Where does ChatGPT get its brand recommendations from?
Two main places.
The first is training data: the public web as it looked when the model was last trained. You can't change that retroactively.
The second is live retrieval: when ChatGPT searches the web to answer a question, it pulls from pages it can fetch right now and cites them directly. That's the part you can actually work on.
The evidence is clear on what retrieval favours. Content with citations, statistics, and quotable structure gets significantly more visibility in AI-generated answers (Aggarwal et al., Princeton, 2023).
And the audience for these answers is growing fast: OpenAI reports hundreds of millions of weekly users, with information-seeking and product research among the top use cases.
To understand the broader context behind why this works, read our explainer on what Generative Engine Optimization (GEO) is.
Seven steps to get your brand mentioned by ChatGPT
- Find the prompts your buyers actually use. Write out 10 to 20 questions a buyer would ask just before choosing a product like yours. Think "best X for Y," "X vs Y," "how do I solve Z." These are the specific prompts you need your brand to appear in.
- Publish an answer-first page for each prompt. Open with the direct answer in 40 to 60 words, use question-shaped H2 headings, include a comparison table, and add a FAQ section at the bottom. Give retrieval something it can quote directly from your own domain.
- Get onto the pages ChatGPT already cites. Run your prompts and look at which third-party sites show up in the answers: directories, review platforms, "best of" roundups. Get your brand listed and described on those pages.
- Keep your brand description consistent across the web. Use the same one-sentence description on your website, LinkedIn, directories, and press coverage. AI models build entity understanding from repeated, consistent signals across sources.
- Add quotable facts and numbers. Publish specific, verifiable claims: pricing, customer numbers, benchmark results. Engines are far more likely to cite a concrete figure than a vague marketing claim.
- Add structured data markup. FAQ schema, Organisation schema, and Article schema give engines clean, machine-readable context about your brand and content. It takes an afternoon to add and pays off for months.
- Measure, then adjust. Re-run your prompt set weekly, track how brand coverage changes per engine, and direct your next round of work at the prompts where competitors appear and you don't.
Which tactics move fastest?
| Tactic | Effort | Time to impact | Why it works |
|---|---|---|---|
| Answer-first pages for buyer prompts | Medium | 2 to 6 weeks | Gives retrieval a directly quotable source on your own domain |
| Listings on already-cited third-party sites | Low | 1 to 4 weeks | Piggybacks on pages the engine already trusts |
| Consistent brand description everywhere | Low | Ongoing | Builds entity recognition over time |
| FAQ and Organisation schema | Low | 2 to 4 weeks | Creates clean extraction targets for engines |
| Waiting for training data to update | None | 6 to 12+ months | Outside your control — don't count on it as a strategy |
How do you verify ChatGPT is actually mentioning you?
Asking once is an anecdote, not a measurement. Generated answers change between sessions, accounts, and days.
A prompt that returns your brand name this morning may not tomorrow, and a single positive result tells you nothing about whether your coverage is improving.
What you need is repeated, dated scans: the same prompts, asked the same way, across ChatGPT, Gemini, and Perplexity, with results stored over time. Only then can you see whether your content work is actually moving the needle.
Lead Rescue runs those scans automatically every day and reports brand coverage, average mention position, share of voice against your tracked competitors, and which URLs each engine cited. You get the trend, not just a snapshot.
For guidance on which numbers to watch and how to interpret them, read our guide on how to track brand visibility in AI search.