Answer Engine Optimisation: Getting Cited by AI in 2026

TL:DR: Answer engine optimisation is the work of making a page easy for search engines and AI answer systems to retrieve, understand, and quote, and most of it is ordinary technical and editorial hygiene. Google documents no special requirements for AI Overviews or AI Mode beyond being indexed and eligible for a snippet, and its guidance tells site owners to ignore llms.txt files, content chunking, and rewriting for AI. What the vendors do document is crawlable pages, clear structure, factual sentences that survive being lifted out of context, and content that adds something the existing results do not. Treat any tactic you cannot trace to a vendor document as a hypothesis, not a rule.

The pitch arrives by email most weeks: guaranteed placement in ChatGPT answers, an AI visibility audit, a monthly retainer. The invoice is specific. The methodology is a slide with arrows on it.

This post goes through what Google, OpenAI, Anthropic and Bing publish about their own systems, marks where documentation ends and inference begins, and sets out the habits that survive every document we could find.

If you are commissioning the site work, what a small business site actually costs to run covers the build side.

What "answer engine optimisation" actually means in 2026

Answer engine optimisation is the practice of preparing a page so that a system which answers a question directly, rather than returning ten links, can use it as a source. That covers three different systems which are often discussed as one: the AI features inside a search engine, the crawlers that feed a chat assistant's search, and the model weights that decide what an assistant says from memory.

The distinction matters because the controls differ. Google's AI Overviews and AI Mode draw on the same index and ranking systems as ordinary Search. ChatGPT's search features depend on OpenAI's crawler permissions. A model answering from training data has already been trained, so nothing you change today affects that answer.

The other distinction is between documentation and folklore. Several widely repeated AEO tactics are contradicted by the vendor that supposedly rewards them, and others are simply unmeasured. The citations behind each row appear below.

What Google documents about AI answers

Google's position is specific, and the opposite of what most AEO advice assumes. To appear 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", per Google's AI features documentation, last updated 10 December 2025.

The mechanism is retrieval, not a separate ranking system. Google describes AI Overviews and AI Mode as using retrieval-augmented generation over its existing index, plus "query fan-out", which issues concurrent related queries to gather more sources, per Google's guide to optimizing for generative AI features, last updated 10 July 2026. A page that ranks for the sub-queries inside a fan-out can be retrieved even when it never ranks for the original question. That is an ordinary keyword-coverage problem.

The same guide has a section titled "Mythbusting generative AI search: what you don't need to do", which addresses AEO and GEO by name. It says you can ignore llms.txt files and special markup, chunking content, rewriting for AI systems, and chasing inauthentic mentions. On structured data: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add."

Two operational details follow. A site must be included in the Search generative AI features setting in Search Console to be eligible for display there, and performance is reported in the Generative AI performance report. Google's earlier AI features page adds that AI appearances are counted inside the ordinary "Web" search type, so your existing reporting already contains them.

The FAQ markup case, as a worked example

The FAQ rich result was deprecated in May 2026: Google said it would no longer appear in Google Search from 7 May 2026, and removed the documentation in June 2026, per Google's Search Central documentation changelog. Schema.org still defines the type: an FAQPage is "a WebPage presenting one or more Frequently asked questions", per schema.org's FAQPage definition.

So the markup remains valid and harmless, and the Google rich result it used to earn is gone. Any agency selling FAQ schema as an AI visibility tactic in late 2026 is selling a feature switched off seven months ago. Markup that matches visible content is still worth having, which is a different argument from the one on the invoice.

What OpenAI and Anthropic document about their crawlers

Chat assistants do not use one crawler, and the naming is where confusion starts.

OpenAI runs separate user agents with separate jobs. OAI-SearchBot "is for search" and is used "to surface websites in search results in ChatGPT's search features", while GPTBot signals whether crawled content may be used for training. The settings are independent: "a webmaster can allow OAI-SearchBot in order to appear in search results while disallowing GPTBot", and a robots.txt change takes about 24 hours, per OpenAI's crawler documentation. ChatGPT-User covers actions a person initiates, where OpenAI notes robots.txt rules may not apply.

Anthropic uses three robots for three jobs. ClaudeBot collects content that may contribute to model training, Claude-User handles retrieval when a person asks Claude a question, and Claude-SearchBot indexes content for search quality. Blocking Claude-SearchBot "may reduce your site's visibility and accuracy in user search results", per Anthropic's crawler support page. All three honour robots.txt, and Anthropic supports Crawl-delay.

The practical consequence: blocking every AI crawler is a decision to be absent from those answers, and allowing all of them is a decision to be used for training. Those are two choices, made per vendor, and worth writing down rather than leaving at whatever the CDN default happens to be.

What Bing documents about grounding and citations

Bing is the most explicit of the four about citations. Its webmaster guidelines now state that they describe "how Bing discovers, crawls, indexes, evaluates, and surfaces content across Bing search experiences, Copilot, and grounding API results", and that following them helps a URL become eligible for "grounding results and citations", per Bing's Webmaster Guidelines.

The guidance is conventional, which is the point. Crawl efficiency, indexing accuracy, URL consolidation, content clarity and structure, and authority signals are the listed supports for grounding eligibility. IndexNow and XML sitemaps are named as discovery methods, with accurate lastmod values and ETags as freshness signals.

Bing's guidelines also carry the clearest published warning about gaming the systems. Markup that is "irrelevant, inaccurate, or misleading may be ignored and can affect trust and eligibility for enhanced features", and content "designed to manipulate or interfere with language models used by Bing or Copilot may result in reduced visibility or removal". A tactic that only works if the model does not notice it is a documented liability.

The habits that hold up against the documentation

Four things recur across the documents that matter here, and they are unglamorous.

Be crawlable, and check it yourself. Indexed and snippet-eligible is the whole technical bar for Google's AI features, and crawl access is the bar everywhere else. Confirm the page renders as HTML with the content present, that robots.txt and any CDN rule allow the crawler you want, and that the canonical is correct.

Put the answer in the first sentence of each section. This is editorial, not technical, and no vendor requires it. It is also what makes a passage quotable: a sentence that begins with the answer survives extraction, while three clauses of setup do not. Write each section as if a person will read only that paragraph.

Make numbers specific, and date them. "Starter seats start at $20 per month" is citable. "Affordable pricing" is not. Where a figure comes from a vendor page, name the page and the date you read it, as the HubSpot pricing breakdown does.

Add something the existing results do not have. Google's framing is "non-commodity" content: a first-hand review rather than a summary of what already ranks, or a worked example rather than a definition. This appears in Google's guidance and in Bing's trust signals in different words, and it cannot be automated.

Two supporting habits are worth adding because they cost almost nothing: keep a visible author and a dateModified that reflects real edits, and use tables for comparisons so a figure sits next to the thing it describes.

How to measure this without inventing numbers

The share of searches that produce an AI-generated answer is not published by any of the four vendors in the documentation cited here. Percentages quoted in most decks come from third-party panel estimates or a vendor's own marketing, and should be labelled as such.

What you can measure directly:

• The Generative AI performance report in Search Console, which Google documents as the place where AI feature performance is reported.
• Search Console's ordinary Performance report, since AI appearances are counted in the "Web" search type.
• Bing Webmaster Tools and IndexNow submissions, which Bing names as discovery and freshness signals.
• Server logs filtered for the documented crawler names: OAI-SearchBot, GPTBot, ChatGPT-User, ClaudeBot, Claude-User and Claude-SearchBot. A crawler hit is not a citation, but a crawler that never arrives is a page that cannot be cited.

The honest summary is that citation volume is a lagging, noisy signal. Track whether cited pages are crawled, whether the quoted passage was the sentence you wrote, and whether the traffic converts.

Integrating with the rest of your stack

The work above is content and infrastructure, and both live in one place: your repository or CMS. That is the argument for keeping the site generator, the schema and the sitemap under one roof, so a page change and its metadata ship together rather than drifting across two vendors.

Three practical checks. Serve the content as HTML rather than assembling it client-side, covered in why page speed and rendering decide who gets found. Keep internal links crawlable in plain anchor tags, since both Google and Bing name internal linking as a discovery path. Generate structured data from the same source as the page, so the markup cannot contradict the text.

If your site is a custom build, the features page lists what ships with a BareStack build, and the comparison page sets it against a hosted stack on cost and control. How a headless storefront was built for a client is a worked example of the same architecture choice, and the BareStack FAQ covers hosting and maintenance.

Bottom line

Answer engine optimisation is the disciplined version of SEO: crawlable pages, clear headings, specific facts with dates, and content that adds something the top results lack. Google documents no special technical requirements for its AI features beyond being indexed and snippet-eligible, and it tells site owners to ignore llms.txt, chunking, and AI-specific rewriting. Check every tactic against a vendor document before you pay for it.

FAQ

What is answer engine optimisation?

Answer engine optimisation is the practice of preparing content so that AI answer systems can retrieve, understand, and quote it in a direct response. It covers search engines' AI features, the crawlers behind chat assistants, and the editorial habits that make a sentence quotable. Most of it overlaps with technical and on-page SEO, because the answer systems are built on the same index.

Does answer engine optimisation actually work?

The documented parts work, because they are the same crawl, index and clarity signals search engines already use. Google states that AI Overviews and AI Mode rely on its existing ranking systems, so a page that ranks well and answers precisely is eligible to be cited. The undocumented parts, such as special AI files or rewritten copy, have no published mechanism behind them and should be tested rather than bought.

Is llms.txt worth adding?

Not for Google. Its guidance states that you do not need to create "AI text files" and that Google Search does not use them, per Google's own AI optimization guide. The file is cheap to publish and some other services may read it, so adding one is harmless. Paying an agency to build one as a visibility tactic is where the value disappears.

How do I get my website cited by ChatGPT?

Allow OAI-SearchBot in robots.txt, since that crawler is what surfaces sites in ChatGPT's search features, and keep the pages crawlable and clearly structured. Allow GPTBot only if you also want the content used for model training, because the two settings are independent. Then publish pages that answer a specific question well enough to be worth quoting, since retrieval still selects on relevance.

Do I need FAQ schema for AI search?

No. Google states that structured data is not required for generative AI search and that there is no special schema.org markup to add. The FAQ rich result was also deprecated in May 2026, so the markup no longer produces a Google search feature. FAQPage remains a valid schema.org type, and the questions themselves are still useful because they match how people phrase queries.

What is the difference between SEO and answer engine optimisation?

SEO aims to rank a page in a list of results, and AEO aims to have a passage from that page quoted in a generated answer. The inputs overlap: crawling, indexing, page structure, clarity and trust. The output format rewards short self-contained answers, specific figures and tables more than a long narrative.

---

Photo by Campaign Creators on Unsplash

Photo by lilartsy on Unsplash