AI Search Optimization in 2026: Get Cited by ChatGPT and Google

What is AI search optimization, and how is it different from SEO?

AI search optimization is SEO for a new kind of reader: an engine that writes the answer itself and names a few pages as its sources. The groundwork is shared, and Google says plainly that optimizing for its generative AI features is still SEO.

In a classic results page, ten links compete for a click and the searcher does the reading. In an AI answer, the engine does the reading: it searches, opens several pages, writes one reply and cites a handful of them. Being one of those cited pages is the new prize, because the citation puts your business name inside the answer even when nobody clicks.

Most of what helps a page rank also helps it get cited. A page that search engines cannot crawl, index or understand will not be quoted by anyone, and Google asks owners to judge paid AEO or GEO services by the same standards as any other SEO help. If your website has never had basic SEO attention, run the AI Website SEO Tools check of titles, structure and content alongside the AI work.

Three things are new, and they are what the rest of this page covers. First, each AI engine has its own bots, and a website can welcome Google while quietly refusing ChatGPT or Claude. Second, engines lift passages rather than whole pages, so each important page needs an answer that makes sense on its own. Third, engines lean heavily on what other websites say about you, so your reputation elsewhere matters as much as your own copy.

How do AI engines choose the pages they cite?

AI engines cite pages in two steps: a search finds candidate pages for the question, then the model picks the passages that answer it most clearly and credibly.

It helps to separate what a model learned in training from what it reads live. Training data is the text a model learned from months before you ask it anything, and it shapes what the model already believes about well-known subjects. When ChatGPT, Perplexity, Claude, Copilot or Google's AI features look something up, their search bots fetch current pages and the answer cites those pages with links. The citations you can win, and the visits you can measure, come from that live search, which is why the search bots matter far more to a business than the training bots.

Google describes one more detail that changes how you plan pages. Its AI Overviews and AI Mode may use query fan-out, issuing several related searches across subtopics and data sources to build one response. A question such as "which accountant in Leeds handles sole traders" can fan out into searches about fees, services and reviews, and a different page can be cited for each part, so one clear page per real customer question gives you more chances than one long page that answers everything vaguely.

None of the engines publishes a formula for citations, and Google states that no third-party tool has access to its internal ranking or AI systems. What the engines' own advice repeats is a short list of qualities. The page must be reachable by the engine's search bot, it must contain a direct answer, it should back that answer with facts or sources, and the business behind it should be identifiable and consistent across the web.

Which AI bots should your robots.txt let in?

Let every AI search bot in, and decide separately about training bots. Blocking a training control such as GPTBot, ClaudeBot or the Google-Extended token does not remove you from that engine's search answers.

Most AI companies now run separate bots for separate jobs, and each reads its own line in robots.txt. OpenAI lists OAI-SearchBot for ChatGPT search, GPTBot for content that may train its models, ChatGPT-User for pages a person asks ChatGPT to visit, and OAI-AdsBot for pages submitted as ChatGPT ads. It states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though they can still appear as navigational links, while disallowing GPTBot only signals that content should not be used for training. Anthropic's support page describes the same split for Claude: ClaudeBot for training, Claude-SearchBot for search and Claude-User for user requests.

Perplexity's crawler documentation lists PerplexityBot, which surfaces and links websites in its results and is not used to crawl content for AI foundation models, and Perplexity-User for user requests. Google's AI Overviews and AI Mode are part of Google Search and use Googlebot. Google-Extended is a separate robots.txt token that controls whether content may train future Gemini models and ground answers in Gemini Apps and Vertex AI, and Google states that Google-Extended does not affect a site's inclusion in Google Search or its ranking there.

The user-request fetchers behave differently. OpenAI says robots.txt rules may not apply to ChatGPT-User because a person started the request, and Perplexity says Perplexity-User generally ignores robots.txt for the same reason. Anthropic, on the other hand, says disabling Claude-User stops Claude retrieving your content for a user's question, so leave these fetchers allowed unless you have a clear reason.

Whether to block training bots is a business decision rather than a search one. Allowing them lets future models learn your name, services and expertise, while blocking them keeps your writing out of training sets, which matters most to publishers whose content is the product. For most service businesses and shops allowing everything is simpler, and either choice is fine as long as the search bots stay allowed. A training-only opt-out reads User-agent: GPTBot with Disallow: / on the next line, and no rule at all for OAI-SearchBot.

Is a firewall or CDN blocking AI bots without you knowing?

A firewall, security plugin or content delivery network can refuse AI bots even when robots.txt welcomes them. Check how your homepage answers each bot, not only what robots.txt says.

Robots.txt is a request that bots read before they crawl, but the decision to serve a page is made by your server and anything in front of it. Content delivery networks such as Cloudflare offer settings that block or challenge AI crawlers, hosts apply bot protection to whole servers, and WordPress security plugins often rate-limit visitors they do not recognise. Any of these can answer an AI search bot with an error or a challenge page, and the engine then has nothing to read.

The symptom is silent, which is why it lasts. The website looks normal to every human visitor and Google may still index it, yet ChatGPT or Claude answers questions about your business from other websites. A useful check requests your homepage as each bot, with its real user agent, and confirms that a normal page comes back.

Fix a block in the tool that made it. In a CDN or firewall, allow the search bots by name and, where the provider supports it, by verified identity; OpenAI and Perplexity publish their bots' IP ranges as JSON files for exactly this, and OpenAI's help pages say a host or CDN must accept its search bot's addresses for pages to be eligible. Anthropic does not publish IP ranges for its bots and says blocking them by IP address may not work correctly, so robots.txt is the way to set its rules. After any change, request the homepage as each bot again and allow about a day for the engines to notice.

How do you get your website cited by ChatGPT?

ChatGPT cites pages that OAI-SearchBot can reach and that answer the user's question clearly. Allow OAI-SearchBot, make sure nothing blocks it, then give each important page one question to answer well.

When ChatGPT searches, it reads candidate pages and cites the pages it used inline, while its Sources panel also lists other relevant links. OpenAI's documentation is clear that OAI-SearchBot decides whether your pages can appear there, and that ChatGPT-User is not used to decide what appears in search. Without a live search, ChatGPT answers from training data, which may be out of date and comes with no link to you.

What earns the citation is answer-first writing with concrete details. Pages that state the answer in their first lines, name the business and the place, and give specifics such as your own prices, hours, service areas or steps are easier to quote than general marketing copy. For comparison and "best" questions, your mentions on lists and directories count as much as your own page, and dated, current information helps an engine trust that the page is still true.

To see where you stand, ask ChatGPT the questions your customers ask, with search switched on, and note which websites it cites. Use 10 to 20 questions, from "what does a conveyancer charge" to "best wedding florist near Bath", and repeat the same set each month. Answers vary, so read the pattern over months, not single replies.

How do you get pages into Google AI Overviews and AI Mode?

Google AI Overviews and AI Mode draw on pages that are indexed and eligible to show a snippet. Google says no extra files, markup or special optimization are needed beyond normal SEO.

Google's guide is unusually direct. There is no special schema.org markup to add, no AI text file to create, no need to break content into tiny pieces and no ideal page length, and rewriting pages to chase every long-tail variant is unnecessary because its systems understand meaning. What it asks for is unique, useful content written from your own knowledge, on a page that is crawlable, quick and easy to use on any device.

An AI Overview appears on an ordinary results page to give the gist of a question, while AI Mode is a conversational mode for questions that need more exploration or comparison. Both can fan a question out into several searches and both cite pages as supporting links. Search Console also has a Search generative AI control under Settings that includes a website in these features by default, so check it once, because a property set to exclude receives no impressions from them.

Google's snippet controls are where owners limit themselves by accident. Its robots meta tag documentation says nosnippet prevents content being used as a direct input for AI Overviews and AI Mode, that max-snippet limits how much may be used. The data-nosnippet attribute keeps a single span, div or section out of snippets in Search, though Google's AI wording names only nosnippet and max-snippet. A site-wide nosnippet left over from an old theme or plugin quietly stops every page being a direct input for AI answers; to leave the features altogether Google points to noindex for a page or the Search Console control for the whole site.

Google adds practical points for businesses. Real images and video can be brought into AI responses, a current Google Business Profile helps products and services show, and Merchant Center feeds do the same for shop products. It also warns that mass-producing pages for query variations, mainly to manipulate rankings or AI responses, breaks its scaled content abuse policy.

How do Perplexity and Claude choose their sources?

Perplexity and Claude both search the web live and link the pages they use, so letting PerplexityBot and Claude-SearchBot reach your website comes first.

Perplexity's answers usually list numbered sources beside the text. Its documentation recommends that site owners allow PerplexityBot and says robots.txt changes can take up to 24 hours to show. It has outlined its approach (its own index, pages split into small parts that are each scored against the question, and a priority on fresh content) without publishing ranking weights, and owners observe that the pages it quotes tend to answer directly and carry specific facts.

Claude, the assistant made by Anthropic, searches the web when a question needs current information and cites the pages it used. Anthropic says blocking Claude-SearchBot may reduce your visibility and accuracy in its search results, while ClaudeBot, its training crawler, is a separate choice that also honours the Crawl-delay rule. Blocking ClaudeBot keeps future material out of training, but only Claude-SearchBot and Claude-User affect whether Claude can cite you today.

Because neither company publishes a formula, the approach is the same for both. Allow the search bots, keep pages fast and readable without heavy scripts, open each page with a direct answer, and back claims with numbers, dates and named sources. When a competitor is cited and you are not, compare the two pages line by line: the difference is usually a clearer first answer, a more specific fact or a mention on a website the engine trusts.

How do you get your website cited by Microsoft Copilot?

Microsoft says its Copilot experiences are powered by Bing, so being cited by Copilot starts with being crawled, indexed and trusted by Bing.

Many websites have never been added to Bing Webmaster Tools, and that is the first fix. Verify the website, submit its sitemap and check that Bingbot can crawl the pages you care about; Bing says it respects robots.txt and its other supported controls, so the same firewall checks apply. Once verified, the AI Performance report shows Total Citations, Average Cited Pages, the grounding queries used to retrieve your content and citations for each page.

Bing's announcement of that report gave concrete advice. Pages cited for specific grounding phrases tend to have a clear subject focus, headings, tables and FAQ sections help AI systems reference key information, and examples, data and cited sources build trust. It recommends regular updates, consistency between text, images and video, and Bing Places for Business so local details stay current.

Bing also recommends IndexNow, a protocol that tells search engines the moment a page is added, updated or removed. Under the IndexNow rules, a participating engine shares the URLs it receives with the other participating engines, so one notification reaches all of them. For a business that changes prices, hours or stock, it helps engines find the change sooner, though IndexNow says a submission does not guarantee crawling or indexing, and Google is not among the participating engines.

How do you write a page that AI engines can quote?

Write each important page answer-first: the question in a heading, a direct answer of 40 to 60 words beneath it, then the evidence and detail.

Engines quote passages, and the one most likely to be lifted answers the question completely on its own. That means the answer names its subject instead of saying "it", states the fact plainly in the first sentence and stops before the sales pitch. A heading written as the customer's question, such as "How long does a boiler service take?", tells readers and engines exactly what the next paragraph answers.

Google says there is no need to split content into tiny pieces and no ideal page length, and answer-first writing agrees with that. It is not about fragmenting a page; it is about putting the conclusion before the explanation, the way a good professional answers a client. After the answer come the conditions, steps, exceptions, examples from your own work and sources for any figures, which is what makes the page trustworthy enough to cite.

The examples only show the pattern, and every figure on your own page should be your real one. Link to the regulator, standard or study behind a claim, date information that changes, and say who wrote or checked the page when expertise matters. A named author or reviewer helps readers and engines alike judge whether the source deserves trust.

Frequently asked questions still earn their place. Google stopped showing FAQ rich results from 7 May 2026, so FAQ markup no longer changes how a page looks in results, but a good FAQ answers the follow-up questions people ask and gives engines more self-contained passages. Bing names FAQ sections among the structures that help AI systems reference information, so write each question the way a customer would say it and answer it in two to four sentences.

Does structured data help AI engines understand your business?

Structured data helps engines confirm who your business is and what it offers, but Google says it is not required for AI features and no special markup exists for them.

Structured data is code in the page, usually JSON-LD using the schema.org vocabulary, that labels facts such as your name, address, hours, logo and services. Google recommends keeping it as part of overall SEO because it still makes pages eligible for rich results, and other engines do not publish what they read from it. Treat it as a way to state your facts unambiguously, not as a switch that produces citations.

Two types do the most for a business. Organization markup on the homepage, or on one page that describes the business, names the business, website, logo and contact details, and its sameAs property lists your official profiles, such as Google Business Profile, LinkedIn, a trade body listing or a company register entry, which helps an engine connect them. LocalBusiness markup, or a more specific type such as Dentist, Plumber or Restaurant, adds the address, phone number, opening hours and location that local questions depend on; name the areas you serve in the page text as well.

Keep the markup matched to the page and to your listings. Every fact in it should also be visible on the page and agree with your Business Profile and directory entries, down to the phone number. Owners often find that mismatched sources lie behind an engine giving old hours or a wrong address, so change every copy on the same day.

Does your website need an llms.txt file?

An llms.txt file is optional and low priority: Google Search ignores it, and none of the crawler pages from OpenAI, Anthropic or Perplexity says it is used to choose sources.

The llms.txt idea is a Markdown file at the root of a website that lists its most useful pages for language models. Google's guide says such files will neither harm nor help visibility in Google Search, because Google Search ignores them. The other engines' crawler documentation describes robots.txt rules and bots, and says nothing about reading llms.txt to decide what to cite.

The difference from robots.txt is fundamental. Robots.txt is an established standard that every search bot named on this page reads before crawling, and it grants or refuses access, while llms.txt grants or refuses nothing. If your platform makes one easy to publish it does no harm, but the hour is better spent on clear answers at the top of your five most important pages.

Why do mentions on other websites matter for AI answers?

AI engines answer "which business should I use" questions largely from what other websites say. Being named on trusted lists, directories and press pages matters as much as your own pages.

Ask any engine for the best accountant or roofer in a town and look at the sources: usually directories, local guides, comparison articles, review platforms and forum threads. An engine cannot take a business's word that it is the best, so it looks for others saying so. In what owners observe, the businesses that appear most are named consistently, with the same name, place and specialism, on several independent websites.

The work is ordinary reputation building with this in mind. Get listed in the directories and trade bodies your customers use, offer expert comment to local and trade press, and make sure "best of" lists for your area know you exist. Pages that mention you without linking are worth finding too, and the AI Backlink Checker lists them beside the sites that link to you.

Community discussions on sites such as Reddit appear often in AI answers, but they reward genuine participation only. Answering questions in your field under your real name, with useful detail and no sales pitch, builds the mentions engines quote. Google warns that seeking inauthentic mentions is not as helpful as it might seem, and its spam policies apply to AI features as to the rest of Search, so planted recommendations can cost you the trust you are building.

How does a local business show up in AI answers?

A local business shows up in AI answers through its Google Business Profile, its Bing Places listing, its reviews and a website that states clearly what it does and where.

Local questions such as "emergency electrician near Didsbury" depend on facts engines take from business listings. Google says Business Profiles can help products and services be visible in AI responses, and Bing advises Bing Places so address, hours and contact details stay current. Complete every field, choose the most specific category, add real photos and keep holiday hours updated.

Reviews shape what engines say about a business as well as whether they mention it. An engine summarising reviews repeats what customers say most often, so reviews that mention the service, the place and the outcome give it specific material. Ask every satisfied customer with a direct link, reply calmly to critical reviews, and never buy reviews or offer anything in return for one, which Google's policies forbid.

Your website then ties the facts together. Each service should have its own page that opens with an answer naming the service and the area, written in plain text rather than only in an image or map. Your name, address and phone number should match your profiles exactly, with LocalBusiness markup repeating the same facts.

How do you measure whether AI engines cite your website?

Measure AI search visibility three ways: the engines' own reports, AI referral visits in your analytics, and a monthly sample of real questions asked to each engine.

Two engines now report citations directly. Search Console's Generative AI performance report shows impressions of links to your website in AI Overviews and AI Mode by page, country, date and device, and Search Console Help says it reached all websites worldwide on 31 August 2026. Bing Webmaster Tools' AI Performance report does the same for Copilot and Bing's AI summaries; on a quiet website either may stay blank for a while.

For the other engines, use referral visits in Google Analytics 4. In Reports, Acquisition, Traffic acquisition, set the dimension to session source and look for chatgpt.com, perplexity.ai, claude.ai, gemini.google.com and copilot.microsoft.com. Since 13 May 2026, Google Analytics 4 groups visits from AI assistants into a built-in AI Assistant channel, so check where each source lands; clicks from AI Overviews and AI Mode count as Organic Search. Some visits from AI apps arrive without referral information and count as direct, so these numbers are a minimum.

The third measure is the question panel from the ChatGPT section: the same 10 to 20 questions asked to each engine monthly, with cited sources recorded in a spreadsheet. It is the only way to see competitors' citations and the questions where you are missing. Be wary of any service claiming to show your exact position inside AI answers, since Google says no third-party tool has access to its ranking or AI systems.

What does the AI Search Visibility Checker show you?

The AI Search Visibility Checker reads your website the way ChatGPT, Perplexity, Claude, Google AI and Copilot do, and scores how quotable each page is.

It checks robots.txt for each engine's search bot and asks for your homepage as each bot, which catches firewalls and CDNs that refuse a bot robots.txt allows. Each page gets an AI Citation Score out of 100 across five areas, and the checker shows the question the page can win and the passage an AI engine would quote. Five pages are read with no account, and up to 50 with an account for 1 credit, which is about $0.65.

The website as a whole gets an AI Visibility Score, made of the page scores (55%), the AI search bots let in (20%) and how often the business is named on other sites (25%). The score predicts how quotable your pages are; it does not measure whether engines quote them today, which only the engines' reports and your question panel can show. Each area is explained on the method page, and together with Search Console and Bing Webmaster Tools it tells you which pages to fix first.

Your first week of AI search optimization

Spend the first week opening the door to every AI search bot, fixing your five most important pages and starting the measures you will track. An hour or two a day is enough.

Day one matters most, because no engine can cite a page its bot cannot reach and nothing else works until that is fixed. Days two to five repair the pages and listings that engines quote most, starting with those that answer your customers' main questions, and days six and seven set the baseline you will compare against. After the first week, rewrite two more pages a week, earn one new mention a week on a website your customers trust, and repeat the question panel and the checker every month.