What Actually Gets You Cited by AI Answer Engines in 2026

More and more buyers now get their answer without clicking a single blue link. They ask ChatGPT, Perplexity, Google's AI Overviews, or Gemini, read the synthesised reply, and act on it. Being one of the sources those tools quote is the new version of ranking, and a whole discipline has grown up around it: answer engine optimisation, or AEO, sometimes called generative engine optimisation (GEO).

Most of the advice attached to it is hype. A cottage industry is already selling AEO "tricks", magic files, and dashboards, and a good deal of it does nothing measurable. This is a guide to the opposite: what the evidence actually supports, what genuinely earns a citation, and which popular tactics are theatre. The short version is that there is no trick, and that turns out to be good news, because it means the work is knowable.

First, understand what an answer engine actually does

When you ask one of these tools something current, it does not recite the answer from memory. The large language models that teams now build into their own products are frozen at a training cutoff, so for anything live the tool searches the web first. It fans a question out into several sub-queries, pulls a shortlist of pages from a search index, reads the live HTML on those pages, and composes an answer from the handful it judges most relevant and trustworthy, with links back to them.

The specifics matter, because they tell you exactly where citations come from:

• ChatGPT increasingly runs a live web search instead of answering from memory, fanning a question out into several sub-queries and reading the results before it replies.
• Google's AI Overviews are built on its ordinary search index, not a separate one, though the tie to classic rankings is loosening. 2026 analyses of where AI Overview citations rank vary by method but sit well below their 2025 levels: Ahrefs found only about 38% of citations came from the top ten organic results, down from roughly three-quarters a year earlier. Search is still the substrate, but ranking first no longer guarantees the citation.
• Perplexity is unusually citation-heavy and rewards clearly structured, question-shaped pages.

The implication is the whole game: there is no separate "AI index" you optimise for. These engines read the same web you already have, mostly through the same search infrastructure. AEO is therefore not a new channel bolted onto SEO. It is largely the same machine, with a few new sensitivities.

The biggest myth: llms.txt

The single most-promoted AEO tactic is adding an llms.txt file, a hand-curated map of your site meant to spoon-feed AI crawlers. It sounds obviously correct. The data says it does close to nothing for the engines anyone actually cares about.

• Adoption is only about 10% of sites, and, more tellingly, that same 300,000-domain study found no relationship between having an llms.txt file and how often a site gets cited by AI.
• A larger Ahrefs study of 137,000 domains found 97% of llms.txt files received no requests at all over the study window, and of the small share that did, most of the traffic came from ordinary non-AI bots.
• Google has said so directly. Its own Search guidance states you "don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities)", because Search does not use them, and Gary Illyes reiterated at Search Central Live in July 2025 that Google has no plans to support it.

The reason is simple once you picture the mechanism above: the crawlers already read your real HTML directly. A separate, hand-written file they would have to trust, and that anyone could stuff with whatever claims they liked, is not something they need or particularly want.

There is one honest caveat. llms.txt is not useless everywhere. Developer and agent tooling does read it: coding assistants, IDE agents, and some tools built on the emerging agent protocols. So it is cheap, forward-looking hygiene for a future where agents transact with your site directly. It is just not a lever that gets you quoted in ChatGPT today. Add it if you want; do not expect traffic from it.

That is the tell for AEO advice in general. If a tactic is a file or a tag you add once and never have to earn, treat it with suspicion.

What actually earns a citation

Strip away the noise and three things earn citations, in rough order of leverage. None of them are new. The AI layer has just raised the stakes on getting them right.

1. Be readable to a machine

This is the floor, and more sites fail it than you would expect. The words a buyer needs have to be present in the HTML the crawler receives, not assembled afterwards by JavaScript that many crawlers will not run. That means server-rendered content, quick load times, and a robots.txt that actually lets the AI crawlers in: GPTBot, ClaudeBot, PerplexityBot, Google-Extended and the rest. A surprising number of sites block these by accident and then wonder why they never appear. If a crawler cannot read the page, nothing else on this list matters.

2. Structure content the way engines lift it

Answer engines quote passages, not whole pages, so the pages that win are built as answers:

• Clear, question-shaped headings, with a direct, self-contained answer in the first sentence or two underneath. A reader, or a model, should get the answer without needing the rest of the page.
• Structured data. Schema markup for FAQs, articles, organisations, and author credentials hands engines machine-readable facts and the credential signals they weigh before quoting a source. It will not manufacture authority, but it removes ambiguity about what a page says and who stands behind it.
• Unambiguous identity. Make it obvious who you are and what you do, with a consistent name and details, an Organization schema, and links to profiles the engine already trusts. Ambiguity gets you dropped in favour of a source the engine is certain about.
• Specifics over adjectives. Named facts, dates, numbers, and methods get quoted. Vague marketing prose does not.

3. Be genuinely authoritative

This is the hard one, and the reason the "tricks" avoid it: it cannot be faked in an afternoon. Look again at how the engines choose. AI Overviews cite what already ranks. ChatGPT leans on consensus sources. Perplexity rewards pages that other credible pages cite. All three are proxies for a single question: does the rest of the web treat you as a source worth trusting?

That is earned off your own site, not on it. Third-party reviews, credible directory listings, mentions and links from places your buyers already read, a real presence in the communities where your topic gets discussed. There is no file for any of it. It is the actual work, and it is why a site with immaculate markup but no external standing stays invisible while a plainer, better-regarded one gets quoted.

The uncomfortable bottom line

AEO comes down to a short formula: sound technical fundamentals, content shaped as answers, and genuine authority. The genuinely new part is small. Structure your answers cleanly, keep your identity data tidy, and let the AI crawlers in. Everything else is the same discipline that has always separated the sites that get found from the ones that do not.

The hype exists because "add this file" is easier to sell than "earn a reputation". But every credible study points the same way: the engines read the real web and quote the sources it already trusts. You get cited by being genuinely worth citing, and by making it technically effortless for a machine to do so. The first half is engineering. The second half is time and substance. Neither one ships in a plugin.

Sources

• Ahrefs, 97% of llms.txt files never get read (137,000 domains, June 2026), and where AI Overview citations rank.
• SE Ranking, llms.txt shows no clear effect on AI citations (300,000 domains).
• Search Engine Land, Google says normal SEO is all you need for AI Overviews, and llms.txt won't be used.
• Originality.AI, analysis of where Google AI Overview citations rank.