Start With What Google Actually Says
In May 2026 Google published a dedicated guide to optimizing for generative AI features in Search. It is the first authoritative source on this topic, and it is worth reading before any third-party GEO advice, because a good deal of that advice is now contradicted by it.
The guide's central claim is blunt: there are no additional requirements to appear in AI Overviews or AI Mode, and no special optimizations for them. To be eligible as a supporting link, a page must be indexed and eligible to be shown in Search with a snippet. That is the whole technical bar.
That is genuinely good news, because it means the work is familiar. It also means several popular tactics are a waste of a sprint. This post covers what is left once you remove them.
What You Can Stop Doing
Google explicitly names these as unnecessary:
- Machine-readable files for AI. In Google's words, you do not need to create new machine readable files, and Google Search itself does not use them. That includes llms.txt.
- Special schema markup for AI. Structured data is not required for generative AI search, and there is no special schema.org markup for it.
- Chunking your content. There is no requirement to break your content into tiny pieces.
- Hitting a word count. There is no ideal page length.
- Writing in an "AI-friendly" voice. You do not need to write in a specific way just for generative AI search.
- Chasing mentions. Seeking inauthentic mentions across the web is, in Google's phrasing, not as helpful as it might seem.
Structured data is still worth implementing, but for its actual job: rich result eligibility and helping Search understand entities and relationships. Treating it as a citation lever is the mistake.
1. Understand Query Fan-Out
This is the mechanism that should change how you plan content. Google documents that AI Overviews and AI Mode use a query fan-out technique: rather than answering the query you typed, the system issues multiple related searches across subtopics and data sources, then synthesizes across the results. Because more supporting pages are identified during generation, a wider and more diverse set of links can surface than in a classic web search.
The practical consequence is that the unit of competition is the topic, not the keyword. A page can be pulled into an answer for a query it does not target, because it was the best result for one of the fanned-out subqueries. Comprehensive topic coverage is not a soft "authority" idea here. It is how you become eligible for more of the fan-out.
2. Answer the Question, Early and Plainly
This advice survives the 2026 guidance, but for a different reason than usually given. It is not that models read your page in chunks and score them. It is that a page which states its answer clearly is easier for any reader, human or machine, to extract a correct answer from, and Google's quality guidance has always rewarded that.
So: put the answer near the top of the section that promises it, then support it. Keep claims self-contained enough to be quoted without surrounding context. This costs nothing and helps regardless of which surface is reading.
What you should not do is shred a coherent article into disconnected fragments because someone told you models prefer chunks. Google says that is not required, and it makes your page worse for the humans who do click.
3. Publish Something That Is Not a Commodity
Google's guide leans hard on this, and it is the one instruction with real teeth. The recommendation is to create non-commodity content: material that is not simply a recycling of what is already indexed, and that a generative model could not have produced by itself. The examples given are first-hand ones, such as a review based on personal experience.
This is the uncomfortable part of the advice, because it cannot be executed by a checklist. If your page is the fourteenth summary of the same publicly available facts, there is no structural trick that makes a synthesis engine prefer it. Original data, first-hand testing, and a genuine point of view are the differentiators.
4. Get the Fundamentals Right
The rest of Google's list is unglamorous and mostly technical:
- Meet the Search technical requirements, and be snippet-eligible
- Be crawlable, including JavaScript-rendered content
- Reduce duplicate content
- Use semantic HTML
- Support text with relevant, high-quality images and video
- Provide good page experience, including low latency
For ecommerce and local businesses, Google also points to Merchant Center and Business Profile data as inputs to its AI experiences.
5. Measure It Properly
Since June 2026 Search Console has included Search Generative AI performance reports, covering your site's visibility in generative AI features on Search. That is now your primary source of truth for the Google surfaces, and it is first-party data rather than inference.
Pair it with direct observation. AIORadar's AI Overview Tracker checks whether an AI Overview appears for a given query and market, whether your domain is cited, and which sources appear instead, keeping each result tied to its query, market, device, and check time. The two answer different questions: Search Console tells you what happened across your whole site, and tracking tells you what the SERP looks like right now for the queries you care about.
A Note on ChatGPT and Perplexity
Everything above concerns Google. Other assistants behave differently, and the same playbook does not transfer cleanly.
Independent analyses published through 2026 report that brand mentions across third-party high-trust domains correlate considerably more strongly with citation rates in LLM answers than backlinks do, and that ChatGPT invokes web search on only about a third of queries, answering the rest from training data. If those findings hold, a large share of ChatGPT visibility is not reachable by editing your own pages at all. It is earned through presence on the sites and in the conversations that fed the model.
That is a digital PR and product-review problem more than an on-page one, and it is worth budgeting separately. Note that Google's warning about inauthentic mentions still applies: the goal is to be genuinely present in your category, not to manufacture citations.
What to Actually Track
Pick a query set that matters commercially, then watch three things monthly: whether those queries return an AI Overview at all, whether you are cited when they do, and who is cited when you are not. The third is the most useful, because a recurring competitor tells you it is a content gap, and a recurring review site or publication tells you it is a brand-presence gap. Those two problems have completely different fixes. Our step-by-step content gap analysis for AI search turns that list into a prioritized plan.