How to audit AI visibility

What is an AI visibility audit?

An AI visibility audit is a structured test of how your brand appears inside generated answers. It is not a keyword ranking report. The unit of measurement is the prompt: a buyer question asked to an AI engine.

The audit should show whether your brand is cited, mentioned without a citation, absent, misrepresented, or replaced by a competitor. It should also show which sources the engine used so the next fix is grounded in evidence rather than opinion.

• Brand cited: the answer links to or names your domain as a source.
• Brand mentioned: the brand appears but is not cited as a source.
• Brand absent: the answer does not name you.
• Competitor replacement: another brand is recommended where you should be considered.
• Source gap: cited pages support competitors or the category but not your brand.

Step 1 — Define the prompt set

The prompt set determines the quality of the audit. A broad random list creates noise. A strong list mirrors buyer behavior: category discovery, alternatives, comparisons, problem-solving, use cases, and proof questions.

Keep the prompt set stable. If the prompt changes every time, you cannot tell whether visibility improved or the test simply moved the target.

• Category prompts: best AI visibility tools, top citation monitoring platforms.
• Alternative prompts: Profound alternative, Semrush alternative for AI visibility.
• Comparison prompts: RankEcho vs Profound, AI citation tracker vs rank tracker.
• Problem prompts: why ChatGPT does not mention my brand.
• Use-case prompts: AI visibility tool for agencies, SaaS, ecommerce, or enterprise.
• Proof prompts: how to prove AI citation changes after a fix.

Step 2 — Choose engines and run conditions

Different engines behave differently. ChatGPT, Perplexity, Claude, Gemini, Copilot, and Google AI Overviews may use different retrieval layers, citation rules, and source preferences. A serious audit should not assume one engine represents the whole market.

Run prompts under consistent conditions. Record the date, engine, prompt, response, cited URLs, brand mentions, competitor mentions, and any notes about uncertainty.

• Use the same prompt wording across engines.
• Record the exact response, not only the score.
• Track engine-by-engine variance.
• Separate live retrieval behavior from slower model-memory behavior when possible.
• Avoid changing prompt wording during before/after tests.

Step 3 — Record the right metrics

A useful audit records more than a visibility score. The score is the summary; the evidence is the prompt-level table behind it.

The minimum metrics are citation rate, mention rate, share of voice, competitor replacement, owned citation share, third-party source share, and source-type mix. Each metric answers a different question about visibility.

• Citation rate = prompts where your brand is cited / total tracked prompts.
• Mention rate = prompts where your brand is named / total tracked prompts.
• Share of voice = your appearances compared with named competitors.
• Competitor replacement = prompts where a competitor is recommended instead of you.
• Owned citation share = citations to your domain / total citations involving you.
• Source mix = owned, review, forum, directory, publication, docs, social, or community sources.

Step 4 — Classify the source layer

Source classification turns a raw answer into an action plan. If the engine cites your owned page, the next fix may be on-page. If it cites third-party roundups where competitors appear and you do not, the fix may be external source coverage.

The source layer is especially important for commercial prompts because AI systems often prefer corroborated third-party evidence over brand-owned claims.

• Owned source: your website, docs, blog, or product pages.
• Review source: G2, Capterra, Product Hunt, marketplace listings, or review articles.
• Community source: Reddit, forums, Stack Exchange, or niche communities.
• Editorial source: articles, roundups, guides, and publications.
• Directory source: category listings or vendor databases.
• Documentation source: help docs, developer docs, changelogs, or API references.

Step 5 — Diagnose the gap behind each loss

Every failed prompt should be assigned a likely gap type. This prevents generic recommendations and makes the next fix concrete.

The main gap types are access, extraction, entity, source, prompt-fit, competitor, and proof gaps. A single prompt can have more than one gap, but one is usually the highest-leverage first fix.

• Access gap: crawlers cannot fetch or read the content.
• Extraction gap: the answer exists but is vague, buried, or hard to lift.
• Entity gap: the brand-category relationship is inconsistent or weak.
• Source gap: external evidence supports competitors but not you.
• Prompt-fit gap: the prompt does not match your current content or strongest use case.
• Competitor gap: another brand owns the comparison, alternative, or category source layer.
• Proof gap: no before/after re-test exists after a fix ships.

Step 6 — Prioritize fixes by commercial value

Do not fix every gap in order of appearance. Prioritize the prompts closest to buying decisions: category shortlists, alternative searches, comparison questions, and problem prompts where a competitor is recommended.

A low-intent educational prompt may be useful for authority, but a competitor replacement in a category prompt is usually more urgent.

• Highest priority: competitor recommended for a high-intent prompt.
• High priority: brand absent from category or alternative prompt.
• Medium priority: brand mentioned but not cited.
• Medium priority: inaccurate description of your product or category.
• Lower priority: broad educational prompts with weak commercial intent.

Step 7 — Ship one fix package

A fix package is a concrete set of changes tied to one prompt gap. It should be specific enough for a writer, developer, marketer, or PR lead to execute.

The package may include a direct answer block, schema, an internal link update, a crawler-access correction, a comparison section, a new page brief, or an off-site source plan.

• On-page fix: answer block, H2 structure, FAQ, schema, and internal links.
• Technical fix: robots.txt, CDN behavior, rendering, canonical, or indexability.
• Entity fix: brand/category consistency, About page, Organization schema, external profiles.
• Source fix: earn inclusion on pages the AI already cites.
• Comparison fix: neutral comparison or alternative page for buyer evaluation.

Step 8 — Re-test and prove movement

The proof step is what makes the audit operational. After a fix ships, re-run the exact same prompt under the same conditions and compare the before/after answer.

A good proof report does not overclaim. It shows the baseline, the fix shipped, the re-test result, the time window, the citation or mention change, and the confidence level.

• Baseline response and cited URLs.
• Fix package shipped and date shipped.
• Re-test response and cited URLs.
• Citation, mention, and competitor replacement change.
• Likely pathway: retrieval, training-dependent, or unknown.
• Confidence note based on sample size and repeatability.

What a complete AI visibility audit report includes

A premium audit report should be readable by executives and actionable for operators. It needs a summary score, but the real value is the prompt-level evidence and the fix backlog.

The best report makes the next action obvious: which prompt matters, why the brand lost, what to fix, and how to prove whether the fix worked.

• Executive summary: citation rate, mention rate, share of voice, and competitor loss.
• Prompt table: prompt, engine, brand outcome, competitors, cited URLs, and source type.
• Gap diagnosis: access, extraction, entity, source, prompt-fit, or competitor gap.
• Fix backlog: prioritized changes with owner and difficulty.
• Proof plan: re-test schedule and success criteria.
• Limitations: what the audit can and cannot conclude.

How RankEcho helps

RankEcho turns AI visibility auditing into a repeatable operating system. It runs the prompt set, records citation and competitor outcomes, classifies source gaps, generates fix packages, and supports re-tests after changes ship.

The goal is not only to measure visibility. The goal is to find what to fix next and prove whether the work changed the answer.

Frequently asked questions