Why brand mentions are the real AI metric
When a buyer asks ChatGPT for the best tool in your category, the shortlist it produces is the modern equivalent of page one. Being on that shortlist is binary and high-value: there is no position eight consolation prize. That makes 'how often are we named, and how are we described' a cleaner business metric than any ranking average.
It also surfaces problems ordinary SEO reporting hides. A brand can rank well in Google and still be described by ChatGPT with two-year-old pricing, a discontinued product line, or a competitor's positioning — because those are the facts the retrievable web currently agrees on.
Run a proper mention audit
Build a panel of prompts before you look at any results, so you are not tempted to cherry-pick the flattering ones. Four categories cover most buying journeys:
- Unbranded discovery — 'best tools for X', 'how do I solve Y'.
- Category comparison — 'X vs Y', 'alternatives to Z'.
- Branded verification — 'what is [brand]', 'is [brand] any good', '[brand] pricing'.
- Objection handling — 'is [brand] secure', 'does [brand] integrate with [tool]'.
Run each prompt in a fresh session with memory and personalisation off, otherwise you are measuring your own history. Log four fields: prompt, mentioned yes/no, cited with a link yes/no, and which competitors appeared. Repeat weekly. Within a month the noise averages out and the trend is readable.
Do the same panel on Perplexity, Gemini and Copilot. Overlap is high, and a source that works on one engine usually works on the others.
Diagnosing why ChatGPT says what it says
When an answer is wrong or missing, ask ChatGPT for its sources on that specific answer. The cited URLs are the diagnosis. Three patterns cover nearly every case:
| Symptom | Likely cause | Fix |
|---|---|---|
| Never mentioned in category prompts | You are absent from the roundup and comparison pages the engine retrieves | Earn placement in those specific pages; publish your own comparison with real data |
| Mentioned but described wrongly | Stale third-party profiles outrank your own current pages | Update directories and profiles, add dateModified, restate the fact plainly on-site |
| Mentioned without a link | Your site is not retrievable or lacks a page that answers that prompt | Check crawler access, then publish a page whose heading matches the prompt |
Growing mentions deliberately
- Publish one page per high-value prompt, with the prompt itself as the H1 phrasing.
- State facts about your brand — founding, pricing model, integrations, security posture — in plain, extractable sentences on a stable URL.
- Keep every off-site profile boilerplate identical so corroboration is unambiguous.
- Write the comparison pages your buyers ask for, honestly, including where you lose. Engines retrieve balanced comparisons more readily than sales pages.
- Add dateModified and keep it truthful; stale-looking pages lose to fresher rivals in synthesis.
Expect a lag. Retrieval-driven mentions can shift within weeks of a re-crawl, while descriptions baked into background knowledge only change with training cycles. Judge the programme on the trend of your panel share over a quarter, not on any single conversation.