An AI search audit is a structured assessment of how often and how accurately a brand is cited across conversational answer engines — ChatGPT, Perplexity, Claude, and Gemini. Run it by firing a representative set of prompts across each engine, recording whether the brand appears, how it is described, and who else is cited instead. The first three fixes that follow almost always involve entity consistency, content structure, and third-party citation gaps.
Key Takeaways
- An AI visibility audit covers four conversational answer engines — ChatGPT, Perplexity, Claude, and Gemini — and measures citation frequency, accuracy, and share of voice against named competitors.
- The audit begins with a prompt set, not a crawl: you need to ask the engines the questions a real buyer would ask, then record what they say and who they cite.
- Entity inconsistency — a brand name, address, or description appearing differently across directories and third-party sources — is the single most common reason a brand is absent or mis-described in AI answers.
- Answer-first content structure is required for AI extraction: a page that buries its key claim in paragraph three will rarely be lifted verbatim by an answer engine, even if it ranks well in traditional search.
- Third-party citations (mentions in credible external sources) carry more weight in generative engine optimisation than on-site content alone; an audit that only checks your own pages misses the most important signal.
- A single audit is a snapshot; the pattern that matters — citation drift, competitor displacement, share-of-voice shifts — only becomes visible through repeated sampling over time.
Why an AI Search Audit Is Different From an SEO Audit
An AI search audit is not a crawl of your own website — it is an interrogation of what AI engines believe about your brand, based on everything they have ingested. Traditional SEO audits check your pages for technical issues: broken links, missing meta tags, slow load times. An AI visibility audit checks something harder to see: whether the engines that now answer your customers’ questions have an accurate, confident, and favourable picture of who you are. Those are different problems with different fixes.
This distinction matters practically. A brand can have a technically clean website and still be invisible in AI answers — because the engines are drawing on third-party sources, aggregators, review platforms, and the broader web of mentions, not just the brand’s own pages. Understanding what generative engine optimisation is — building off-site authority and citation footprint so AI systems recommend and cite a brand — is the conceptual foundation before you run a single prompt.
Step 1: Build a Prompt Set That Reflects Real Buyer Intent
The prompt set is the core instrument of the audit, and most first attempts are too narrow. Start with three prompt categories: category-level queries (“best [service type] in [city]”), problem-aware queries (“how do I choose a [service type]”), and brand-direct queries (“tell me about [brand name]”). Aim for 15–25 prompts total — enough to surface patterns without making the audit unmanageable.
Write prompts the way a buyer actually speaks, not the way a keyword tool formats them. “What’s the best accountancy firm for a small e-commerce business in Manchester” will produce a different answer than “accountancy firm Manchester” — and the conversational form is what your customers are actually typing. Include at least two or three prompts that name your closest competitors directly, because those answers will tell you exactly how the engines position the competitive landscape.
Step 2: Run Each Prompt Across All Four Engines and Record What You Find
Run every prompt across ChatGPT, Perplexity, Claude, and Gemini — not just one. The engines draw on different training data, use different retrieval mechanisms, and produce meaningfully different answers for the same prompt. A brand that appears confidently in Perplexity may be absent in Claude, or described differently in Gemini. Treating one engine as representative of all four is one of the most common mistakes agencies make when they run a quick manual check and call it an audit.
For each prompt and each engine, record: (1) whether the brand is cited at all; (2) how it is described — name, category, location, key claims; (3) which competitors are cited alongside or instead; (4) whether any description is inaccurate or outdated. A simple spreadsheet with these four columns per engine is enough for a first audit. What you are building is a citation map — a picture of where the brand exists, where it is absent, and where it is being displaced.
This is genuinely time-consuming to do manually across four engines and 20 prompts. The real difficulty is not the individual query — it is the cross-engine, cross-prompt pattern recognition, and the fact that AI answers are non-deterministic: the same prompt can return a different answer on consecutive runs. What looks like a clean result on one pass may be an outlier. That is exactly why AI citation and share-of-voice tracking exists as a discipline rather than a one-time exercise — the pattern only becomes reliable through repeated sampling.
Step 3: Interpret the Results — What the Patterns Actually Mean
Three patterns account for the majority of what a first audit surfaces, and each points to a different fix.
Pattern 1 — The brand is absent across most prompts. This almost always means one of two things: either the brand has insufficient third-party citation coverage (not enough credible external sources mention it in a way the engines can find and trust), or the brand’s entity is poorly defined — the engines do not have a confident, consistent picture of what it is. Check whether the brand name appears consistently across directories, review platforms, and industry publications. Inconsistency in name, address, or category description fragments the entity signal and suppresses citation.
Pattern 2 — The brand appears but is described inaccurately or incompletely. This is an entity authority problem. The engines are pulling from sources that carry outdated or partial information. The fix involves updating the authoritative sources the engines are most likely drawing on — Google Business Profile, Wikidata, industry directories — and ensuring the brand’s own pages describe it clearly and consistently, with structured data (FAQPage markup, Organisation schema) that makes the description machine-readable, as documented in the Google Search Central structured data guidance.
Pattern 3 — Competitors are cited consistently where the brand is not. This is a share-of-voice gap, and it is the most commercially significant finding. It means the engines have enough signal about competitors to recommend them confidently, but not enough about the brand. The fix is a combination of third-party citation building (earning mentions in the sources the engines trust) and content restructuring so the brand’s own pages are extractable.
The First Three Fixes: A Prioritised Remediation Order
After running the audit, most brands face more issues than they can address at once. Prioritise in this order, because each layer depends on the one before it.
Fix 1 — Resolve entity inconsistency. Before anything else, make the brand’s name, description, category, and location consistent across every source the engines are likely to index: your own site, Google Business Profile, Wikidata, major directories, and review platforms. A brand that appears as “Hartley & Co”, “Hartley and Co Ltd”, and “Hartley Co” across different sources is, from an engine’s perspective, potentially three different entities. Consolidate to one canonical form and propagate it.
Fix 2 — Restructure key pages for answer extraction. This is where what answer engine optimisation is becomes practical. Answer engines extract content that is structured to be extracted: answer-first paragraphs, clear headings that match the question being asked, and concise definitions that can be lifted verbatim. A page that buries its key claim three paragraphs in will rarely be cited, even if it is technically authoritative.
Here is what that looks like in practice. Take a hypothetical agency client, Hartley & Co, a commercial cleaning firm. Their “About” page currently reads:
Before: “Established in 2009, Hartley & Co has grown from a small family operation into one of the region’s most trusted providers of cleaning services for offices, retail spaces, and industrial facilities. Our team of over 80 trained operatives brings a commitment to quality that has earned us long-term relationships with clients across the North West.”
That paragraph is competent marketing copy. An answer engine asked “who are the best commercial cleaning companies in the North West” will not lift it, because the key claim — what the company does, where, and why it is credible — is distributed across the whole paragraph rather than stated in the first sentence.
After: “Hartley & Co is a commercial cleaning company serving offices, retail spaces, and industrial facilities across the North West of England, with over 80 trained operatives and more than 15 years of operation.”
The improvement is specific and nameable: the after version is answer-first and self-contained — an engine can lift that single sentence verbatim and it stands alone as a complete, accurate description. The before version requires the reader to assemble the answer from across the paragraph; the after version hands it over in the first sentence.
Fix 3 — Build third-party citation coverage in the gaps the audit identified. Once entity and content are in order, the highest-leverage remaining work is earning mentions in the external sources the engines draw on: industry publications, trade directories, review aggregators, and credible press coverage. This is the off-site authority work at the core of generative engine optimisation services — and it is the fix that takes the longest, which is exactly why it should start immediately after the entity and content work is done, not after.
Common Misconceptions About AI Search Audits
Myth: Running the audit once is enough. The mechanism that sustains this belief is the analogy to a technical SEO audit, which does produce a stable list of fixable issues. AI answers are not stable — they shift as training data updates, as competitors earn new citations, and as the engines’ retrieval mechanisms evolve. A single audit is a useful baseline, but the pattern that actually drives decisions is the trend over time. Agencies that run a one-time audit and report it as ongoing monitoring are selling a snapshot as a service.
Myth: If a brand ranks well in Google, it will appear in AI answers. The mechanism here is legacy tooling and institutional habit — most agencies built their reporting around blue-link rankings, and it is easier to extend that report than to build a new one. But AI engines do not simply mirror search rankings. They weight third-party citations, entity clarity, and content extractability in ways that do not map cleanly onto traditional ranking signals. A brand can hold the top organic position and still be absent from AI answers for the same query.
Myth: Schema markup alone will fix AI visibility. Schema markup helps engines parse and extract structured content — the schema.org FAQPage specification is a genuine signal — but it is necessary, not sufficient. An engine that does not have enough third-party evidence to trust a brand will not cite it confidently regardless of how clean the markup is. Schema is the last-mile extraction layer, not the foundation.
What to Do With the Results if You Are an Agency
For agencies managing AI visibility across multiple clients, the audit output is both a diagnostic and a client communication tool. The citation map — which prompts cite the client, which cite competitors, and what the engines say — is the most concrete evidence available that AI search visibility is a real, measurable problem worth solving. It is also the baseline against which future progress is measured, which is what makes it defensible as a retainer deliverable rather than a one-time project.
The practical challenge at scale is that running 20 prompts across four engines for ten clients manually is not a sustainable workflow. The pattern recognition across engines, the non-determinism of AI answers, and the need for repeated sampling over time are the reasons dedicated tooling exists for this work. If you are building out an AI visibility practice, the audit methodology above is the right starting point — but the operational reality of running it across a client roster is what separates a pilot from a practice.
The most useful thing you can do right now is run the audit on one client — ideally one where you already suspect a gap — and let the results shape the conversation. Start with a free AI visibility audit to see where a brand currently stands across ChatGPT, Perplexity, Claude, and Gemini before you build the remediation plan.
Frequently Asked Questions
What is an AI search audit?
An AI search audit is a structured assessment of how often and how accurately a brand is cited across conversational answer engines — typically ChatGPT, Perplexity, Claude, and Gemini. It involves firing representative buyer-intent prompts across each engine and recording citation frequency, description accuracy, and share of voice against competitors.
How many prompts do I need for a first AI visibility audit?
A first AI visibility audit typically requires 15–25 prompts covering three categories: category-level queries (“best [service] in [location]”), problem-aware queries (“how do I choose a [service]”), and brand-direct queries (“tell me about [brand name]”). Include prompts that name key competitors to reveal how engines frame the competitive landscape.
Why do I need to run prompts across all four engines, not just one?
ChatGPT, Perplexity, Claude, and Gemini draw on different training data and retrieval mechanisms, producing meaningfully different answers for the same prompt. A brand cited confidently in Perplexity may be absent or mis-described in Claude. Treating one engine as representative of all four produces an incomplete and often misleading picture of AI search visibility.
What are the most common reasons a brand is absent from AI answers?
The two most common causes are insufficient third-party citation coverage — not enough credible external sources mention the brand — and entity inconsistency, where the brand name, category, or description appears differently across directories and platforms. Both fragment the signal AI engines need to cite a brand confidently.
How is an AI search audit different from a traditional SEO audit?
A traditional SEO audit crawls a brand’s own website for technical issues. An AI search audit interrogates what answer engines believe about a brand based on third-party sources, entity signals, and content extractability — factors that do not map cleanly onto blue-link ranking signals. A brand can rank first in Google and still be absent from AI answers.
How often should an AI visibility audit be repeated?
AI answers are non-deterministic and shift as training data updates, competitors earn new citations, and engine retrieval mechanisms evolve. A single audit establishes a baseline; meaningful pattern recognition — citation drift, competitor displacement, share-of-voice shifts — requires repeated sampling over time, typically monthly for active remediation programmes.