How to Appear in AI Overviews: What Google Actually Uses in 2026

Most advice about how to appear in AI Overviews is written as if Google had built a second search engine with its own secret rules. It did not. Google’s own documentation, updated through 2026, says the AI summaries at the top of results are built on the same ranking and quality systems as the blue links underneath them. The work that gets a page cited is, almost entirely, the work that gets a page found.

That is good news and bad news. The good news is that you do not need to buy a new tool, install a plugin, or rewrite your site in a special dialect for machines. The bad news is that there is no shortcut either. A page that is not indexed, cannot show a snippet, or does not answer the question better than the alternatives will not be lifted into an AI Overview, no matter how much structured data it carries.

This guide is for owners and marketing leads at US businesses who want a clear plan, not a list of hacks. It explains how Google picks the pages it links from an AI Overview, walks through five gates every page has to pass, shows what to stop worrying about, and ends with how to measure the result with data Google actually provides. Every claim about how Google works is tied to Google’s own pages, with links, so you can check it yourself.

What is an AI Overview, and how does Google choose its sources?

An AI Overview is the generated summary Google shows above some search results, with links to the pages it drew from. AI Mode is the conversational version, where a person can ask follow-up questions in a chat-like view. Both are features of Google Search, not separate products, and both rely on the Search index.

Google describes the mechanics in two places. Its page on AI features and your website explains eligibility and controls. Its newer guide, Optimizing for generative AI search, published in May 2026 and last updated in July 2026, explains how the systems work and which popular tactics it does not need.

Read together, they describe a three-step process. First, the model receives a question. Second, it runs what Google calls query fan-out: a set of concurrent, related searches across subtopics, generated by the model itself, to gather more material than one search would return. Third, it retrieves pages from the index using Google’s core ranking systems and writes a response grounded in what it found, linking the pages that support it. In machine learning terms, this is retrieval-augmented generation: the answer is generated, but the facts are retrieved from indexed pages first.

The link pool is wider than page one

Because fan-out runs several searches at once, an AI Overview can link pages that rank for a sub-question even when they do not rank for the main query. Google’s AI features page says this lets it show a wider and more diverse set of supporting links than a classic results page. A small business that will never outrank a national publisher for a broad query can still be the best answer to one narrow sub-question inside it.

Not every query gets an AI Overview

Google says AI Overviews appear only when its systems judge them to be additive to classic results. Many searches never trigger one: navigational queries, simple lookups, and many commercial searches are often answered better by a plain list of links. If the query you care about does not show an AI Overview, there is nothing to appear in, and the right goal for that query is the ordinary organic result.

If you want the vocabulary behind all this, our explainer on the difference between AEO, SEO and GEO covers the terms. The short version is that Google itself now treats them as one discipline.

How to appear in AI Overviews, according to Google’s own guidance

Google is unusually direct on this point. Its AI features page says there are no additional requirements to appear in AI Overviews or AI Mode, and no special optimizations necessary. The 2026 guide goes further and says that optimizing for generative AI search is optimizing for the search experience, and that best practices for SEO continue to be relevant.

Here is what Google lists as worth doing, condensed from both pages:

• Meet the Search technical requirements. The page has to be indexed and eligible to be shown with a snippet. Google states this as the basic condition for appearing as a supporting link.
• Allow crawling. Googlebot must be able to fetch the page, which includes your robots.txt file and any CDN or hosting layer that might block it.
• Make content findable through internal links. Pages nobody links to are slow to be discovered and weak in ranking.
• Provide a good page experience. Usable on every device, fast to load, and easy to tell the main content from everything around it.
• Keep important content in text. Text that only exists inside an image, a slider or a script is harder to retrieve and quote.
• Offer a unique point of view. The 2026 guide asks for content that adds something, not content that restates what every other page already says.
• Support text with high-quality images and video where they help, and keep structured data consistent with what is visible on the page.
• Keep Business Profile and Merchant Center data current if you are a local business or sell products.

Notice what is missing from that list. There is no AI-specific file, no AI-specific schema, no word count, no required format for answers, and no instruction to write for robots. Everything on it would have been good advice for SEO five years ago. That is the single most useful thing to understand before spending money on AI search: the fundamentals are the strategy.

Google’s AI Overviews are built on the same ranking and quality systems as its organic results, so the pages it cites are, first, pages it can find, index and rank.

That does not mean every SEO tactic transfers. Some popular habits, like stuffing an exact phrase into every heading, add nothing to an AI summary because the model reads meaning, not phrase matches. Google’s guide says its systems understand synonyms and general meaning, so you do not need to write in a specific way for them. Clear writing beats keyword-shaped writing in both places.

The five gates a page passes before it can be cited

When a business asks why it never appears in AI Overviews, the answer is almost always one of five gates, and they fail in order. A page that fails gate one cannot be helped by work on gate four. We use this ladder as a diagnostic because it stops teams from polishing answers on pages Google cannot even show.

1. Gate 1: crawlable. Googlebot can fetch the page. robots.txt allows it, the server returns 200, and nothing at the CDN or firewall level blocks Google’s crawler. Check it with the URL Inspection tool in Search Console.
2. Gate 2: indexed. The page is in Google’s index under the URL you expect, with the canonical you intended. Pages marked noindex, pages canonicalized to another URL, and pages Search Console lists as discovered but not indexed are out of the pool.
3. Gate 3: snippet-eligible. The page can be shown with a text snippet. A nosnippet rule, a very low max-snippet value, or a site-wide opt-out from generative AI features removes the page from AI Overviews even while it still ranks.
4. Gate 4: competitive for a sub-question. The page ranks well enough, through Google’s core systems, for the main query or one of the sub-questions the model generates during fan-out. This is where content quality, internal links, page experience and reputation all show up.
5. Gate 5: answerable. The page states the answer in plain text, close to the heading that asks the question, in a passage that makes sense on its own. A page that buries the answer in a long story is retrievable but hard to use.

A practical note on gate two. A redesign or migration that changes URLs without proper redirects can knock a whole section of a site out of the index for weeks. If AI Overview visibility dropped after a relaunch, start with redirects. Our guide to migrating a website without losing SEO rankings lists the checks.

Query fan-out: why sub-questions decide which pages get cited

Fan-out is the part of the process that changes strategy the most. When someone asks a broad question, the model does not run one search. It breaks the question into narrower ones, searches them at the same time, and combines what it finds. Each of those narrower searches is a separate chance to be retrieved.

Take a business owner who searches for whether they should redesign their website or start over. A fan-out for that question is likely to touch several angles: what a redesign includes, what changes in a rebuild, what drives the effort, how long each tends to take, what happens to rankings, and how to decide. Nobody outside Google can see the exact sub-queries, but you can predict most of them from evidence you already have: Google autocomplete, the People Also Ask box, related searches, and the questions customers ask you on calls.

The planning move is simple. List the sub-questions, then decide which page on your site answers each one, and make sure that answer exists in plain text. We call this a fan-out map. It looks like this:

Two things happen when you build this map honestly. First, you find gaps: sub-questions customers clearly ask that no page on your site answers. Second, you find overlaps: three pages that half-answer the same question and compete with each other. Both are fixable, and fixing them helps classic rankings at the same time, because you end up with one clear owner for each question.

This is also where a small site can beat a large one. A national publisher may own the broad query. It rarely owns every narrow, practical sub-question a buyer in your field asks. Be the best answer to those, consistently, and you give the fan-out something to retrieve.

How do you write a section an AI Overview can use?

Once a page is eligible and competitive, the remaining question is whether the answer is easy to find and easy to use. This is not about writing for robots. It is about the same habits that make a page useful to a person who skims, which is most people.

Lead with the answer

Put the direct answer in the first one or two sentences under the heading that asks the question. Then add the context, conditions and exceptions. A reader who stops after two sentences should leave with the right answer, and a model looking for a grounded statement finds one immediately.

Make passages stand on their own

Write paragraphs of two to four sentences that make sense out of context. Name the subject instead of saying “it” or “this” across paragraph breaks. A passage that begins “This is why it matters” means nothing when lifted; a passage that begins “A noindex tag removes a page from AI Overviews because” is complete.

Use questions as headings where they are natural

Headings phrased the way people ask help both readers and retrieval. Do not force it. “Pricing” is a fine heading on a pricing page. On a guide, “What drives the cost of a custom website?” tells everyone what the section answers.

Use tables and lists for comparisons and steps

Comparisons belong in tables and procedures belong in numbered steps. They are easier to scan and they keep related facts together, which reduces the chance of a summary mixing up which fact belongs to which option.

State facts once, the same way everywhere

Your business name, what you do, where you are and how long you have operated should read the same on every page and in your structured data. Inconsistent facts force any system, human or machine, to guess which version is right. Our guide on getting your business cited by ChatGPT covers entity consistency in more depth, since the same principle applies across AI assistants.

Add something the consensus does not

Google’s 2026 guide asks for a unique point of view. In practice, that means first-hand experience, a worked example, a process detail other pages skip, or a reasoned disagreement with common advice. A page that only restates the top results gives the model nothing it could not get elsewhere.

None of this requires a special format. Google’s guide states that there is no requirement to break content into tiny pieces. Write well-organized pages for people; the structure that helps them is the structure that helps retrieval.

Which technical settings quietly block AI Overviews?

Several settings remove a page from AI Overviews while leaving it in regular results, which is why they go unnoticed. Check these before anything else.

The details of the first three come from Google’s documentation on the robots meta tag and X-Robots-Tag, which states that nosnippet and max-snippet also govern use in AI Overviews and AI Mode. That page notes an important side effect: nosnippet removes the text description from your normal search listing too, so it is a blunt tool.

Two other blockers are common on small business sites. The first is content that only appears after a script runs. Google can render JavaScript, but heavy client-side rendering delays and sometimes loses content, and Google’s guide points site owners to its JavaScript SEO practices. The second is a security layer that challenges or rate-limits Googlebot. Some hosting firewalls treat aggressive crawling as an attack. If Search Console shows crawl errors you cannot explain, ask your host whether Google’s crawlers are being challenged.

Page experience belongs on this list too, even though it is not a switch. A slow, unstable page is less competitive at gate four. Our explainer on Core Web Vitals and how they affect SEO covers the three measurements Google uses and how to read them.

Google-Extended is a different control

Google-Extended is a robots.txt token that controls whether content is used to train and ground Google’s AI models in other Google systems. Google’s AI features page lists it separately from the Search controls. Blocking Google-Extended is not the tool for removing a site from AI Overviews; the Search controls in the table above are.

What should you ignore? llms.txt, chunking and AI rewrites

A large share of the advice ranking for this topic sells tactics Google has now said it does not use. That does not make every tactic worthless for every AI system, but for Google’s AI Overviews the 2026 guide is explicit. Here is how the popular claims compare with what Google publishes.

Structured data still matters, just not as a ticket into AI Overviews. It makes pages eligible for rich results, it states facts in a consistent machine-readable form, and Google asks that it match the visible text. Our beginner’s guide to schema markup covers which types are worth adding and how to validate them.

An llms.txt file is a similar case. Some AI assistants outside Google may read it, and it costs little to maintain, but it will not move anything in AI Overviews. If someone sells you an llms.txt setup as an AI Overview strategy, the claim does not match Google’s documentation.

AI Overviews and AI Mode: does the same work cover both?

Mostly, yes. Both features rely on the Search index, both use fan-out, and both are governed by the same eligibility rules and controls. The differences are in how people use them and in the models behind them. Google’s AI features page notes that AI Overviews and AI Mode may use different models and techniques, so the set of links each shows for the same question can differ.

The practical difference is the follow-up. In AI Mode, a person refines the question step by step: first the broad need, then constraints, then a comparison, then a next step. Each turn is a new chance to be retrieved, and the later turns tend to be narrower and more commercial. Pages that state who a service is for, what it involves, what the cost depends on and what the next step is give those later turns something concrete to use.

That is the same structure a good service page should already have. If your service pages answer those four points in plain text, you are serving AI Mode follow-ups and human buyers with one piece of work. If they read like a list of adjectives, it may be the page and not the traffic that limits your leads, which our guide on why a website is not getting leads unpacks.

If you would rather have someone review your service pages against those four points, book a Blueprint Briefing and bring the URLs; we will tell you which questions each page answers today and which it skips.

Why are AI Overviews not showing up for your site?

There are two different problems hiding behind this question, and they have different fixes. Either the query does not trigger an AI Overview at all, or it does and your site is not among the links.

When the query shows no AI Overview

• The query is not one Google treats as needing a summary. Navigational, simple factual and many shopping queries often show none. Google says the feature appears only when it adds something.
• The result varies. The same query can show an AI Overview on one search and not on the next, and results differ by location, device and signed-in state.
• The viewer has hidden it. Browser extensions and some settings remove AI Overviews from the page, so a colleague’s screenshot may differ from yours.

When an AI Overview shows, but your site is not linked

One pattern is worth calling out. A page that ranks first organically is not automatically the page an AI Overview links. Fan-out widens the pool, and the model picks passages that support specific statements in its answer. A page that ranks well but hides its answer under a long introduction can lose the citation to a lower-ranked page that states it in two sentences. That is a gate five problem, and it is usually the cheapest one to fix.

How do you measure AI Overview visibility in Search Console?

For years, the honest answer was that you could not, at least not separately. AI Overview traffic was folded into the Web search type of the Performance report with no way to isolate it. In 2026 that changed. Google introduced a Search generative AI performance report in Search Console, first tested with a subset of sites and, according to Google’s help page, rolled out to all websites worldwide as of August 31, 2026.

Here is what the report shows, according to Google’s help documentation:

• Impressions for your site in generative AI features on Google Search, meaning AI Overviews and AI Mode. Search Labs experiments are excluded.
• Dimensions for pages, countries, dates and devices, with a search type filter that separates typed searches from multimodal ones such as Google Lens.
• Counting rules that differ from what most people expect. In the chart, several links to your site in one response count as one impression for the property. Grouped by page, the table counts per page. Totals can differ between the two views for that reason.

Three cautions keep the numbers honest. First, the report shows impressions, not clicks, so it tells you how often you were shown, not what that exposure produced. Second, it may not appear at all if your site has not received enough generative AI impressions yet, which is common for small sites and says nothing about a penalty. Third, the same activity is still included in the overall Performance report, so do not add the two together.

Connect visibility to results

Impressions are a visibility metric. The business question is whether visibility turns into calls, forms and bookings. That answer lives in your analytics, not in Search Console. Track conversions as events in GA4, compare landing pages that gained AI impressions with those that did not, and look at the quality of the leads, not only the count. Google’s AI features page states that clicks from AI Overviews tend to be higher quality, citing longer time on site; your own data is the only way to see whether that holds for your business.

If analytics is not set up yet, start there. Our guide to tracking website analytics to improve SEO walks through the basics.

Local businesses, images and video in AI Overviews

Google’s 2026 guide includes guidance that is easy to overlook if you only think of AI Overviews as text summaries.

Business Profile data feeds local answers

For businesses that serve customers in a place, Google says keeping Business Profile information current can help products and services be visible in AI responses. That means the basics: correct name, address and phone, accurate categories, real hours, a description of the actual services, and photos of real work. The same facts should match your website exactly. Our guide to local SEO for websites covers how the profile and the site support each other.

Images and video support the text

Google recommends adding high-quality images and video where they are relevant, and AI features can show them. The practical rules are the same as for image search: original images rather than generic stock, descriptive file names and alt text, and images placed next to the text they illustrate. Our list of image optimization best practices has the details. Multimodal searches, such as a photo search through Google Lens, are reported separately in the generative AI report’s search type filter, which gives you a way to see whether images are contributing.

Should you ever opt out of AI Overviews?

Since 2026, site owners have a direct choice. Google announced a Search Console control on June 3, 2026 that lets a site decide whether it appears in, and helps ground responses in, AI Overviews, AI Mode and AI Overviews in Discover. Google says sites that opt out will not receive traffic or impressions from those features, and that the control is not used as a ranking signal for results outside them.

For most businesses that want customers to find them, opting out is the wrong default. If your buyers ask Google questions before they buy, an AI Overview is one more place your answer can be seen, and opting out hands that space to whoever is left. The case for opting out is narrower:

• Your content is the product. Publishers whose revenue depends on people reading the full page may decide summaries cost them more than they return.
• Licensing or legal constraints. Some content is licensed under terms that do not allow this kind of reuse.
• Accuracy concerns in a regulated field. A business may prefer that a summary not paraphrase sensitive guidance, and may use nosnippet or max-snippet on specific pages instead of the whole site.

If you do restrict anything, restrict the smallest scope that solves the problem: a section with data-nosnippet, a page with nosnippet, and only then the site-wide setting. Measure before and after with the generative AI report and your analytics, so the decision rests on your numbers rather than on a general opinion about AI.

How GoSocial approaches AI Overview visibility

GoSocial is a custom web design company in Boca Raton, Florida, building websites for businesses across the United States since 2016. We optimize websites for traditional search (SEO), answer engines (AEO) and generative AI search (GEO) on one foundation, which is the same conclusion Google’s own 2026 guide reaches: it is one discipline, not three.

In practice, that means the work in this guide is built into the site rather than added afterward. We build both custom-coded websites and custom WordPress websites, and in both cases the content is in the HTML, the pages are fast and indexable, and every service page states who it is for, what it involves and what the next step is. Our custom web development process runs from discovery and strategy through architecture, design, development, testing, launch and measurement, and the fan-out map described above is part of the strategy stage.

Every GoSocial website launches with GA4 and Google Search Console connected, tracking form submissions, phone clicks, WhatsApp clicks and bookings, so AI visibility can be compared with actual leads from the first day. Our SEO, AEO and GEO services describe the ongoing side, and one published example is Jortman, a jewelry brand whose visibility grew in Google and in AI search environments, appearing on the first pages for relevant searches.

We also say what we cannot do. No one can guarantee a citation, and we do not. We do not promise a performance score either; we build for the metrics and show the measurements. If you are weighing a rebuild against working with the site you have, our comparison of custom websites and WordPress is a useful starting point, since either path can pass all five gates when it is built well.