AI Search Optimization in 2026: How SEO, AEO and GEO Fit Together

Most advice on AI search treats it as a new trick. In practice it is SEO with three extra demands: answers that can be lifted, a name machines are sure about, and mentions on sites the engines already trust. This guide covers each part, what is known as of October 2026, and a 90-day plan for a business starting from nothing.

What changed: blue links, AI Overviews, AI Mode and chat answers

For twenty years a search returned ten blue links, and the job was to be one of them. Since May 2024 that page has changed in stages. Google added AI Overviews, a written summary above the links. ChatGPT added live web search in October 2024. Google AI Mode, a chat-style tab of its own, reached all US users in May 2025 and has since rolled out to many more countries. In December 2025 Google removed the Q&A section from Business Profiles and replaced it with Ask Maps, where Gemini answers questions about a business.

All of these still send people to websites, but fewer people, and to fewer sites. When the answer is written on the results page, the person who only wanted a fact has no reason to click. The person who needs a dentist, a CRM or a builder still clicks, and they click the sources the answer named. So the prize has moved. Ranking fifth below an AI answer is worth less than it was. Being one of the two or three sources inside the answer is worth more.

Where people search now

Each surface works a little differently, and each one shows up in a different place in your reports:

What did not change is the starting point. Every one of these systems begins with a search index. If your page cannot be crawled, is not indexed, or says what it does in vague terms, no AI-specific work will get it quoted. Google says this directly: there are no additional requirements to appear in AI Overviews or AI Mode beyond being indexed and eligible to show with a snippet. The rest of this guide builds on that.

How answer engines pick a source

Answer engines differ in the details, but most follow the same five steps. Knowing them tells you where your effort counts and where it doesn't.

Retrieval: the engine searches first

When you ask a question, the model does not read the whole web. It runs searches against an index, Google's own, Bing's or one the company built, and takes back a short list of pages. A page that ranks nowhere for the question, or for any smaller question inside it, rarely makes that list. This is why classic SEO still carries most of the weight. Crawlable pages, clear titles, internal links and a few good backlinks decide whether you are even in the room.

Query fan-out: one question becomes many

AI Mode and similar systems split a long question into several smaller searches and run them at the same time. Google says its Deep Search feature can run hundreds of searches for a single question. A buyer's question with three or four conditions in it becomes a handful of separate searches, and a different page can win each one. More detail in What is query fan-out?

Fan-out is the strongest argument for covering a topic in depth. One general service page can answer one of those searches. A set of pages that each answer one of them well, linked together, gives the engine six chances to land on your site.

Grounding: the passage has to hold up

Once it has the pages, the model looks for passages that answer each part of the question and that it can back up. It prefers text that states the answer plainly, uses full names, and agrees with other sources. A passage that contradicts everything else on the topic is a risk for the engine, so it usually gets dropped. Then the model writes its reply and attaches links to the passages it used.

Two things follow from this. The unit that wins is a passage, so each section of a page has to make sense on its own. And agreement across sources matters: if your site, your Business Profile and three directories all say the same thing about you, the engine can repeat it with confidence.

Engines also disagree with each other more than most people expect. In a March 2026 study of 680 million AI citations, Averi found that ChatGPT and Perplexity shared only 11% of the domains they cite. Showing up in one engine tells you little about another, so check each one that matters to your customers.

SEO, AEO and GEO: one job in three parts

Agencies sell these as three separate services. They make more sense as three layers of one job, each with its own question.

AEO, answer engine optimization, is about the passage: making a section of your page easy to lift as an answer. GEO, generative engine optimization, is about the brand: getting your name included when an AI answer recommends businesses, tools or people. Both sit on top of SEO, because neither works for a page the engine never retrieves.

In practice the layers overlap. A clear, well-structured page helps all three. A well-known brand helps rankings as well as mentions. The split is most useful for diagnosis:

• You rank but are never quoted: work on AEO. Your answers are probably buried under introductions, or spread across several paragraphs.
• You are quoted for definitions but never named when someone asks for a recommendation: work on GEO, which is mostly off-site.
• You are neither ranked nor quoted: start with SEO. Check indexing, titles and whether a page for that question exists at all.

You will also see AIO, which some people use for AI optimization and others for AI Overviews. When someone says AIO, ask which one they mean before agreeing to anything.

Entity first: making machines sure who you are

An entity is a thing a machine can identify without doubt: a person, a company, a place, a product. Search engines keep a graph of them, and AI answers lean on it when they decide whom to name. When an engine is unsure who you are, it hedges, mixes you up with someone who has a similar name, or leaves you out.

This site had the problem itself. Until October 2026 the site title said Jaspreet Singh, the domain said Nagi, and the homepage schema described a company where it should have described a person. Each signal on its own looks minor. Together they told machines there might be two different people. Fixing that came before writing a single article.

Five steps to a clear entity

1. Pick one exact name and use it everywhere: site title, About page, schema, Google Business Profile, LinkedIn, YouTube and directories.
2. Make your About page the entity home. Say who you are, what you do, where and since when in the first paragraph, and link to every official profile.
3. Add Person or Organization schema to that page, or LocalBusiness if customers visit you, with sameAs links to the same profiles. The free Schema Markup Generator writes the code for you.
4. Make each profile link back to the website. A link in both directions is what confirms the match.
5. Clean up old mentions: former business names, old phone numbers and old addresses in directories. Engines read disagreement as doubt.

For local businesses

Your Google Business Profile is the strongest entity signal you control. Use the same name, address and phone number on the profile, in the website footer and in the schema, character for character. With Q&A gone from Maps, Ask Maps builds its answers from your profile, reviews, photos and website, so fill in services, attributes and hours completely, and put your customers' common questions and their answers on your site.

Writing pages that get quoted

Once a page is retrieved, the engine still has to find a passage it can use. Most pages make this harder than it needs to be. Here is the same question answered by two pages:

The pattern that works is simple to describe. Put the question in the heading. Answer it in the first 40 to 60 words below. Then give the detail: steps, a table, an example. Use full names for things, include one specific fact, and make each section readable on its own.

What the research says

The 2023 paper that named GEO, by researchers from Princeton, Georgia Tech, the Allen Institute for AI and IIT Delhi, tested nine ways of rewriting a source page. Citing sources, adding quotations and adding statistics raised a page's visibility in generated answers the most, by up to 40% in their benchmark. Keyword stuffing did badly.

Formats engines lift cleanly

• Definitions, for questions that start with what is.
• Numbered steps, for questions that start with how to.
• Comparison tables, wherever a buyer weighs two or more options.
• Short lists, when an answer has several parts of equal weight.
• A few real questions and answers at the end of the page.

Depth beats volume

Because of fan-out, a thin page can answer one small search at best. A topic covered by a pillar page and a set of linked cluster pages gives the engine many places to land. Each cluster page answers one question in full and links back to the pillar. This guide is the pillar for AI search on this site, and the glossary entries linked throughout it are the first of its clusters.

Show that a person checked it

Show a named author with a real About page, a visible updated date, and links to your sources. Google's guidance from 1 October 2026 asks publishers to fact-check AI-assisted content by hand before it goes live, titles, meta descriptions, schema and alt text included. Run drafts through the AI Writing Tells Checker as well. Text that reads like unedited AI loses readers long before it loses rankings.

Mistakes that keep pages out of answers

Off-site signals AI engines trust: YouTube, Reddit, reviews, press

AI answers about the best anything are built mostly from what other sites say. Your own site tells the engine what you do. Other sites tell it whether you are any good, and that second part decides recommendations.

None of this has a shortcut that lasts. Paid places on best-of lists and bought reviews break Google's policies and tend to get found. The slow version, earning mentions, is also the one that keeps working when an engine changes its model, as ChatGPT did when GPT-6 arrived on 7 and 8 October 2026.

A useful habit: every time you answer a customer's question well in person or on WhatsApp, ask whether that answer belongs in three places. A section on your website, a short video, and a reply in a forum where the same question comes up. One good answer, published three times, covers the site, YouTube and the discussions engines read.

Technical access: crawlers, llms.txt, speed, rendering

None of the writing matters if the crawlers cannot read the page. Check these five things.

1. Let the right bots in

Each AI company runs its own crawlers, and most run more than one. The search bots decide whether you can be cited. The training bots decide whether your content is used to build future models. They are separate choices.

Check your robots.txt at yoursite.com/robots.txt, and check your firewall too. Security plugins and CDN settings can block AI bots without telling you, and Cloudflare began blocking AI crawlers by default for new sites in July 2025. OpenAI says that blocking GPTBot does not remove a site from ChatGPT search, as long as OAI-SearchBot is allowed.

2. Put the content in the HTML

Google renders JavaScript before it reads a page. Many AI crawlers don't. A 2024 analysis by Vercel found that the major AI crawlers it studied fetched JavaScript files but did not run them. If your text, prices or reviews only appear after a script loads, those crawlers see an empty page. View the page source (not the inspector) and check that the main content is there.

3. Keep pages fast and error-free

Slow pages and server errors get crawled less often. Open the Page indexing and Crawl stats reports in Search Console, fix any 5xx errors, and make sure the pages you most want cited return a clean 200 response in under a second or two.

4. Know the snippet controls

Google's AI features follow the same controls as normal snippets: nosnippet, data-nosnippet, max-snippet and noindex. Use data-nosnippet on a passage you want kept out of AI answers, knowing it also hides that passage from regular snippets. Most businesses never need these.

5. Add llms.txt if you like, without expecting much

llms.txt is a proposed file that lists a site's most useful pages for AI tools. Google representatives have said Search does not use it, and no major AI search engine has confirmed using it to choose citations. It takes about 20 minutes and does no harm. If you add one, check your server logs after a month to see which bots actually request it.

Set up Bing Webmaster Tools as well. Copilot runs on Bing's index, and Bing supports IndexNow, which lets you tell it about a new or updated page the same day.

A note on tools: I co-founded wptaskify, so weigh this with that in mind. If your WordPress site is connected to it, you can ask Claude or ChatGPT to show you robots.txt and llms.txt and make changes to both. The manual route works just as well: most SEO plugins have a robots.txt editor, and llms.txt is a plain text file you upload to the site's root folder.

Measuring AI visibility on a small budget

Search Console now shows AI impressions but not AI clicks, and no tool sees every answer. You can still build a clear picture for free with three sources.

Search Console

Since 31 August 2026 every site has a Generative AI performance report in Search Console. It shows how often your pages appeared in AI Overviews and AI Mode, by page, country, device and date, but it has no clicks and no queries. Clicks from AI features still sit inside the normal Web search report. Use the two together: look for queries where impressions rose while clicks fell, then search them yourself to see whether an AI answer now sits above your result.

GA4 referrals

Since 13 May 2026 GA4's default channel group has a built-in AI Assistant channel for visits from assistants such as ChatGPT, Gemini and Claude. It doesn't include AI Overviews or AI Mode, and some assistants can still land in Referral, so also open Reports, then Acquisition, then Traffic acquisition, and filter the session source for chatgpt, perplexity, copilot, gemini and claude. ChatGPT adds utm_source=chatgpt.com to many of its links, so its visits are easy to spot. These numbers are small for most sites, but they are visitors who arrived with a specific question, and they tend to convert well.

A 40-prompt panel

This is the part paid tools charge for, done by hand once a month. Write 40 questions a real customer would ask, in their words, in four groups of ten. Run each one in ChatGPT, Perplexity, Google AI Mode and Copilot, in a fresh or logged-out session, and record whether your business is named, linked or absent. Save a screenshot of each answer.

Answers vary from one run to the next, so a single result means little. The trend over three months is what counts: how many of the 40 name you, in which engine, and which of your pages get linked. Keep it in a simple sheet with one row per prompt and one column per engine per month.

Paid trackers such as Semrush's AI toolkit, Ahrefs Brand Radar and Profound run thousands of prompts for you. They earn their cost once you have a budget and a baseline. Start with the manual panel so you know what you would be paying to automate.

How the work changes by type of business

The layers are the same for everyone, but the weight shifts. A plumber, a software company and an online shop win AI answers in different places.

Local service businesses have the shortest path. Their buyers ask near-me and best-in-city questions, and the answers lean on Business Profiles and reviews that the business already controls. A clinic with a complete profile, steady reviews and a price page for each treatment is ahead of most competitors within a quarter.

B2B and software companies face the hardest comparison questions. Buyers ask which tool suits a team of five, or how one product compares with another, and the answers draw on review sites, comparison articles and forum threads. Honest comparison pages on your own site help, as long as they name the cases where a competitor is the better fit. Engines quote fair comparisons and skip one-sided ones.

Online shops depend more on product data. Clear product titles, full specifications, prices, stock status and return policies in the page HTML give shopping answers what they need. Product and Offer schema helps machines read those details, and reviews on the product page do the persuading.

Personal brands and consultants sit closest to the entity problem described above. The work is mostly about one clear name, an About page that states expertise plainly, and a body of work published under that name: articles, talks, videos and interviews that engines can connect back to one person.

A 90-day plan for a business starting from zero

This is the order I use with service businesses. It assumes a working website, a Google Business Profile and a few hours a week.

Weeks 1 to 2: access and entity

1. Check robots.txt and your firewall against the bot table above.
2. Submit your sitemap in Search Console and Bing Webmaster Tools.
3. Choose one exact business name and fix it on the site, Business Profile and main profiles.
4. Rewrite the About page as your entity home and add schema with sameAs links.
5. Complete every field of your Google Business Profile: services, attributes, hours, photos.

Weeks 3 to 6: answer pages

1. List 30 real questions from calls, WhatsApp chats, emails and Search Console queries.
2. Rewrite your top five service pages answer-first, with prices or price ranges.
3. Add a comparison table wherever buyers weigh options.
4. Write one pillar page for your main service and four to six cluster pages, each answering one question in full.
5. Run every page through a fact check and the AI Writing Tells Checker before it goes live.

Weeks 7 to 10: off-site

1. Ask every happy customer from the last six months for a Google review, and reply to all reviews.
2. Record three short videos that answer your most common questions.
3. Find two lists that rank for your service and ask to be reviewed.
4. Answer ten questions in forums or groups where your customers ask them, under your own name.
5. Pitch one useful story or number to local or trade media.

Weeks 11 to 13: measure and repeat

1. Run the 40-prompt panel and compare it with the first run.
2. Compare Search Console clicks and impressions with the 30 days before you started.
3. Update the pages that slipped, and the ones that are cited for the wrong reason.
4. Pick the next five questions and plan the next quarter.

Most businesses see the first citations during weeks 3 to 6, because rewriting pages that already rank gives the engines a better passage to quote. Recommendations by name usually take longer, since they depend on the off-site work.

Questions people ask

Is SEO dead because of AI search?

No. Every answer engine pulls pages from a search index first, and the pages that rank for the smaller searches are the ones that get cited. What has changed is that a ranking no longer guarantees a click.

How long does it take to show up in AI answers?

Pages that already rank can be cited within days of an update, because the engines search live. Being named in recommendations takes months, since it depends on reviews and mentions on other sites.

Do I need a separate plan for ChatGPT and Google?

The work is mostly the same, but check each engine you care about. They cite different sites: in Averi's March 2026 study, ChatGPT and Perplexity shared only 11% of the domains they cite.

Can I pay to appear in AI answers?

Ads are appearing in and around some AI answers, and they are labelled as ads. The citations and brand mentions this guide covers cannot be bought.

Does schema help AI search?

It helps machines read a page and confirm who is behind it. It does not guarantee a citation. FAQ rich results stopped showing on 7 May 2026, but the markup is still valid and still read.

Should a small business block AI crawlers?

Most should allow the search bots, because blocking them removes the site from those answers. Blocking training bots such as GPTBot is a separate decision about whether your content is used to train models.

Keep reading

Every guide in this section

This pillar organises the whole AI Search library on the site, and the index below grows as each cluster guide is published.

Sources