A report template that keeps AI visibility evidence honest.

1. Executive summary

Open with the scope and the decision the report supports. State the collection window, platforms or modes, markets, languages, prompt count, competitor set, and any important limitations. Then summarize only the movements that repeated, affected a material question, or introduced a factual risk.

Keep the summary separate from interpretation. A report can say that recommendation coverage changed or that a cited source disappeared; it should label the cause as a hypothesis until the team has inspected the evidence and ruled out platform or sampling changes.

• Reporting window and last updated date.
• Platforms, modes, countries, languages, and prompt portfolio version.
• Top repeated movements, risks, and open questions.
• One sentence on what the report does not prove.

2. Baseline and metric definitions

Define every metric beside the result. Mention rate is not recommendation rate. Citation rate is not the same as a visible endorsement. Answer position needs a documented counting rule. Accuracy requires a list of facts and a reviewer, not a vague sentiment label.

Use a small metric stack: mention, recommendation, position, competitor share of voice within the defined set, citation or source coverage, factual accuracy, and stability. Keep the raw answer or an auditable excerpt beside the aggregate so a reviewer can inspect the reason for movement.

3. Evidence table

Create one row for each material prompt or signal. The evidence table should make it possible to move from a report line to the answer, cited URL, competitor, and page that the team may improve. Store the collection conditions even when a platform does not expose every field.

• Prompt and intent: discovery, recommendation, comparison, alternative, implementation, or accuracy.
• Platform and mode: the exact AI surface and model context when available.
• Brand outcome: absent, mentioned, recommended, compared, cited, or inaccurate.
• Competitors and position: which brands appeared and what role they received.
• Sources: cited URL, source domain, claim supported, and freshness review.
• Reviewer note: what changed, what is unknown, and whether the result repeats.

4. Diagnosis and action register

Classify the gap before assigning work. An absent brand may need category clarity or external evidence; a mention without a citation may need a more specific source; a wrong fact may need an authoritative correction; and a volatile result may need more sampling instead of a rewrite.

Give each action one owner, one reason, one canonical URL or source target, one publication or outreach date, and one validation date. Do not change multiple unrelated variables if the goal is to learn which intervention mattered.

• Diagnosis: access, coverage, clarity, authority, freshness, accuracy, or variance.
• Action: improve, create, verify, distribute, or watch.
• Owner and due date.
• Expected evidence and re-check scope.

5. Validation and leadership takeaway

Close the report with the next observation date and the rule for calling an action successful. Re-run the same prompt set where possible, compare the answer and sources, and keep Search Console or analytics observations separate but connected by topic and URL.

Leadership should see the business implication and the confidence level: high-confidence factual risk, repeated competitive gap, directional movement, or no material change. Avoid claiming that a single generated answer proves market demand, revenue attribution, or a permanent ranking.

Frequently asked questions