Your weekly AI visibility review: a decision log for small businesses
Your business appears in an AI answer on Monday and disappears on Thursday. Should you rewrite your website? Not on that evidence alone. A useful weekly review records what changed, separates observation from interpretation, and ends with one justified action.
This is a practical review template for small service businesses. It complements an initial AI search visibility audit and the 20-prompt visibility test. Its purpose is narrower: deciding what to do next without treating a single answer as a reliable trend.
Keep three kinds of evidence separate
Google’s generative AI performance report records impressions in its supported Search AI features. Google states that the rollout reached websites worldwide on 31 August 2026, but a site may still lack a visible report when it has insufficient impressions. The report is not a measure of your presence across every AI assistant.
Your second evidence source is a repeatable set of manual checks. Record the exact question, service, language, date, visible answer and cited URLs. A mention without a link and a link to your website are different observations. Keep branded questions separate from questions that never name your company.
The third source is your own commercial evidence: enquiries, qualified conversations and sales. An AI mention does not establish that it caused any of them. Preserve the source information you actually have; leave attribution unknown when it is unknown.
Copy this decision log
Use one row for one observation. Do not combine Hungarian and English results or several AI services into a single unnamed percentage.
| Field | What to record |
|---|---|
| Review scope | Service, language, market, device and comparison dates |
| Evidence | Export or saved answer, exact question and cited URL |
| Observation | What you can directly see, without explaining the cause |
| Possible explanation | A hypothesis that still needs checking |
| Next action | One specific page check or content improvement |
| Owner and review date | Who acts, and when the observation will be repeated |
Download the blank CSV decision log. Keep screenshots and exports in your own folder, with filenames that match the rows. The template deliberately has separate observation and hypothesis columns: that small separation makes it harder to turn a guess into a marketing claim.
A worked example without invented results
Imagine a small accounting practice whose service page says it supports small businesses but never states whether it handles webshops. In a saved answer to a webshop-specific question, the business is absent. This is a hypothetical example, not an observed client result.
The observation is simply “not mentioned in this saved answer.” The hypothesis is “the website may not explain this service clearly enough.” The action is to check the actual offer with the business owner, then add accurate scope, exclusions and a useful example to the existing service page if appropriate.
Do not create five nearly identical pages for five versions of the question. Google’s AI search guidance keeps established SEO principles relevant and emphasises useful, original content. Eligibility also does not guarantee a citation. Fix a real information gap for buyers, then repeat the same observation later.
Choose the action from the evidence
An inaccessible or unindexed page: investigate the technical issue first. The Search Console getting-started guide explains how the service supports monitoring and diagnosis.
An inaccurate description of your business: compare the answer with your service page and public business details. Correct your own inconsistent information before assuming the model needs more content.
One fluctuating answer with no corroborating signal: record it and repeat the check. Avoid a website rewrite based on one result.
A repeated buyer question your page does not answer: improve the relevant page with accurate details, then log exactly what changed. Make the answer useful to a prospective customer even if no AI system ever quotes it.
End with one decision
Keep the review small enough to repeat. Choose a few commercially relevant questions and pages, maintain comparable conditions, and record changes in the business as well as the website. A new offer or different target market can explain a change in results.
The deliverable is a short decision note: what was observed, what remains uncertain, what will change and when it will be checked again. A visibility score is only useful when its scope and denominator stay explicit.
If you want help choosing the first pages and questions to review, start with a Let’s talk about AI. Bring one important service page and the questions customers ask before they enquire.
