How this engine builds an answer
Plain version: ChatGPT answers partly from memory and partly by looking things up. When it looks things up, it names sources; when it answers from memory, it does not, but it is still describing your brand from what the web has said about you over years.
Specialist version: two layers determine the output. The parametric layer is the model's learned representation of your entity, built from broad web text and effectively fixed between model updates. The retrieval layer is triggered situationally and pulls live pages into the context window. Your measurement has to separate the two, because a mention without any cited source is a parametric win and a cited mention is a retrieval win - and they respond to entirely different work.
Why it matters for your brand
ChatGPT has the widest general consumer and professional reach of the assistants, which means it is often the first place your category gets narrowed to a shortlist. Because much of its answering is parametric, it is also the slowest engine to change - and therefore the one where an early, consistent presence pays back longest.
It is also the engine where inaccuracy hurts most, since answers frequently arrive with no citation for the reader to check.
What moves visibility here
| Signal | Why it matters on this engine | What to do about it |
|---|---|---|
| Breadth of third-party description | Learned knowledge is built from how widely and consistently the web describes you. | Earn genuine coverage across review sites, directories, trade press and community threads. |
| Entity consistency | Conflicting category descriptions weaken classification, so you are omitted from category answers. | One category sentence, identical on your site, profiles, listings and press boilerplate. |
| Crawlability for AI agents | Browsing cannot retrieve pages your robots rules block. | Check robots.txt permits the AI crawlers you want citing you, and that key pages render server-side. |
| Answer-shaped pages | Retrieved pages are quoted from the top; buried answers are skipped. | Lead with a direct answer in the first 80 words, add specifics, dates and FAQPage schema. |
| Comparison coverage | Comparison prompts are the highest-intent ones and are usually answered from third-party round-ups. | Publish honest X vs Y pages and get into the round-ups where rivals appear alone. |
How to track it
- Step 01Fix the prompt set
25-50 real buyer questions across category, comparison, problem and product intent. Freeze it within the cycle.
- Step 02Sample repeatedly
Run each prompt many times, because one answer is a single draw, not a fact.
- Step 03Record two things per answer
Whether you were named and how, and whether any sources were cited - parametric and retrieval visibility are different problems.
- Step 04Store verbatim
Keep the full answer text so sentiment, position and factual errors can be audited later.
- Step 05Baseline, then compare
Cycle one is the reference point. Judge everything after it against that, per engine.
- Report ChatGPT separately from the other engines.
- Track how you are described, not just whether you appear.
- Log whether the answer cited anything at all.
- Fix stale third-party facts before writing new content.
- Give it time: parametric change takes multiple cycles.
- Judge visibility from one manual check.
- Assume a Google ranking transfers to a ChatGPT recommendation.
- Manufacture mentions to game the learned layer.
- Block AI crawlers and then expect to be cited.
- Report a swing without checking the sample behind it.
What to report
For leadership: your ChatGPT presence rate versus last cycle, your share of voice against the leading competitor in the same answers, and any factual error being repeated. For the working report: the split between cited and uncited mentions, the prompts where competitors appear and you do not, and the domains that show up when it does browse.
Compare against the other engines rather than a universal benchmark - see Perplexity for the contrast with a retrieval-first engine.
The full measurement model, including sampling and confidence, is in the AI visibility tracking pillar guide, and the optimisation framework is in AEO visibility tracking.