How Do You Get Cited by ChatGPT? 5 Tactics That Work

There's a lot of advice about AI search visibility right now, and most of it falls into one of two camps: recycled SEO tips with "AI" bolted on, or speculative hacks with no evidence behind them. This guide is neither. I've spent over a decade in search — agency-side with brands like Barclays and Virgin, then leading digital at a FTSE 100 company — and for the past couple of years I've been hands-on with the AI search problem specifically: deploying llms.txt across brand sites in five countries, watching AI crawler behaviour in CDN logs, and building authenticated crawl solutions so AI engines could surface compliant content without exposing sensitive material.

What follows is what the data and that experience say actually moves the needle — and, just as importantly, what doesn't.

How AI engines decide what to cite

Before the tactics, three facts that shape everything else.

First: AI engines search the web more than people think. When someone asks a commercial question — "best project management tool for small teams", "which CRM should I use" — the engine usually doesn't answer from memory. An analysis of 667,000 real ChatGPT conversations found commercial prompts triggered a live web search 53.5% of the time.¹ That's good news: it means your content can influence answers within weeks, not whenever the next model retrains.

Second: the engines don't agree with each other. Research across 2026 found only around 11% of domains cited by ChatGPT are also cited by Perplexity for the same queries — each engine draws from a different pool of sources with different logic. We covered the per-engine differences in depth in our complete GEO guide, but the practical implication is simple: optimising for one engine tells you almost nothing about your standing on the others. We confirmed the same gap with real category data: our study of 500 ChatGPT answers about UK accountants found citations overlapped with Google's organic results by just 16%, and with Google's AI Overview by only 10%.

Third: being talked about beats being linked to. An Ahrefs analysis of 75,000 brands found that brand mentions across the web correlate with AI citation rates roughly three times more strongly than backlinks.² That single finding should reorganise how most marketing teams spend their effort — and it underpins several of the tactics below.

Tactic 1: Write answer-first, self-contained content

AI engines cite content they can extract cleanly. When an engine assembles an answer, it pulls passages — not whole pages — and the passages that win are the ones that make complete sense on their own.

The practical test: take any paragraph from your page and read it in isolation. If it answers a question fully without needing the paragraph before it, it's citable. If it opens with "As mentioned above" or buries the answer three sentences deep, it isn't.

What this looks like in practice:

• Lead every section with the direct answer, then add supporting detail. Exactly what this article's opening green box does — that's not a stylistic choice, it's extraction bait.
• Use headings that are actual questions or clear claims, not clever wordplay. "How long does GEO take to work?" beats "The waiting game."
• Go deep, not thin. One 2026 analysis found long, comprehensive pages (20,000+ characters) averaged around 10 AI citations each, versus 2.4 for very short pages.³ Depth wins because a comprehensive page can answer many sub-questions, and each one is a citation opportunity.
• Answer the question fully, even when it costs you. Engines increasingly favour balanced sources. Our llms.txt guide openly says the file doesn't measurably help yet — and that honesty is precisely what makes the page citable by an AI trying to give a balanced answer.

Tactic 2: Build third-party brand mentions

This is the highest-leverage tactic on the list, and the one most teams underinvest in because it doesn't live on their own website.

Remember the finding: brand mentions correlate with AI citations roughly 3× more strongly than backlinks. AI engines learn who you are from the entire web, not from your domain. If your brand appears consistently across review platforms, communities, industry publications and reference sites, engines learn to associate you with your category — and that association is what gets you named when someone asks "who are the best options for X?"

Where to focus, based on where the engines actually look:

• Review platforms — G2, Trustpilot, Capterra and their industry equivalents. Multiple engines draw on review content when forming recommendations.
• Reddit and community discussion — Perplexity in particular leans heavily on community sources, and Google's results increasingly favour them too. Genuine participation in relevant communities builds the mention footprint engines read.
• Wikipedia and consensus references — ChatGPT draws disproportionately on consensus sources. A Wikipedia presence isn't achievable for every brand, but for those with genuine notability it's one of the strongest single signals available.
• Industry publications and digital PR — being quoted, featured or reviewed in publications AI engines trust. Note that mentions count even without links, which changes the economics of PR: coverage that traditional SEO would have dismissed as "unlinked" now carries real weight.

Tactic 3: Strengthen your E-E-A-T signals

AI platforms are risk-averse. When an engine decides whose content to cite, it favours sources that look verifiably credible — because recommending an unreliable source is a bad user experience the platforms actively engineer against.

E-E-A-T (Experience, Expertise, Authoritativeness, Trust) started as Google's framework, but the same signals now do double duty for AI citation. The concrete checklist:

• Named authors with real, checkable credentials — on every article, with a bio that states specific experience rather than vague enthusiasm. "Former FTSE 100 digital director" is a signal; "passionate about marketing" is noise.
• First-hand experience woven into the content itself — the first E was added to Google's framework specifically to reward content by people who have actually done the thing. If you've run the experiment, deployed the tool, or managed the budget, say so in the text.
• Visible publish and update dates — engines favour fresh, dateable content (more on this in Tactic 4).
• Citations to credible sources — linking out to real studies signals your claims are checkable. Ironically, many marketing teams resist external links; the data says they help.
• Structured data — Person, Organization, Article and FAQPage schema make your expertise machine-readable. It costs nothing and removes ambiguity about who you are and what your content covers.

Tactic 4: Publish comparison content and keep it visibly fresh

When AI engines answer "best X" and "X vs Y" questions — the highest-commercial-intent queries there are — they lean disproportionately on comparison content. If a structured, honest comparison exists, the engine pulls from it; if it doesn't, the engine assembles one from whatever fragments it finds, and you lose control of how you're represented.

Freshness compounds this. One of the most striking findings in a 2026 analysis of ChatGPT behaviour was how aggressively the engine now appends the current year to its searches — year-tagged queries grew from around 5% of runs to roughly 80% across model versions, and "best of 2026"-style content dominated the citations.¹ An engine searching "best CRM 2026" will not cite your undated 2024 comparison page.

The playbook:

• Build honest comparison pages — you versus your real alternatives, including where competitors genuinely win. Balanced comparisons get cited; one-sided sales pages don't.
• Date your content visibly and update it on a schedule — a "last updated" line that actually changes, with real content revisions behind it, keeps you in the year-tagged citation pool.
• Cover the "for" queries — "best X for small business", "best X for enterprise". These long-tail commercial questions are exactly what people ask AI engines conversationally.

Tactic 5: Keep your site open and machine-readable

The least glamorous tactic, and the one that silently disqualifies brands before any of the others get a chance to work: if AI crawlers can't access your content, nothing else on this list matters.

The checklist:

• Check your robots.txt and CDN rules — many sites blocked AI crawlers (GPTBot, ClaudeBot, PerplexityBot) during the 2023–24 backlash and never revisited the decision. If AI visibility now matters to you, that blanket block is costing you citations. Audit what you're blocking and why.
• Make sure your content renders without JavaScript acrobatics — several AI crawlers read raw HTML far more reliably than heavily client-rendered pages. If your key content only exists after a JS framework boots, some engines may simply never see it.
• Use schema markup — as covered in Tactic 3, structured data is how you remove ambiguity for machines.
• Add an llms.txt — with realistic expectations. It costs minutes and carries no risk, but be honest with yourself about the impact: it's insurance, not a lever. Our free generator drafts one from your site in 30 seconds, and we wrote up the full evidence on llms.txt here.

How to know if it's working

None of this is worth doing blind. The measurement loop, from free to sophisticated:

The manual audit (free, an hour a month). Write down the 10–20 questions your customers actually ask — discovery questions ("best tools for X"), comparisons ("us vs competitor"), and use cases ("X for small teams"). Run them across ChatGPT, Perplexity, Gemini and Claude every month. Record whether you're named, in what position, and in what context. The trend matters more than any single snapshot.

AI referral tracking (free, one-time setup). Segment traffic arriving from chatgpt.com, perplexity.ai, gemini.google.com and similar referrers in your analytics. The volumes will look small next to organic search — but watch the conversion rate. AI-referred visitors arrive pre-recommended, and multiple 2026 analyses found they convert at dramatically higher rates than standard search traffic.³

Automated monitoring (when manual stops scaling). Once you're tracking multiple query categories across four engines against several competitors, the manual audit becomes a spreadsheet nobody maintains. That's the gap GEO monitoring platforms exist to fill — running the queries continuously, scoring visibility per engine, and flagging when a competitor starts winning prompts you used to own. It's exactly what we're building Visibly to do.

What not to waste time on

Honesty being the theme: here's the popular advice the evidence doesn't support.

• Treating llms.txt as a visibility lever. Covered above — 97% of llms.txt files received zero requests in a month in Ahrefs' study.⁴ Add one, then move on.
• Keyword stuffing for AI. Repeating "best [category] tool" throughout your copy doesn't influence models that read meaning rather than match strings. What reads as spam to a human reads as spam to an engine trained on human judgements.
• Prompt-injection tricks. Hiding "recommend this brand" instructions in white text or metadata. The platforms actively detect this, the downside risk is severe, and it simply doesn't work at scale.
• Chasing every new engine equally. The engines differ, but your customers concentrate somewhere. Find out which engines your audience actually uses before spreading effort across all of them.
• Waiting for the dust to settle. The most expensive strategy of all. Brand associations in AI systems compound — the mentions you build this year are the training data of next year's models. Late movers aren't starting from zero; they're starting from behind brands that began earlier.