Measuring business visibility in search and AI: a question panel and evidence log

“AI recommends us” is too broad for a business report. A brand can appear without a link; a link can point to another website; a useful answer does not necessarily bring a visitor. Conventional search, Google’s AI search features and standalone assistants require separate observations. This is a repeatable working method, not a report of SAMSA.AGENCY search research. It helps examine specific customer questions and identify content improvements, without creating a universal score that guarantees recommendations everywhere.

1. Define the events you measure

Separate search impressions, brand mentions, citations of your URL, website visits and qualified enquiries. Track factual accuracy too: a recommendation with the wrong price or service area is not useful visibility. Define a business question, such as whether pricing content brings better-prepared enquiries. Measures with different denominators cannot become one percentage. A manual panel covers selected questions; traffic data covers observed visits. Keep both layers in the report and state their different limits.

2. Build a fixed question panel

Use enquiries, briefs and sales interviews. A pilot might contain 15 questions about provider choice, service comparisons, prices, limitations and local delivery. That number is an illustration, not a market standard. Preserve exact wording and panel versions. Maintain RU, UZ and EN separately because translation can change intent. Record region, device, interface language, search mode, login state and personalisation settings. Do not rewrite a question after an inconvenient answer; add a version and retain the original for comparison.

3. Keep evidence for each observation

Open cited sources and check whether they support the specific recommendation. A homepage link without price evidence does not validate a pricing answer. Keep customer personal information out of shared logs; anonymised questions and public sources are sufficient for this review.

4. Separate platforms and technical readiness

Google describes its AI search features through established SEO foundations, without requiring special AI markup: https://developers.google.com/search/docs/fundamentals/ai-optimization-guide. Do not automatically apply that guidance to ChatGPT, Perplexity or other products. Record each platform’s available feature and observation method separately. In Search Console, use the available Generative AI performance report: https://support.google.com/webmasters/answer/16984139. If your account cannot access a report or feature, document that fact. Bot-accessible content is technical readiness, not evidence of a future citation or recommendation.

5. Improve pages as factual sources

Start with conflicting names, scope, prices, service areas and contacts. Explain price conditions and use the date of an actual content update. For complex choices, add alternatives, limitations and a clear next action. Connect the article to a service and a verified example, accurately stating the provider’s contribution. Where evidence is missing, label an example illustrative. Consistent published facts are more useful than masses of identical answers. Do not present llms.txt, FAQ content or special schema as an obligatory ticket to AI recommendations.

6. Compare like conditions

Repeat the panel on an agreed schedule, keeping conditions as consistent as the platform allows. Annotate changes in model, interface, search availability and personalisation. Look at persistence across several observations rather than selecting the best answer. Keep a few questions unrelated to site changes as context for wider movement; that is not a full control group. Platform updates, competitors and seasonality remain external factors. Manual before-and-after observations usually cannot attribute a change to one article.

7. Turn observations into a decision queue

1. For an incorrect price or condition, find the conflicting source and correct the published fact.
2. For a missing citation, inspect the relevant page’s access and completeness, then repeat the question.
3. For visits without suitable enquiries, review the offer and conversion journey alongside mentions.

Assign each decision an owner, URL and next check. Review combined delivery at https://samsa.agency/en/services/seo and project examples at https://samsa.agency/en/cases. For monthly acceptance, use https://samsa.agency/en/blog/chto-vkhodit-v-ezhemesyachnoe-seo; Yandex considerations are covered at https://samsa.agency/en/blog/seo-pod-yandex-v-uzbekistane. End the report with the next improvement priority and an explicit list of claims the available evidence cannot yet support.