How to appear in ChatGPT: a step-by-step playbook to get cited by the AI

How ChatGPT decides who to recommend (short version)

Before the steps, the minimum whys. ChatGPT combines three signals to decide which brands to mention in an answer: training data (what it learned up to its cutoff), real-time retrieval (what it searches when the prompt requires it) and content structure (how easily it can extract citable information from your site). The five steps below move those three levers, in order of impact.

Step 1: entity mapping for your brand

Your goal here is for the model to understand exactly what your brand is — category, location, use cases, competitors, value proposition — and to associate it with the right entities. Without this step, everything else is built on sand.

1. List the 5-10 entities you want to be associated with (category, geography, key use cases).
2. Audit your site: does every page make clear which entity it belongs to and how it relates to the others?
3. Create/update your Wikidata profile (free, high authority for LLMs).
4. Unify your company description across every public channel — the model penalizes inconsistency.

Step 2: citable content on your site

LLMs cite short, self-contained fragments. Your job is to structure every relevant page so any block is extractable without extra context. Practical rule: if the first paragraph of each section doesn't answer the H2 question on its own, rewrite it.

• Every H2 is a clear question or a specific statement.
• Every section opens with a 40-80 word answer that closes the idea.
• After that, develop as much as you want for humans — but the opening block has to be able to stand alone.
• Lists and tables are gold: LLMs retrieve them with very high fidelity.

Step 3: schema and structured data

Implementing Organization, Service, FAQPage, Article and BreadcrumbList in JSON-LD is the lever with the best effort/impact ratio. Two days of a competent dev and you raise the probability of literal quoting by 30-40% over the next 8 weeks.

Don't get bogged down in micro-decisions (exact type, which optional properties). Start with the basic types correctly implemented and measure. Then tune.

Step 4: build external authority

Your content and your schema are what you control. External authority — who mentions you, where, in what context — is what the model uses to decide whether to trust what you say. It's the slowest lever to move and the hardest, which also makes it the most defensive.

1. Identify 5-10 relevant outlets/channels in your category (sector publications, newsletters, podcasts).
2. Design 2-3 substantive PR angles — not generic press releases; ideas with real editorial value.
3. Request and line up 2-3 cross-citations with partners, customers and allies who carry real authority.
4. If you have interesting proprietary data (studies, benchmarks), publish it openly: it's a citation magnet.

Step 5: monitor and measure

Without monitoring, steps 1-4 are faith. Define a battery of 50-100 prompts representative of your category, run it weekly against ChatGPT, Perplexity and Google AI, and log mention rate, relative position and sentiment. Do this from day one and in 8 weeks you have an honest read of what's working.

Typical mistakes that pulverize the work

• Inconsistent brand description. If LinkedIn says one thing, your site another and directories a third, the model never consolidates the entity.
• Bad schema implementation. Better no schema than broken schema — validators flag the content as unreliable.
• Optimizing vanity prompts. "What's the best AI company in the world?" doesn't bring customers; "How do I automate customer support in a US SMB?" does.
• Not iterating. GEO isn't "I set it up and see what happens". It's a weekly loop — whoever stops measuring stops improving.