Write five questions a buyer would ask without naming you. Run each on ChatGPT, Gemini and Perplexity. For every answer, record whether you were named, whether your site was linked, and who else was named. That is 15 answers and about 30 minutes. Answers change from run to run, so read the result as a pattern.
Write five questions the buyer would ask
Two category questions (“best [what you sell] for [who you serve]”), one comparison (“[a competitor] vs [another]”), one problem question, one local or situational question. Never use your own name: the point is whether the engine brings you up unprompted.
Keep them short, the way a buyer types. Phrasing matters: in a June 2026 vendor study, short conversational prompts produced brand mentions far more often than long structured ones on the same topic (Semrush, 9 Jun 2026).
Run each once on each engine
Use a fresh conversation each time, with web search switched on where the engine offers the choice. Five questions on three engines is 15 answers, which fits in about 30 minutes.
Record two things separately
Named: your brand name appears in the text. Linked: your own site appears as a source. Never merge them: an engine can use your page and not say your name, or name you without linking. Also note your position in any list and every competitor named.
Insight
Named and linked are two different results. Keep them in two columns, never one number.
Read the sources
For each answer, note which domains were cited. If the same third-party pages keep appearing, those pages, not your own site, are what the engine is reading about your category. Check that they describe you correctly.
Fill in one table
| Question | ChatGPT | Gemini | Perplexity |
|---|---|---|---|
| Best [product] for [customer] | named · linked | named | – |
| [Competitor] vs [competitor] | – | named | linked |
| … |
The entries above are made up to show the layout. Filled in, fifteen cells with two marks each tell you more than any single score.
If you have another hour
Run every question twice more, in fresh conversations, and add a long version of each with extra context about the buyer. Then a cell reads “named in 2 of 3, linked in 1 of 3”, which is far harder to dismiss as an outlier.
What this cannot tell you
Fifteen answers give you a first read. Different engines and phrasings give different results, and a single check is a snapshot. To see a trend, repeat it the same way in about six weeks. That is why our Full Package includes a 45-day re-audit with the same questions and engines.
Want it done for you?
The free Live Check asks one buying question on one engine and shows whether you were named, where, and who was named instead. The Core Audit runs 20 questions across four engines.