website is bad. It’s not that the entire industry of AI-assisted shopping is wrong. Dell was measuring one website. Adobe was measuring aggregate traffic across many retailers. Dell looked at their own conversion data, saw flat numbers, published the number. Adobe looked at the set of websites AI models can read and cite, saw a channel inversion, published that.
If your conversion numbers look like Dell’s, don’t wait for the channel to mature. Audit the website. Dell’s admission is a diagnostic about dell.com. Adobe’s data is about where the channel is going. Don’t confuse them.
How AI-Assisted Research Shortens The Purchase Funnel
Traffic growth the way we were trained to think about it in the last 30 years, that doesn’t matter at all anymore.
Impressions. Sessions. Unique visitors. Page views. The vocabulary that defined SEO and CRO practice from 1998 to 2024. All of it assumed traffic meant humans arriving to decide. You grew top-of-funnel, so more humans entered deliberation. You optimized the funnel so more of them converted. That was the arithmetic.
AI-referred traffic doesn’t work like that.
When someone clicks through from ChatGPT, Perplexity, or Gemini, they’ve already done their research inside the assistant. They compared options. They asked follow-up questions. They landed on a shortlist. The click to your website is the last step in a decision, not the first. Adobe’s numbers reflect this: 12% higher engagement, 48% longer time per visit, 37% higher revenue per visit. That’s not a better funnel. It’s a shorter funnel. Most of the consideration happened off your website.
If you’re optimizing for volume (more impressions, more sessions, more referrals), you’re optimizing for the old economy. The retailers winning this 393% growth are the ones the AI assistants actually cite, link to, and send pre-qualified buyers to. That’s a legibility problem, not a visibility one.
Technical Audit For AI Crawlers And JavaScript Readability
Two things you can verify this weekend, without tools, without a team, without budget.
Disable JavaScript. Fresh browser profile, JavaScript off, reload a product page. Is the price there in the HTML? The name? The stock status? The buy button? Most AI crawlers that index pages for citation don’t execute JavaScript, or execute it inconsistently. If the critical facts need JavaScript to render, the AI can’t cite what it can’t see, and your page won’t surface as a reference in the assistant’s answer.
Check the answer-first test. Does your product page lead with what the thing is, what it costs, and whether it’s available? Or does it lead with brand nav, hero imagery, lifestyle copy, and a carousel? AI models retrieving and summarizing your page pick up the first dense, structured facts they find. Humans tolerate brand theater. AI indexers don’t scroll past it to find the price.
If both check out, flat AI numbers are a distribution problem. You’re not being referred. Work on that separately. If either fails, it’s an architecture problem. The 393% is passing you by.
Legibility Vs. Optimization For AI Referral Traffic
AI-referred traffic doesn’t reward optimization. It rewards legibility. Those are not the same thing.
