Getting cited by ChatGPT and Perplexity: the GEO guide

Ask ChatGPT "which real-estate photographer would you recommend on the Costa Daurada?" or "how do I choose an SEO consultant?". Somebody gets cited. The question that matters for your business: why not you? This guide isn't theoretical — every technique described here runs on the site you're reading, and part of my traffic already arrives from visitors sent by ChatGPT and Perplexity. Here's the method, from server to writing style.

First, understand how an AI "chooses" whom to cite

Two distinct mechanisms, two workstreams:

1. Training: models remember what they read during learning. Slow, diffuse, outside your direct control.
2. RAG (live retrieval): ChatGPT Search, Perplexity and AI Overviews query the web in real time and cite sources. This is where your game is played — and the criteria look like classic SEO pushed to its logical extreme: accessible, structured, factual, trustworthy content.

Practical consequence: GEO doesn't replace your SEO, it rewards it. A site invisible on Google will be invisible to Perplexity too. If your foundations are shaky, start there before optimising for AI.

Workstream 1: open the technical doors

• Allow AI crawlers in your robots.txt

  GPTBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot, Google-Extended: each is controlled individually. Blocking them 'on principle' removes you from the conversation — OpenAI documents its bots and user-agents at platform.openai.com/docs/bots.

• Publish an llms.txt file

  An emerging standard (spec at llmstxt.org): a Markdown file at your site root presenting your key pages to models. Ten minutes of work — this site has one at amory-studio.com/llms.txt if you want a live example.

• Check your site renders without JavaScript

  Several AI crawlers read raw HTML. A site whose content only appears after JavaScript execution is unreadable to them — test with 'view source', not the inspector.

• Keep structured data clean

  Schema.org markup (Article, FAQPage, LocalBusiness…) helps engines understand who you are and what you claim — the same markup that powers your Google rich results. Reference: schema.org.

The technical references for this workstream: the llms.txt specification, the OpenAI bots documentation and the Schema.org vocabulary.

Workstream 2: write to be quoted

Generative engines assemble answers from fragments. Your goal: make every section of every page a self-contained, quotable fragment.

Workstream 3: become the source AIs prefer

Models preferentially cite what looks like a primary source: original data, lived experience, identifiable expertise. Three accessible levers for an independent:

1. Publish your own numbers: your real prices, average turnaround times, a mini-comparison of your local market. A "what does X cost in 2026" page with real figures becomes THE extractable reference of your niche.
2. Sign and embody: name, bio, concrete experience on every piece (E-E-A-T). Google's helpful-content guidance — documented on Google Search Central — also shapes what AIs pick up.
3. Keep your local footprint consistent: AIs cross-reference your site with your listings and directories. An inconsistent NAP blurs your entity — see our guide on NAP and local citations.

Measure: AI traffic is already visible

In your analytics (Umami, Plausible, GA4), build a referrer segment for chatgpt.com, perplexity.ai, claude.ai and copilot.microsoft.com. Two tracking habits: the monthly trend of those visits, and a quarterly manual test — ask ChatGPT and Perplexity the 5 questions your clients ask you, and note who gets cited. That's your next-generation ranking, and it moves fast. Site speed matters in the crawl equation too — see our take on what the PageSpeed score really means.