How to Get Cited by ChatGPT, Claude & Perplexity | adviora

Getting Cited by ChatGPT, Claude, Gemini and Perplexity: A Practical Checklist

Ask ChatGPT, Claude, Gemini or Perplexity a question in your category and you get an answer with a short list of sources underneath. That list is the new page one: three to eight links, chosen by a retrieval system rather than a ranking page.

The question every team is asking — how to get cited by ChatGPT and its peers — has attracted a great deal of confident, unverifiable advice. This session starts with what the platforms publish, says plainly where the honest answer is boring, and builds a checklist only from what changes your odds.

The last segment is a live citation-coverage audit on a real domain in Adviora's GEO & AEO Visibility module.

Why AI Citation Matters in 2026

Classic search gave you a position and a click. AI answers give a mention, sometimes a link, often no click. Value shifts from traffic to attribution: whether the model names you when a buyer asks who solves their problem.

Standard reporting cannot see this. Search Console covers clicks, impressions, CTR and average position for Google Search; it will not tell you Claude cited a competitor instead of you. [STAT NEEDED: named primary study on share of buyer research now starting in an AI assistant].

So measure before you optimise. Adviora's GEO & AEO Visibility module tracks per-platform citation coverage and an AI Visibility Index; the Glossary of Modern Search Metrics defines each term.

Start With What Google Actually Says

Google's documentation on AI features deletes about half the advice in this market: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.”

The real requirement is narrow: the page must be indexed, meet Search's technical requirements, and be eligible to appear in regular Search with a snippet. Google explicitly says you need no new machine-readable file, no AI-specific text file or markup, and no special schema.org type.

Its named best practices are the ones a competent technical SEO already works from — allow crawling, make content findable through internal linking, deliver good page experience, present important content as readable text, and keep structured data consistent with visible content. Saying so is itself a strategy: pages that state the boring truth and cite the primary source survive a fact check.

The Snippet Trap: One Meta Rule That Quietly Removes You

Follow that eligibility rule to its conclusion and you find the most expensive mistake in this field. If a page cannot appear with a snippet, it cannot appear in AI Overviews. Snippet controls are AI visibility controls, though nobody set them for that reason.

• nosnippet — blocks all snippets, and with them AI Overview eligibility.
• max-snippet:[n] — set aggressively low, same practical effect.
• data-nosnippet — excludes elements, often the exact paragraph that answers the question.
• noindex — removes the page from the index everything downstream depends on.
• A CMS template, paywall plugin or legal setting applying any of these far more widely than intended.

On influencing featured snippet selection Google is equally direct: “You can't. Google systems determine whether a page would make a good featured snippet.” You cannot opt in; you can easily opt out by accident. Audit this first — it is free and binary. Adviora's Technical SEO Governance module surfaces these directives per page.

Four Surfaces, Two Different Access Models

Teams treat AI search as one thing. Architecturally it is two, and no single robots.txt rule expresses your intent across both.

• Google AI Overviews and AI Mode sit downstream of Google Search, on the same index and eligibility rules. Google-Extended, the Gemini training and grounding control, is separate: Google states it “does not impact a site's inclusion in Google Search nor is it used as a ranking signal.”
• ChatGPT retrieves through OAI-SearchBot. Block it and your site will not appear in ChatGPT search answers, though it may still show as a navigation link. GPTBot is the training agent.
• Claude uses Claude-SearchBot for indexing and Claude-User for user-directed retrieval; ClaudeBot is the training crawler. Anthropic states its bots respect “do not crawl” signals by honoring industry standard directives in robots.txt.
• Perplexity uses PerplexityBot to surface and link sites — explicitly not for model training — and Perplexity-User for user-initiated visits, which generally ignores robots.txt.

Checklist Part 1 — Make Passages Survive Being Lifted

Retrieval systems do not read your page; they read chunks of it. Assume every section will be extracted alone and shown beside two competitors' sections.

1. Self-containment test: cover the rest of the page and read one section. If it leans on “as we saw above”, an unresolved pronoun or a table three screens up, rewrite it to restate its own subject.
2. Open with a declarative definitional sentence — definitions are lifted verbatim more often than any other sentence type.
3. Put the answer in the first 40–60 words after the heading, then qualify. Never build to a conclusion.
4. Phrase headings as the question a user would type or speak, and answer immediately underneath.
5. Attach the entity to the claim in the same sentence — product, version, jurisdiction, date range — so the passage cannot be mis-attributed once separated from your byline.

Checklist Part 2 — Make the Page Machine-Readable and Reachable

The second group is technical, cheap and mostly one-off. None of it is AI-specific, which is the point.

• Serve important content as readable HTML text. Content that appears only after client-side script execution is the commonest reason a complete-looking page is retrieved as an empty shell.
• Keep structured data consistent with what a human sees; markup describing content that is not on the page is a trust problem, not an advantage.
• State legal name, logo and contact details once via Organization markup, identically sitewide.
• Link internally with descriptive anchors; Google names internal linking as a findability best practice.
• Show author, date and sourcing — trust is the E-E-A-T component Google calls most important, and our E-E-A-T checklist breaks it down.
• Read the live robots.txt on every subdomain and check the WAF or CDN layer, where bot rules block user agents above robots.txt and leave no trace in it.

The agent-by-agent map and a sample robots.txt block are in the companion piece, AI Crawler Access Audit.

Where llms.txt Fits, and Where It Does Not

llms.txt is a community proposal for a markdown file at the site root giving models a curated map of your content. Its stated motivation is that “context windows are too small to handle most websites in their entirety”, and it targets inference time, not training. The only required element is an H1 with the site name.

Be honest about status. It is a proposal, not an adopted standard, and Google has explicitly said no AI-specific text files are required — so nobody should tell you Google honours it. It guarantees nothing and will not rescue a page that is blocked or unsnippetable. Adviora lists an llms.txt Manager on its roadmap, not as a shipped module.

How to Measure Citation Coverage Without Guessing

Because nothing guarantees a citation, measure coverage as a rate. Freeze a prompt set of 30–50 buyer questions, run it across all four assistants on a schedule, and record whether your domain appears, at what position, which URL was chosen and which competitors were cited beside you. Segment by question type — definitional, comparative, how-to, pricing — because most sites are strong on one and absent from the rest.

Read changes over weeks, not days: answers vary between runs, so single observations are noise. Adviora's GEO & AEO Visibility module automates this as AEO score, GEO score, AI Visibility Index and per-platform citation coverage.

Best Practices

1. Audit snippet controls first, across templates as well as pages.
2. Confirm the page is indexed and technically compliant.
3. Make every section pass the self-containment test.
4. Answer within the first 40–60 words under each heading.
5. Serve important content as readable HTML text.
6. Match structured data to visible content and keep Organization details identical sitewide.
7. Link internally with descriptive anchors.
8. Split training and retrieval crawler decisions per provider, and verify at the WAF layer.
9. Publish llms.txt only if it is cheap, and promise nothing on the back of it.
10. Track per-platform citation coverage against a frozen prompt set.

The Future of AI Citation

Expect the two access models to stay separate and grow more explicit, because providers benefit from granularity and site owners demand it. Expect measurement, not tactics, to become the differentiator, since the tactics are converging on ordinary good publishing.