AI visibility: how to appear in ChatGPT, Gemini and Perplexity answers

A growing share of your customers never see a search results page. They ask ChatGPT, Gemini or Perplexity - or they read Google's AI Overviews box and stop there. If the assistant's answer recommends you, you win the customer before a single blue link is clicked. If it recommends a competitor - or has never heard of you - you're invisible at the exact moment of intent.

Being present in those answers is called AI visibility, and the practice of earning it is Answer Engine Optimization (AEO) - also known as Generative Engine Optimization (GEO). It is the fastest-moving front in search, and most businesses aren't watching it at all. This guide covers how answer engines choose what to say, and the concrete steps that make your brand citable.

How answer engines decide what to recommend

Assistants don't rank pages; they compose answers. Three mechanisms feed them:

• Training knowledge - what the underlying large language model absorbed about brands and categories during training. Slow to change, but shaped by everything public about you.
• Retrieval (RAG) - for questions that need fresh facts, engines fetch live pages and ground the answer in them. Perplexity does this for almost everything; ChatGPT and Gemini do it whenever the question smells current. If your pages can't be fetched, parsed and quoted, you can't be cited.
• Citations - when an engine names or links a source, that's an LLM citation - the AEO equivalent of ranking #1. Citations concentrate on pages that answer questions directly, carry clear structure, and agree with what other sources say about you.

The practical upshot: AI visibility is earned on your website, in your reviews, and in the consistency of your public footprint - the same raw material as SEO, read by a different kind of machine.

Step 1: measure where you stand

You can't optimise what you can't see. Before changing anything, get a baseline on two fronts:

• Your site's readiness. AI92 publishes a free AI Search Readiness Grader - it checks a URL for the signals answer engines depend on (crawlability, structure, answerability) and grades it, no signup. Run it on your homepage and your top product or service page.
• What the engines currently say. Ask ChatGPT, Gemini and Perplexity the questions your customers ask ("best [category] in [city]", "alternatives to [competitor]") and note whether you appear, how you're described, and who is cited instead. Different engines answer differently, so check more than one.

On paid AI92 Plans this measurement runs continuously: the platform tracks whether the assistants mention you for the questions that matter in your market, how often - a share of voice for AI - and how that compares, engine by engine, against tracked competitors. From the Purple Plan up, Sentinel alerts notify you the moment your rankings or AI-engine mentions move. But even done by hand once a quarter, the baseline is worth the hour.

Step 2: make your site fetchable by AI crawlers

Retrieval-based answers start with a fetch. Two files on your server decide whether that fetch succeeds:

• robots.txt - AI engines send their own crawlers (GPTBot, Google-Extended, PerplexityBot and others), and many sites block them accidentally - often via an old blanket rule. AI92's free robots.txt checker shows what your file actually allows, crawler by crawler.
• llms.txt - an emerging convention: a plain-text map at your domain root that tells language models what your site offers and which pages state it best. It's cheap insurance, not magic. The free llms.txt generator builds one from your inputs; ai92.ai publishes its own at /llms.txt.

While you're at it, make sure your key pages render their substance as real HTML text - an answer engine quoting your pricing can't read it out of a JavaScript-only widget or an image.

Step 3: structure content so it can be quoted

Answer engines assemble responses from fragments. Pages win citations when a fragment can stand alone:

• Answer the question in the first sentence of a section, then elaborate. The heading "How much does X cost?" followed immediately by the number is quotable; three paragraphs of preamble are not.
• Use headings that are questions or claims, not slogans. "Works with WhatsApp and Telegram" beats "Communication, reimagined."
• Publish structured data. Schema.org markup (Organization, FAQPage, Article, Product) hands engines your facts - name, prices, ratings - in machine-readable form instead of forcing them to guess. AI92's free schema validator checks what your pages currently declare.
• Keep one page per real question. A focused page on a specific service in a specific city is far more citable than a mega-page that mentions everything once. A glossary of your domain's terms is a classic AEO asset for the same reason.

Step 4: build the reputation the engines cross-check

Engines corroborate. Before recommending a business they weigh what others say - reviews, ratings, mentions across the public web. Two implications:

• Reviews are AEO. Volume, recency and your responses all shape how models describe you, especially for local queries where assistants lean on business-profile data. Keeping your Google Business Profile active and answered is answer-engine work, not just local-SEO work.
• Consistency matters. Your name, category, address and story should match across your site and profiles - contradictions make a model hedge, and a hedging model recommends someone else. (Models also simply make things up - the industry politely calls it hallucination - and a strong, consistent public footprint is your best defence against being misdescribed.)

Step 5: keep classic SEO running - it's the same substrate

AEO does not replace SEO; it reads the same signals through a different lens. Site health, Core Web Vitals, keyword-focused content and clean information architecture still decide whether you're findable - by crawlers of either kind. Treat being found as one job with two surfaces: the classic one you already think about, and the AI one most of your competitors aren't watching yet.

This is exactly how AI92 treats it internally: every Plan runs the classic side - site audits, tracked keywords, Core Web Vitals and content tuned for how AI search reads your site - and from the Standard Plan up, the AI-Visibility dashboards show exactly how ChatGPT, Gemini, Perplexity and Claude describe and cite your brand versus your competitors, engine by engine, including how many AI answers have been analysed. Different engines can say different things about you; a gap on one engine is a targeted opportunity, not a verdict.

Quick wins this week

If this is all new, five actions cover the highest-leverage ground in an afternoon:

1. Run the AI Search Readiness Grader on your homepage and fix what it flags first.
2. Check your robots.txt against the AI crawlers with the robots checker - unblocking a bot is a one-line change.
3. Publish an llms.txt with the generator.
4. Add or validate FAQPage and Organization structured data with the schema validator.
5. Ask three assistants your three most valuable customer questions and write down who they cite - your baseline for next quarter.

The honest timeline

AI visibility compounds slowly. Making your site fetchable and quotable pays within weeks on retrieval-based engines like Perplexity; shifting how trained models describe you takes months of consistent public signals. Start measuring now precisely because the payoff lags - every month unmeasured is a month a competitor may be quietly becoming "the" answer in your category.

For the deeper mechanics - including per-engine tracking and how share-of-voice is scored - the free SEO & AI Visibility course in AI92 Academy covers the whole discipline in 13 video lessons. For how this fits the broader autonomous cycle, see What is autonomous marketing? - and if you want the measurement running for you every cycle, the Plans that include it start at $250/month.