Google Announces A New Era For Voice Search via @sejournal, @martinibuster

Google announced an update to its voice search, which changes how voice search queries are processed and then ranked. The new AI model uses speech as input for the search and ranking process, completely bypassing the stage where voice is converted to text.

The old system was called Cascade ASR, where a voice query is converted into text and then put through the normal ranking process. The problem with that method is that it’s prone to mistakes. The audio-to-text conversion process can lose some of the contextual cues, which can then introduce an error.

The new system is called Speech-to-Retrieval (S2R). It’s a neural network-based machine-learning model trained on large datasets of paired audio queries and documents. This training enables it to process spoken search queries (without converting them into text) and match them directly to relevant documents.

Dual-Encoder Model: Two Neural Networks

The system uses two neural networks:

1. One of the neural networks, called the audio encoder, converts spoken queries into a vector-space representation of their meaning.
2. The second network, the document encoder, represents written information in the same kind of vector format.

The two encoders learn to map spoken queries and text documents into a shared semantic space so that related audio and text documents end up close together according to their semantic similarity.

Audio Encoder

Speech-to-Retrieval (S2R) takes the audio of someone’s voice query and transforms it into a vector (numbers) that represents the semantic meaning of what the person is asking for.

The announcement uses the example of the famous painting The Scream by Edvard Munch. In this example, the spoken phrase “the scream painting” becomes a point in the vector space near information about Edvard Munch’s The Scream (such as the museum it’s at, etc.).

Document Encoder

The document encoder does a similar thing with text documents like web pages, turning them into their own vectors that represent what those documents are about.

During model training, both encoders learn together so that vectors for matching audio queries and documents end up near each other, while unrelated ones are far apart in the vector space.

Rich Vector Representation

Google’s announcement says that the encoders transform the audio and text into “rich vector representations.” A rich vector representation is an embedding that encodes meaning and context from the audio and the text. It’s called “rich” because it contains the intent and context.

For S2R, this means the system doesn’t rely on keyword matching; it “understands” conceptually what the user is asking for. So even if someone says “show me Munch’s screaming face painting,” the vector representation of that query will still end up near documents about The Scream.

According to Google’s announcement:

“The key to this model is how it is trained. Using a large dataset of paired audio queries and relevant documents, the system learns to adjust the parameters of both encoders simultaneously.

The training objective ensures that the vector for an audio query is geometrically close to the vectors of its corresponding documents in the representation space. This architecture allows the model to learn something closer to the essential intent required for retrieval directly from the audio, bypassing the fragile intermediate step of transcribing every word, which is the principal weakness of the cascade design.”

Ranking Layer

S2R has a ranking process, just like regular text-based search. When someone speaks a query, the audio is first processed by the pre-trained audio encoder, which converts it into a numerical form (vector) that captures what the person means. That vector is then compared to Google’s index to find pages whose meanings are most similar to the spoken request.

For example, if someone says “the scream painting,” the model turns that phrase into a vector that represents its meaning. The system then looks through its document index and finds pages that have vectors with a close match, such as information about Edvard Munch’s The Scream.

Once those likely matches are identified, a separate ranking stage takes over. This part of the system combines the similarity scores from the first stage with hundreds of other ranking signals for relevance and quality in order to decide which pages should be ranked first.

Benchmarking

Google tested the new system against Cascade ASR and against a perfect-scoring version of Cascade ASR called Cascade Groundtruth. S2R beat Cascade ASR and very nearly matched Cascade Groundtruth. Google concluded that the performance is promising but that there is room for additional improvement.

Voice Search Is Live

Although the benchmarking revealed that there is some room for improvement, Google announced that the new system is live and in use in multiple languages, calling it a new era in search. The system is presumably used in English.

Google explains:

“Voice Search is now powered by our new Speech-to-Retrieval engine, which gets answers straight from your spoken query without having to convert it to text first, resulting in a faster, more reliable search for everyone.”

Read more:

​​Speech-to-Retrieval (S2R): A new approach to voice search

Featured Image by Shutterstock/ViDI Studio

Google Says What Content Gets Clicked On AI Overviews via @sejournal, @martinibuster

Google’s Liz Reid, Vice President of Search, recently said that AI Overviews shows what kind of content makes people click through to visit a site. She also said that Google expanded the concept of spam to include content that does not bring the creator’s perspective and depth.

People’s Preferences Drives What Search Shows

Liz Reid affirmed that user behavior tells them what kinds of content people want to see, like short-form videos and so on. That behavior causes Google to want to show that to them and that the system itself will begin to learn and adjust to the kinds of content (forums, text, video, etc.) that they prefer.

She said:

“…we do have to respond to who users want to hear from, right? Like, we are in the business of both giving them high quality information, but information that they seek out. And so we have over time adjusted our ranking to surface more of this content in response to what we’ve heard from users.

…You see it from users, right? Like we do everything from user research to we run an experiment. And so you take feedback from what you hear, from research about what users want, you then test it out, and then you see how users actually act. And then based on how users act, the system then starts to learn and adjust as well.”

The important insight is that user preferences play an active role in shaping what appears in AI search results. Google’s ranking systems are designed to respond not just to quality but to the types of content users seek out and engage with. This means that shifts in user behavior related to content preferences directly influence what is surfaced. The system continuously adapts based on real-world feedback. The takeaway here is that SEOs and creators should actively gauge what kind of content users are engaging with and be ready to pivot in response to changes.

The conversation is building up toward where Reid says exactly what kinds of content engages users, based on the feedback they get through user behavior.

AI-Generated Is Not Always Spam

Liz next affirms that AI-generated content where she essentially confirms that the bar they’re using to decide what’s high and low quality is agnostic to whether the content is created by a human or an AI.

She said:

“Now, AI generated content doesn’t necessarily equal spam.

But oftentimes when people are referring to it, they’re referring to the spam version of it, right? Or the phrase AI slop, right? This content that feels extremely low value across, okay? And we really want to make an effort that that doesn’t surface.”

Her point is pretty clear that all content is judged by the same standard. So if content is judged to be low quality, it’s judged based on the merits of the content, not by the origin.

People Click On Rich Content

At this point in the interview Reid stops talking about low quality content and turns to discussing the kind of content that makes people click through to a website. She said that user behavior tells them that users don’t want superficial content and that the click patterns shows that more people click through to content that has depth, expresses a unique perspective that does not mirror what everyone else is saying and that these kinds of content engages users. This is the kind of content that gets clicks on AI search.

Reid explained:

“But what we see is people want content from that human perspective. They want that sense of like, what’s the unique thing you bring to it, okay? And actually what we see on what people click on, on AI Overviews, is content that is richer and deeper, okay?

That surface-level AI generated content, people don’t want that because if they click on that, they don’t actually learn that much more than they previously got. They don’t trust the result anymore.

So what we see with AI Overviews is that we surface these sites and get fewer what we call bounce clicks. A bounce click is like you click on your site, Yeah, I didn’t want that, and you go back.

AI Overviews gives some content, and then we get to surface deeper, richer content, and we’ll look to continue to do that over time so that we really do get that creator content and not the AI generated.”

Reid’s comments indicate that click patterns indicate content offering a distinct perspective or insight derived from experience performs better than low-effort content. This indicates that there is intention within AI Overviews to not amplify generic output and to uprank content that demonstrates a firm knowledge of the topic.

Google’s Ranking Weights

Here’s an interesting part that explains what gets up-ranked and down-ranked, expressed in a way I’ve not seen before. Reid said that they’ve extended the concept of spam to also include content that repeats what’s already well known. She also said that they are giving more ranking weight to content that brings a unique perspective or expertise to the content.

Here Reid explains the downranking:

“Now, it is hard work, but we spend a lot of time and we have a lot of expertise built on this such that we’ve been able to take the spam rate of what actually shows up, down.

And as well as we’ve sort of expanded beyond this concept of spam to sort of low-value content, right? This content that doesn’t add very much, kind of tells you what everybody else knows, it doesn’t bring it…”

And this is the part where she says Google is giving more ranking weight to content that contains expertise:

“…and tried to up-weight more and more content specifically from someone who really went in and brought their perspective or brought their expertise, put real time and craft into the work.”

Takeaways

How To Get More Upranked On AI Overviews

1. Create “Richer and Deeper” Content

Reid said, “people want content from that human perspective. They want that sense of like, what’s the unique thing you bring to it, okay? And actually what we see on what people click on, on AI Overviews, is content that is richer and deeper, okay?”

Takeaway:
Publish content that shows original thought, unique insights, and depth rather than echoing what’s already widely said. In my opinion, using software that analyzes the content that’s already ranking or using a skyscraper/10x content strategy is setting yourself up for doing exactly the opposite of what Liz Reid is recommending. A creator will never express a unique insight by echoing what a competitor has already done.

2. Reflect Human Perspective

Reid said, “people want content from that human perspective. They want that sense of like, what’s the unique thing you bring to it.”

Takeaway: Incorporate your own analysis, experiences, or firsthand understanding so that the content is authentic and expresses expertise.

3. Demonstrate Expertise and Craft

Reid shared that Google is trying “to up-weight more and more content specifically from someone who really went in and brought their perspective or brought their expertise, put real time and craft into the work.”

Takeaway:
Effort, originality, and subject-matter knowledge are the qualities that Google is up-weighting to perform better within AI Overviews.

Reid draws a clear distinction between content that repeats what is already widely known and content that adds unique value through perspective or expertise. Google treats superficial content like spam and lowers the weights of the rankings to reduce its visibility, while actively “upweighting” content that demonstrates effort and insight, what she termed as the craft. Craft means skill and expertise, mastery of something. The message here is that originality and actual expertise are important for ranking well, particularly in AI Overviews and I would think the same applies for AI Mode.

Watch the interview from about the 18 minute mark:

Google Reminds SEOs How The URL Removals Tool Works via @sejournal, @martinibuster

Google’s John Mueller answered a question about removing hacked URLs that are showing in the index. He explained how to remove the sites from appearing in the search results and then discussed the nuances involved in dealing with this specific situation.

Removing Hacked Pages From Google’s SERPs

The person asking the question was the victim of the Japanese hacking attack, so-called because the attackers create hundreds or even thousands of rogue Japanese language web pages. The person had dealt with the issue and removed the spammy infected web pages, leaving them with 404 pages that are still referenced in Google’s search results.

They now want to remove them from Google’s search index so that the site is no longer associated with those pages.

They asked:

“My site recently got a Japanese attack. However, I shifted that site to a new hosting provider and have removed all data from there.

However, the fact is that many Japanese URLs have been indexed.

So how do I deindex those thousands of URLs from my website?”

The question reflects a common problem in the aftermath of a Japanese hack attack, where hacked pages stubbornly remain indexed long after the pages were removed. This shows that site recovery is not complete once the malicious content is removed; Google’s search index needs to clear the pages, and that can take a frustratingly long time.

How To Remove Japanese Hack Attack Pages From Google

Google’s John Mueller recommended using the URL Removals Tool found in Search Console. Contrary to the implication inherent in the name of the tool, it doesn’t remove a URL from the search index; it just removes it from showing in Google’s search results faster if the content has already been removed from the site or blocked from Google’s crawler. Under normal circumstances, Google will remove a page from the search results after the page is crawled and noted to be blocked or gone (404 error response).

Three Prerequisites For URL Removals Tool

1. The page is removed and returns a 404 or 410 server response code.
2. The URL is blocked from indexing by a robots meta tag:
3. The URL is prevented from being crawled by a robots.txt file.

Google’s Mueller responded:

“You can use the URL removal tool in search console for individual URLs (also if the URLs all start with the same thing). I’d use that for any which are particularly visible (check the performance report, 24 hours).

This doesn’t remove them from the index, but it hides them within a day. If the pages are invalid / 404 now, they’ll also drop out over time, but the removal tool means you can stop them from being visible “immediately”. (Redirecting o 404 are both ok, technically a 404 is the right response code)”

Mueller clarified that the URL Removals Tool does not delete URLs from Google’s index but instead hides them from search results, faster than natural recrawling would. His explanation is a reminder that the tool has a temporary search visibility effect and is not a way to permanently remove a URL from Google’s index itself. The actual removal from the search index happens after Google verifies that the page is actually gone or blocked from crawling or indexing.

Featured Image by Shutterstock/Asier Romero