AI Mode is Google’s most powerful AI search experience, providing answers to complex questions in a way that anticipates the user’s information needs. Although Google says that nothing special needs to be done to rank in AI Mode, the reality is that SEO only makes pages eligible to appear.
The following facts, insights, and examples demystify AI Mode and offer a clear perspective on how pages are ranked and why.
What Is AI Mode?
Google’s AI Mode was introduced on March 5, 2025, as an experiment in Google Labs, then swiftly rolled out as a live Google search surface on May 20. AI Mode is described as its most cutting-edge search experience, combining advanced reasoning with multimodality. Multimodality means content beyond text data, such as images and video content.
AI Mode is a significant evolution of Google Search that encourages users to research topics. This presents benefits and changes to how search works:
- The benefit is that Google is citing a greater variety of websites per query.
- The change is that websites are being cited for multiple queries, beginning with the initial query plus follow-up queries.
Those two factors present challenges to SEO. For example, do you optimize for the initial query, or what can be considered a more granular follow-up query? Most SEOs may consider optimizing for both.
Query Fan-Out
Similar to AI Overviews, AI Mode uses what they call a query fan-out technique, which divides the initial search query into subtopics that anticipate further information the user may need.
Query fan-out anticipates the user’s information journey. So, if they ask question A, Google’s AI Mode will show answers to follow-up questions about B, C, and D.
For example, if you ask, “What is a mechanical keyboard?” Google answers the following questions:
- What is a mechanical keyboard?
- What are mechanical switches?
- What happens when a key is pressed on a mechanical keyboard?
- What are keycaps and what materials are they made from?
- What is the role of the printed circuit board (PCB)?
- How are mechanical switches categorized?
The following screenshot of the AI Mode search result shows the questions (in red) positioned next to the answers, illustrating how query fan-out generates related questions and creates answers for them.
How I Extracted Latent Questions From AI Mode Search Results
The way I extracted the questions that query fan-out is answering was by doing an inverse knowledge search, also known as reverse QA.
I copied the output from AI Mode into a document, then uploaded it to ChatGPT with the following prompt:
Read the document and extract a list of questions that are directly and completely answered by full sentences in the text. Only include questions if the document contains a full sentence that clearly answers it. Do not include any questions that are answered only partially, implicitly, or by inference.Try that with AI Mode to get a better understanding of the underlying questions it generates with query fan-out. This will help clarify what is happening and make it less mysterious.
Content With Depth
Google’s advice to publishers who want to rank in AI Mode is to encourage them to create content that engages users who are conducting in-depth queries:
“…users are asking longer and more specific questions – as well as follow-up questions to dig even deeper.”
That may not mean creating giant articles with depth. It just means focusing on the content that users are looking for. That approach to content is subtly different from chasing keyword inventory.
Google recommends:
- Focus on unique, valuable content for people.
- Provide a great page experience.
- Ensure we can access your content.
- Manage visibility with preview controls. (Make use of nosnippet, data-nosnippet, max-snippet, or noindex to set your display preferences.)
- Make sure structured data matches the visible content.
- Go beyond text for multimodal success.
- Understand the full value of your visits.
- Evolve with your users.
The last two recommendations require further clarification:
Understand The Full Value Of Your Visits
This is an encouragement to focus on delivering the information needs of the user and to note that focusing too hard on the “click” comes at the expense of providing what an “engaged” audience is looking for.
Evolve With Your Users
Google frames this as evolving along with how users are searching. A more pragmatic view is to evolve with how Google is showing results to users.
What Experts Say About Content Structure For AI Mode
Duane Forrester, formerly of Bing Search, advises that content needs to be structured differently for AI search.
He advises:
“…the search pipeline has changed. You don’t need to rank – you need to be retrieved, fused, and reasoned over by GenAI systems.”
In his article titled “Search Without A Webpage,” he expands on the idea that content must be useful as forming the basis of an answer:
“…your content doesn’t have to rank. It has to be retrieved, understood, and assembled into an answer.”
He also says that content needs to be:
“…structured, interpretable, and available when it’s time to answer.
This is the new search stack. Not built on links, pages, or rankings – but on vectors, embeddings, ranking fusion, and LLMs that reason instead of rank.”
When Duane says that content needs to be structured, he’s referring to on-page structure that communicates not just the hierarchy of information but also offers a clean delineation of what each section of content is about.
In my opinion:
- Paragraphs should consist of sentences that build to an idea, with a clear payoff at the end.
- If a sentence doesn’t have a purpose within the paragraph, it’s probably better to remove it.
- If a paragraph doesn’t have a clear purpose, get rid of it.
- If a group of paragraphs is out of place near the end of the document, move it closer to the beginning if that’s where it belongs.
- The entire document should have a clear beginning, middle, and end, with each section serving as “the basis of an answer.”
Itai Sadan, CEO of Duda, recommends:
“Use clear, specific language: LLMs rely on clarity first and foremost, so avoid using too many pronouns or any other vague, undefined references.
Organize your content predictably: Break your content up into sections and use headings, like H2 and H3, to organize the unique ideas central to your article’s thesis.”
Mordy Oberstein, founder of Unify Marketing, explains that the focus on attribution took precedence for the average digital marketer:
“What resonates with the person hasn’t fundamentally changed, and I don’t think we’ve realized that. I think we’ve forgotten. I think we’ve completely forgotten what resonance is as digital marketers because of the advent of two things with the internet:
- Attribution
- The ability to track responses
Businesses were seemingly OK with digital marketers doing whatever it took to get that traffic, to get that conversion, because that’s just the Internet, so everyone just goes along.
Now, with AI Mode, attribution no longer exists in the same way.”
Mordy’s right about attribution. AI Mode cannot be tracked in Google Analytics 4 or Google Search Console. They’re lumped into the Web Search bucket, so there’s no way to tell where it’s coming from. It can’t be distinguished from regular organic search in either GA4 or GSC.
The attribution question is a big issue for digital marketers. Michael Bonfils of Digital International Group recently discussed the issue of attribution from the perspective of zero-click searches.
Bonfils says:
“But the organic side, there is an area … that is zero click. So zero click is for those audience members who don’t know what that means, zero click means when you are having a conversation with AI, for example, I’m trying to compare two different running shoes and I’m having this, ‘what’s going to be better for me?’
I’m having a conversation with AI and AI is pooling and referencing … whatever winning schema formats and content that are out there … but it’s zero click. It’s not going to your site. It’s not going there. So without this data that really affects … organic content strategy.”
And that dovetails with what Mordy is getting at, that SEOs are conditioned to view internet marketing through the “attribution” lens, but that we may be entering a kind of post-attribution period, which is what it largely was pre-internet. So, the old marketing strategies are back in, but they were always good strategies (building awareness and popularity); it’s just that digital marketers tended to engage more with attribution.
Mordy shares the example of someone researching a brand of sneakers, who asks a chatbot about it, then goes to Amazon to see what it looks like and what people are saying about it, then watches video reviews on YouTube, and then goes to AI Mode to review the specs. After all that research, the consumer might return to Amazon and then head over to Google Shopping to compare prices.
He concludes with the insight that resonating with users has always been important, and that very little has changed in terms of consumers conducting research prior to making a purchase:
“That was all happening before. But now the perception is that it’s happening because of LLMs. I don’t think things have fundamentally changed.”
I think that the key insight here is that the research is still happening exactly as before, but what’s changed is that the opportunities to expose your business or products have expanded to multimodal search surfaces, especially with AI Mode.
The screenshot below shows how Nike is taking charge of the conversation on AI Mode with both text and video content.
Connect Your Brand To A Product
It’s becoming evident that connecting a brand semantically to a service or product may be important for communicating that the brand is relevant to whatever you want it to be relevant for.
Below is a screenshot of a sponsored post that’s indexed by Google and is ranking in AI Mode for the keyword phrase “what are ad hijacking tools.”
SEO Makes Content Eligible For AI Mode
SEO best practices are necessary to be eligible to appear in AI Mode. That’s different from saying that standard SEO will help you rank in AI Mode.
This is what Google says:
“To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. There are no additional technical requirements.”
The “Search technical requirements” are just the three basics of SEO:
- “Googlebot isn’t blocked.
- The page works, meaning that Google receives an HTTP 200 (success) status code.
- The page has indexable content.”
Google clearly says that foundational SEO is necessary to be eligible to rank in AI Mode. But it does not explicitly confirm that SEO will help a site rank in AI Mode.
Is SEO Enough For AI Mode?
Google and Googlers have reassured publishers and SEOs that nothing extra needs to be done to rank in AI search surfaces. They affirm that standard SEO practices are enough.
Standard SEO practices ensure that a site is crawled, indexed, and eligible for ranking in AI Mode. But there is implication that the signals for actually ranking in AI Mode are substantially different from standard organic search.
What Is FastSearch?
Information contained in recent Google antitrust court documents shows that AI Mode ranks pages with a technology called FastSearch.
FastSearch grounds Google’s AI search results in facts, including data from the web. This is significant because FastSearch uses different ranking signals from what’s used in the regular organic search, prioritizing speed and selecting only a top few pages for AI grounding.
The recent Google antitrust trial document from early September offers this explanation of FastSearch:
“To ground its Gemini models, Google uses a proprietary technology called FastSearch. … FastSearch is based on RankEmbed signals—a set of search ranking signals—and generates abbreviated, ranked web results that a model can use to produce a grounded response. …
FastSearch delivers results more quickly than Search because it retrieves fewer documents, but the resulting quality is lower than Search’s fully ranked web
results. “
And elsewhere in the same document:
“FastSearch is a technology that rapidly generates limited organic search results for certain use cases, such as grounding of LLMs, and is derived primarily from the RankEmbed model.”
RankEmbed
RankEmbed is a deep learning model that identifies patterns in datasets and develops signals that are used for ranking purposes. It uses a combination of user data from search logs and scores generated by human raters to create the ranking-related signals.
The court document explains:
“RankEmbed and its later iteration RankEmbedBERT are ranking models that rely on two main sources of data: __% of 70 days of search logs plus scores generated by human raters and used by Google to measure the quality of organic search results.
The RankEmbed model itself is an AI-based, deep learning system that has strong natural-language understanding. This allows the model to more efficiently identify the best
documents to retrieve, even if a query lacks certain terms.”
Human-Rated Data
The human-rated data, which is part of RankEmbed, is not used to rank webpages. Human-rated data is used to train deep learning models so they can recognize patterns that correlate with high and low-quality webpages.
How human-rated data is used in general:
- Human-rated data is used to create what are called labeled data.
- Labeled data are examples that models use to identify patterns in vast amounts of data.
In this specific instance, the human-labeled data are examples of relevance and quality. The RankEmbed deep learning model uses those examples to learn how to identify patterns that correlate with relevance and page quality.
Search Logs And User Behavior Signals
Let’s go back to how Google uses “70 days of search logs” as part of the RankEmbed deep learning model, which underpins FastSearch.
Search logs refer to user behavior at the point when they’re searching. The data is rich with a wide range of information, such as what users mean when they search, and it can also include the domain names of businesses they associate with certain keywords.
The court documentation doesn’t say all the ways this data can be used. However, a Google antitrust document from May 2025 revealed that search log (click) patterns only become meaningful when scaled to the billions.
Some SEOs have theorized that click data can directly influence the rankings, describing a granular use of clicks for ranking. But that may not be how click data is used, because it’s too noisy and imprecise.
What’s really happening is more scaled than granular. Patterns reveal themselves in the billions, not in the individual click. That’s not just my opinion; it’s a fact confirmed in the May 2025 Google antitrust exhibit:
“Some Known Shortcomings of Live Traffic Eval
The association between observed user behavior and search result quality is tenuous. We need lots of traffic to draw conclusions, and individual examples are difficult to interpret.”
It’s fair to say that search logs are not used to directly impact the rankings of an individual webpage, but are used to learn about relevance and quality from user behavior.
FastSearch is not the same ranking algorithm as the one used for organic search results. It is based on RankEmbed, and the term “embed” suggests that embeddings are involved. Embeddings map words into a vector space so that the meaning of the text is captured. For SEO, this means that keyword relevance matters less, and topical relevance and semantic meaning carry more weight.
Google’s statement that standard SEO is all that’s needed to rank in AI Mode is true only to the extent that standard SEO will ensure that the webpage is crawled, indexed, and eligible for the final stage of AI Mode ranking, which is FastSearch.
But FastSearch uses an entirely different set of considerations at the LLM level to decide what will be used to answer the question.
In my opinion, it’s more realistic to say that SEO best practices make webpages eligible to appear in AI Mode, but the ranking processes are different, and so new considerations come into play.
SEO is still important, but it may be useful to focus on semantic and topical relevance.
AI Mode Is Multimodal
AI Mode is multimodal, meaning image and video content rank in AI Mode. That’s something that SEOs and publishers need to consider in terms of how user expectations drive content discovery. This means it may be useful to create image, video, and maybe even audio content in addition to text.
Optimizing Images For AI Mode
Something that is under your control is the featured image and the in-content images that go with your content. The best images, in my opinion, are images that are noticeable when displayed in AI Mode and contain visual information that is relevant to the search query.
Here’s a screenshot of images that accompany the cited webpages for the query, “What is a mechanical keyboard?”
As you can see, none of the images pop out or call attention to themselves. I don’t think that’s Google’s preference; that’s just what publishers use. Images should not be an afterthought. Make them an integrated part of your ranking strategy for AI Mode.
Creative use of images, in my opinion, can help a page call attention to itself as useful and relevant. The best images are ones that look good when Google crops them into a square format.
Google AI Mode is multimodal, which means optimizing your images so that they display well in AI Mode search results. Your images should be attractive regardless of whether they are displayed as either a rectangle (approximately 16:9 aspect ratio) or a square (approximately 4:3 aspect ratio).
Mordy Oberstein offers these insights on multimodal marketing:
“AI Mode is looking at videos, images, and yes, you could do all of that. Yes, you should do all of that – whatever is possible to do while being efficient and not getting misdirected or losing focus – yes, go ahead. I’m all for creating authoritativeness through content. I think that’s an essential strategy for pretty much any business.
AI Mode is not just looking at your website content, whether it’s your image content, audio content, whatever it may be, it’s also looking at how the web is talking about you.”
AI Mode Is Evolution, Not Extension
AI Mode is not just an extension of traditional search but an evolution of it. Search now includes text, images, and video. It anticipates follow-up queries and displays the answers to them using the query fan-out technique. This shifts the SEO focus away from keyword inventory and chasing clicks and toward considering how the entire user information journey is best addressed and then crafting content that satisfies that need.
More Resources:
Featured Image: Jirsak/Shutterstock
