Our position: An AI visibility audit only matters if it turns messy answer-engine behavior into specific fixes. Name mentions are weak evidence. The better unit is a prompt row with the exact query, answer role, source links, wrong claims, missing sources, and the page-level action that can improve eligibility, clarity, or trust signals.
What you should leave with
- Build the workbook around prompt-level evidence, not summary impressions.
- Score visibility by the brand's role in the answer: recommended, cited, mentioned, absent, or misdescribed.
- Separate source gaps from answer errors so fixes are assigned to the right page, profile, dataset, or proof asset.
- Every row should end with a repair path, owner, dependency, expected signal, and retest date.
What should an AI visibility audit template actually capture?
An AI visibility audit template should capture the full evidence trail behind each answer: the prompt, market context, platform, answer text, brand role, cited sources, missing sources, competitor presence, wrong facts, and recommended fix. The goal is to make every observation traceable enough that another operator can retest it and understand the repair path.
Copy these fields into the first worksheet: prompt_id, date_tested, platform, geography, language, device_or_mode, buyer_stage, intent_type, prompt_text, expected_brand_relevance, answer_summary, brand_role, competitors_named, cited_sources, missing_sources, wrong_or_unsupported_claims, confidence_note, screenshot_or_export_link, fix_type, fix_owner, retest_date.
Treat each row as audit evidence. Do not average away the details too early. AI Overviews, ChatGPT search, and other answer systems can expose different source links and answer shapes, so the workbook should preserve the exact prompt and result before anyone rewrites copy or argues about rankings.
| Field | What to enter | Why it matters |
|---|---|---|
| prompt_text | The exact prompt tested | Prevents rewriting the question after seeing the result |
| brand_role | Recommended, cited, mentioned, absent, or misdescribed | Scores the value of visibility, not just presence |
| cited_sources | URLs, publication names, or visible source labels | Shows which pages influenced or supported the answer |
| wrong_or_unsupported_claims | Any inaccurate, stale, or unverifiable statement | Turns reputation risk into a fixable item |
| fix_owner | Person or team responsible | Keeps the audit from dying as a report |
Evidence used in this section
Which worksheets belong in the workbook?
Use six worksheets: scope, prompt ledger, answer log, competitor share of voice, Missing Source Map, and fix plan. That structure mirrors how real audit work happens: define the market, test the questions, preserve the answers, compare competitors, diagnose source gaps, and assign repair work with retest dates.
The scope worksheet defines brands, competitors, services, regions, buyer stages, excluded prompts, and test platforms. The prompt ledger stores the approved prompt set before testing. The answer log records exact outputs. Competitor share of voice summarizes answer roles by prompt group, not just raw mention counts.
The Missing Source Map is the most useful worksheet for fixes. It lists pages, profiles, reviews, reports, third-party articles, comparison pages, and structured data opportunities that competitors appear to have and your brand lacks. Use missing-source-map and build-buyer-intent-prompt-set as companion planning references.
- Worksheet 1: Scope
- Worksheet 2: Prompt ledger
- Worksheet 3: Answer log
- Worksheet 4: Competitor share of voice
- Worksheet 5: Missing Source Map
- Worksheet 6: Fix plan
Evidence used in this section
How should the template score visibility?
Score visibility by answer role, not by name mentions alone. A brand that is recommended as the best fit has a different commercial value than a brand buried in a list, cited as a source, or mentioned with a wrong description. The score should reward usefulness, prominence, citation quality, and factual accuracy.
Use a 0 to 4 role score. 0 means absent. 1 means mentioned without useful context. 2 means described accurately but not recommended. 3 means cited or included in a shortlist. 4 means recommended for the prompt's stated need with a relevant source or clear supporting rationale.
Add modifiers instead of pretending the score is perfect. Subtract for wrong facts, stale positioning, weak source fit, or competitor-only citations. Add a note when the answer is informational rather than commercial. For a broader method, compare your sheet with ai-visibility-score-methodology and ai-visibility-audit.
| Score | Answer role | Interpretation |
|---|---|---|
| 0 | Absent | The brand does not appear in the answer |
| 1 | Mentioned | The brand appears but has little influence |
| 2 | Accurately described | The answer knows what the brand does |
| 3 | Cited or shortlisted | The brand is evidence or a viable option |
| 4 | Recommended | The brand is presented as a strong answer to the buyer need |
Evidence used in this section
How do you record sources and wrong information?
Record sources and errors at the prompt-row level. For every answer, capture visible citations, source links, source titles, unsupported claims, wrong facts, missing proof, and whether the cited page actually supports the answer. This keeps source eligibility, answer accuracy, and brand messaging problems separate enough to fix.
Use these columns: source_url, source_type, source_owner, source_supports_claim, citation_position, brand_page_cited, competitor_page_cited, unsupported_claim, incorrect_claim, stale_claim, correct_fact, evidence_needed, recommended_source_fix. The source_supports_claim field should be yes, no, partial, or unclear.
Do not assume a citation means the answer is correct. A source can be visible but weak, or the answer can attach a link that only partially supports the claim. OpenAI guidance emphasizes checking citations and accuracy, and Google guidance makes clear that structured data must match visible page content.
Method boundary: Never overwrite the original answer text after finding an error. Add a correction field beside it so the audit keeps the evidence intact.
Evidence used in this section
What makes the template useful for a fix plan?
The template becomes useful when each problem maps to a repair path, owner, dependency, expected signal, and retest date. A visibility gap might require a clearer service page, better comparison content, review coverage, third-party validation, schema cleanup, crawlable documentation, or correction of stale brand facts.
Add fix columns that force action: issue_type, affected_prompt_group, repair_path, target_asset, owner, dependency, priority, effort, expected_signal, publish_date, retest_date, retest_result. The expected signal should be concrete, such as page cited, answer corrected, competitor gap narrowed, or brand moved from absent to shortlisted.
Separate fixes by control level. Owned-site fixes include page copy, FAQs, structured data that reflects visible content, documentation, and comparison pages. External fixes include analyst pages, partner profiles, review sites, directories, and credible third-party mentions. Use templates, pricing, and sample-report to keep the workbook consistent.
- STEP 1
Classify the issue
Choose source gap, factual error, weak positioning, crawlability issue, competitor proof gap, or measurement gap.
- STEP 2
Assign the repair path
Name the page, source, profile, or proof asset that should change.
- STEP 3
Set the retest
Pick a date and reuse the original prompt so the before-and-after comparison is clean.
Evidence used in this section
What should you not put in the template?
Do not put invented revenue loss estimates, guaranteed ranking claims, private model assumptions, unverified citations, or prompts rewritten after results are known in the template. Those fields make the audit look confident while weakening the evidence. Keep the workbook boring, traceable, and honest enough to survive retesting.
Avoid columns like estimated_lost_sales unless you have a defensible model and clearly label it as an assumption. Do not promise that a fix will make an AI system recommend the brand. You can improve source clarity, factual consistency, crawlability, and proof density, but you cannot control private model behavior.
Also avoid stuffing schema recommendations into every row. Structured data is useful when it accurately represents visible content and fits the page type, but it is not a magic visibility switch. A strong llm visibility audit template should show what changed, why it changed, and what the retest found.
- No invented loss estimates
- No guaranteed AI recommendations
- No private model claims
- No unverified source claims
- No prompt rewrites after results are known
Evidence used in this section
Questions that change the decision
Frequently asked questions
Can I use this as an AI visibility audit spreadsheet?
Yes. The structure is designed for a spreadsheet: one row per prompt result, with separate worksheets for scope, prompt ledger, answer log, competitor share of voice, Missing Source Map, and fix plan. Keep raw answer evidence separate from summary scoring.
How many prompts should I include in the first audit?
Start with 25 to 60 prompts across buyer stages, use cases, alternatives, comparisons, and problem-aware searches. A smaller prompt set is fine if it is stable, documented, and retested consistently. Quality beats a giant list of vague prompts.
Which platforms should the template cover?
Cover the platforms your buyers actually use, usually Google AI features, ChatGPT search, Perplexity, Gemini, and any vertical research tools in your market. Record the platform and mode on every row because answers and source behavior can differ.
Are citations more important than recommendations?
They measure different things. A citation can show source influence, while a recommendation shows answer-level preference. The best audit records both: whether the brand was cited, whether it was recommended, and whether the cited source actually supported the claim.
When should I retest after making fixes?
Retest important prompt groups after the fix is published, indexed or crawlable, and visible to users. For most teams, a 30 to 60 day retest rhythm is practical. Always reuse the original prompt text so the comparison stays clean.
Primary sources and research
Platform documentation supports factual statements. Where we describe an audit method or prioritization rule, that is AnswerMentions' operating judgment and is labeled as such.
- [1]Google Search Central: AI features and your websiteGoogle says AI Overviews and AI Mode build on Search fundamentals and may use query fan-out to surface a wider supporting source set.
- [2]Google Search Central: creating helpful, reliable contentGoogle recommends original information, substantial analysis, clear sourcing, and content that leaves a visitor feeling they learned enough to achieve the goal.
- [3]Google Search Central: structured data policiesGoogle requires structured data to match visible content and makes clear that valid markup does not guarantee a search feature or recommendation.
- [4]Google Search Central: crawler overviewGoogle documents crawler access and robots behavior; public evidence must be reachable before search systems can reliably process it.
- [5]OpenAI: ChatGPT searchOpenAI describes ChatGPT search as providing timely web answers with links to relevant sources and publisher content.
- [6]OpenAI Help: accuracy and citationsOpenAI warns that ChatGPT can produce incorrect facts and fabricated references, so consequential claims should be checked against reliable sources.
- [7]NIST: AI Risk Management FrameworkNIST frames AI risk work around governance, mapping, measurement, and management, a useful model for separating observations from decisions.
- [8]FTC: advertising and marketing basicsThe FTC states that advertising claims must be truthful, non-deceptive, and supported by evidence when appropriate.