Complete one record per question, provider, model, date, and repetition.
Keep the raw answer and citation evidence beside this record in your working system. Separate what the answer did from why you think it did it. That distinction makes a review possible and prevents a proposed action from being mistaken for a measured fact. If you have not defined the question set yet, follow the guide to run an AI visibility audit for your brand before completing the worksheet.
Start with the complete record.
AI VISIBILITY AUDIT RECORD Company: Market/category: Buyer persona: Commercial intent: Question/prompt: Provider/model: Test date (UTC): Mention (yes/no): Recommended (yes/no): Answer position: Citations: Owned citation (yes/no + URL): Competitors considered: Accuracy notes: Evidence gap: Confidence (low/medium/high + rationale): Proposed action: Owner: Acceptance criteria: Retest date:
What to record and why.
Scope the decision
- Company
- The exact brand or product being measured.
- Market
- The category and boundaries used to decide which alternatives are relevant.
- Buyer persona
- The role, company context, and problem represented by the question.
- Commercial intent
- Discovery, comparison, validation, pricing, migration, or another stated decision.
- Question / prompt
- The exact tested wording. Do not silently rewrite it after seeing the answer.
Record the run
- Provider / model
- The named system and model identifier when available.
- Test date
- The measurement date and timezone, preferably UTC.
- Mention
- Whether the answer names the measured brand.
- Recommended
- Whether it presents the brand as a plausible option for the stated need.
- Position
- The answer order only when the format makes position meaningful.
Retain the evidence
- Citations
- Every source URL or source label surfaced with the answer.
- Owned citation
- Whether a citation belongs to the measured company, with the exact URL.
- Competitors
- Vendors considered for the same question—not every organization mentioned.
- Accuracy
- Claims that match, conflict with, or cannot be verified against current evidence.
- Evidence gap
- The smallest supported explanation for the observed recommendation difference.
Turn evidence into work
- Confidence
- Low, medium, or high, plus the evidence and repetitions supporting that judgment.
- Proposed action
- A bounded change tied directly to the gap; “publish more content” is not specific enough.
- Owner
- The person or team accountable for the action.
- Acceptance criteria
- The observable site, evidence, or answer condition that defines completion.
- Retest date
- When the same decision and recorded conditions should be measured again.
Do not upgrade confidence without new evidence.
A clear-looking answer can still be unstable. Increase confidence only when repetitions, source evidence, and comparable provider runs support the same interpretation. Keep disagreements visible.
Automate collection; review consequential interpretation.
RecoProof can automate the bounded crawl, question preparation, benchmark, extraction, and report assembly. Human review still matters when evidence is incomplete, category fit is ambiguous, or a proposed action could change positioning, pricing, or product claims.
Run this automatically with RecoProof