Use several layers of evidence: website changes, search data, sampled AI answers, visits, and inquiries. Record the context of each observation. One answer mentioning your business does not measure visibility for all users or prove revenue growth.
Define what visibility means
A name mention, a source link, a website visit, and a prospective customer are different events. An answer may mention a name without linking. Someone may follow a link without contacting the business. Define the business question before choosing a metric.
| Layer | Question it answers |
|---|---|
| Mention | Does the business name appear in this sample? |
| Citation / link | Does the answer link to a page owned by the business? |
| Visit | Is there a session that can be attributed to that source? |
| Inquiry | Does a relevant prospective customer contact the business? |
Build a repeatable question sample
Use questions from real customer conversations. Do not select only prompts already known to produce your brand. Group needs such as product selection, service comparisons, or location.
For each observation, record the platform, complete question, date, language, location where available, sign-in state, and any cited pages. Note whether web search was used. Preserve the answer evidence and its access limitations.
- Keep the same question set for the next comparison.
- Record answers that omit your business as well as positive ones.
- Do not quietly rewrite a prompt when the result is less flattering.
- Report sample size and observation time alongside results.
What can Google Search Console show?
Google includes traffic from AI features in Search in the overall performance report for the Web search type. Do not label every increase in Web clicks as an AI Overviews result. Combine the data with page and query notes, keeping attribution limits explicit.
Source: Google Search Central — AI features and your website
What about Bing’s AI reporting?
Bing Webmaster Tools provides an AI Performance report for citation activity in the AI experiences it covers. The figures are useful within that platform’s scope, rather than as a measure of all AI use. A difference between periods also does not establish causation.
Use an observation record with context
Use one row per observation, rather than a large number without context. This example describes a record structure; it contains no client results.
| Field | What to record |
|---|---|
| Question and context | Full prompt, platform, language, and time |
| Result | Mentioned / not mentioned; source URL if present |
| Evidence | An inspectable screenshot or answer record |
| Limits | One session, a limited sample, web search on or off |
| Next action | Unclear information or a page worth checking |
Connect observations to decisions
If answers repeatedly use outdated product information, inspect the source and its date first. If a page is being found but inquiries are a poor match, check whether the offer and service limits are clear.
Terlihat.com uses this separation to discuss implemented work and what the evidence actually shows. Continue improvements when they help prospective customers, rather than solely because they produce an attractive screenshot.
Put it into practice.
Use the self-assessment to turn these ideas into a short list of things to check. It does not scan your site or predict rankings.
Open the website self-check