Test real buyer questions, save the complete answers and count mentions separately from website citations.
- Use unbranded service and location questions, with a separate branded accuracy check.
- Record the platform, date, search mode, exact prompt and sources so you can compare later.
- Treat failed tests as unknown and small prompt samples as observations, not market-wide visibility.
- Check your website foundations separately with the free AI visibility checker.
Keep asking about it in
To check whether AI recommends your business, ask realistic buyer questions without your brand name, save the answers and inspect the cited sources. Repeat the same questions under recorded conditions. A website readiness score cannot answer this question on its own.
Start with the service and place your customers search for
Choose one real service and one area you serve. For an emergency plumber in Darwin, a useful first set is:
- Which businesses provide emergency plumbing in Darwin, Australia?
- Which emergency plumbing providers in Darwin have verifiable customer reviews?
- How can I compare local emergency plumbers in Darwin? Give specific examples with sources.
These are sample research prompts, not a list of companies we have tested. Replace the service and location with your own. Keep the questions stable when repeating the test so that a changed question does not masquerade as improved visibility.
Do not ask “Why is My Business the best plumber?” and count the answer as discovery. You supplied the brand. Use a separate branded question, such as “What services does My Business offer, and where?”, to check factual accuracy.
Record the conditions before interpreting the answer
Use a fresh conversation for each question. Record the platform, displayed model where available, date, country or location setting, whether web search ran, and the exact prompt. Record whether you were signed in, without storing account identifiers or personal history in the report.
Consumer assistants can use different models, retrieval and personalisation. An API observation is useful evidence under its own conditions; it is not proof of what every ChatGPT or Claude user sees. Google’s AI features are part of Google Search and also need their own observation. Google’s AI features guidance explains the website eligibility and crawling foundations.
Our AI visibility checker offers an optional three-question Anthropic API sample with web search. It records the requested Australian location, model, time, answer and citation links. It does not simulate consumer ChatGPT or Google AI Overviews.
Count mentions and website citations separately
Use one row per question. A simple record looks like this:
| Field | What to record |
|---|---|
| Exact question | The full text you submitted |
| Test conditions | Platform, model, date, search mode and location |
| Business mentioned | Yes, no, or unknown if the test failed |
| Owned website cited | Yes, no, or unknown; save the actual URL |
| Context | Recommendation, comparison, warning, factual description or unclear |
| Other sources | The cited pages and businesses discussed |
| Evidence | The complete answer or a private screenshot |
A name match is not necessarily an endorsement. It could be a warning or an unrelated business with the same name. A directory citation is also different from a citation to your own domain. Open the sources and check that they refer to the right company and service area.
If two of three successful answers mention you, report “mentioned in 2 of 3 tested answers.” If one request fails, leave it out of the successful-answer denominator and record the failure. Do not convert timeouts into a zero visibility score. Three prompts are too small and too selective to describe your entire market.
Use the evidence to choose the next change
If your business is misidentified, check your visible name, service area, contact information and official profiles. If competitors appear, inspect what the answer cites: their own useful pages, review platforms, trade associations or editorial coverage. A citation tells you what supported that answer, not the secret ranking weight of the source.
Run the free website readiness scan separately. It checks public crawler rules, non-empty metadata, structured details and readable content. Fix a real crawl block or missing service explanation before adding speculative files or mass-producing near-identical articles.
Where the issue concerns Maps rather than an AI answer, use the Google Business Profile troubleshooting guide. Where the issue concerns crawler permissions, read AI search crawlers versus training bots.
Repeat the saved prompt set after a meaningful change and enough time for sources to be rediscovered. Keep both records. Results can vary even when your site does not change, so do not attribute a single new mention to your last edit or promise a ranking date from this sample.