What an AI Discovery Audit Actually Examines

Claims in this piece are observable and methodological — no client or benchmark data used.

Most manufacturers, when they first hear that AI assistants now shape how buyers discover suppliers, ask a reasonable question: fine — so what would you actually look at for us? It’s the right question. “AI Visibility” can sound abstract until you see the specific, concrete things being examined. This article answers that plainly: what an AI Discovery Audit™ actually inspects, what it can tell you, and — just as importantly — what it cannot.

The short version: an audit does not try to make an AI assistant say something flattering about you. It establishes, against realistic buyer questions, what AI assistants can currently find, understand, and accurately convey about your organisation — and where that breaks down. It is a diagnostic, not a promise. What follows is what that diagnosis looks at.

It starts with the buyer’s question, not your company

A weak version of this exercise would ask an assistant “tell me about [Company]” and read the summary. That tests whether the AI can summarise a company it’s been handed — not the situation a real buyer creates. A buyer doesn’t start with your name. They start with a need.

So an audit begins by reconstructing the questions a real buyer in your segment would actually ask — the ones explored in Emerivo’s research on why some manufacturers are shortlisted before the first sales call. For a sterile-injectables CDMO serving regulated markets, that might be “which Indian CDMOs can manufacture sterile injectables for EU and US markets?” — not “tell me about [Company].” The audit works from the buyer’s framing, because that is the situation that actually decides whether you are discovered.

What it examines, across five platforms

An audit runs those buyer questions across the assistants buyers actually use — ChatGPT, Gemini, Perplexity, Claude and Copilot — because they do not behave identically, and the differences between them are themselves informative. Against those responses, it examines a specific set of things:

• Presence — when a buyer asks for a manufacturer with your capabilities, do you appear at all, or only when named directly? Appearing only when already named is very different from being surfaced to a stranger.
• Accuracy — what the assistants say about you: is it correct? Do they attribute the right capabilities, the right regulatory context, the right specialisations — or confuse, outdate, or misstate them?
• Specificity — can they establish your genuine capability in specific terms, or only describe you generically (“a pharmaceutical manufacturer”) in a way that fails to connect you to a buyer’s actual requirement?
• Completeness — which real, commercially important capabilities are missing from what the assistants can establish, even though you genuinely hold them?
• Coherence — do the assistants give a consistent account, or do they contradict themselves and each other, suggesting a fragmented public record?
• Supportability — when asked what public evidence supports their description, what do they point to, and does it actually hold up?

Each of these is an observation, not a score. The audit records what the assistants actually say — verbatim, with the date — because these outputs vary by prompt, platform, and time, and honest diagnosis means capturing what is really there rather than a flattering or alarming summary of it.

It traces findings back to the public record

Observing what AI says is only half of it. The more useful half is examining why. An assistant’s account of you is constructed from what it can find and interpret in your public presence — so where the account is vague, wrong, or missing, an audit looks for the most likely explanation in the public record: a capability buried in a downloadable PDF, a specialisation described only in generic terms, a facility page that contradicts the homepage, an outdated profile still circulating. No one can prove exactly why a given model produced a given answer — the systems are opaque — but you can identify the candidate weaknesses in the public record that plausibly account for it, and those are the things you can actually act on. This is the same mechanism examined in When AI Recommends a CDMO, What Is It Actually Evaluating? — the AI’s description is a readout of your legibility, and the audit’s job is to connect each weakness in the readout to something addressable.

That is what turns an observation (“the assistants can’t establish your HPAPI capability”) into something actionable (“it appears only inside a brochure and isn’t stated in plain text where it can be found”).

What the audit produces

The output is a diagnostic picture, not a promise. Concretely, an audit establishes: where you are currently discoverable and where you are not; where AI assistants describe you accurately and where they are vague, incomplete, or wrong; which genuine capabilities are hard for an outsider to establish; where your public record is inconsistent; and where the specific, addressable gaps lie between what your company genuinely is and what an outside system can currently establish about it.

It is a baseline — a clear account of your current AI Visibility, and a map of the opportunities to improve how accurately and clearly your genuine capability can be established. It is the starting point for improvement, measured against reality.

What an audit does not do — and won’t claim to

Being clear about the limits is part of what makes the diagnosis trustworthy. An AI Discovery Audit does not, and cannot:

• Guarantee that AI assistants will recommend you. No credible organisation can promise that. AI behaviour varies by platform, prompt, and time, and depends on factors outside any supplier’s control.
• Guarantee more enquiries or revenue. It improves how establishable your genuine capability is; it does not purchase a commercial outcome.
• Replace technical evaluation, audits, or qualification. Those remain human, evidence-based processes that happen after discovery.
• Manufacture capability you don’t have. It makes real capability legible. It never invents or inflates — that would fail the only test that matters when a buyer moves to verification.
• Treat AI output as a ranking or a benchmark. It’s a reproducible observation of what can currently be established, not a leaderboard position or a market-share figure.

An audit that promised any of those would be overselling, and the credibility of the whole exercise rests on not doing so.

A version you can run yourself, first

You don’t have to commission anything to see the principle. Take one capability where you have genuine depth, and ask ChatGPT, Gemini, Perplexity, Claude and Copilot a question a buyer in that area would ask. Read the answers against the six things above — presence, accuracy, specificity, completeness, coherence, supportability — and note where each holds up and where it doesn’t. That informal version won’t be as systematic as a full audit, but it will show you, directly, the gap between what your company genuinely is and what an outside system can currently establish. That gap is exactly what a formal audit maps in full. Results vary by platform, prompt and time, so treat it as observation, not proof.

The takeaway

An AI Discovery Audit is, at its core, an honest answer to a question most manufacturers have never been able to ask before: when a buyer researches us through the tools they now actually use, what can they establish — and is it accurate? It examines presence, accuracy, specificity, completeness, coherence, and supportability across the assistants buyers use; it traces every weakness back to the public record where it can be fixed; and it produces a baseline, not a promise. The value is not a flattering report. It is finally being able to see what an outsider sees — and knowing, specifically, what to do about it.

Emerivo is a specialist advisory for pharmaceutical manufacturing and research organisations — TPMs, CMOs, CDMOs and CROs. The AI Discovery Audit™ examines what AI assistants can currently establish about your organisation against realistic buyer questions, and maps the specific, addressable gaps — diagnostic, never a promise of rankings or recommendations. For the wider context, see Why Some Pharmaceutical Manufacturers Are Shortlisted Before the First Sales Call and When AI Recommends a CDMO, What Is It Actually Evaluating?.

Frequently asked questions