The MentionAudit Methodology

Every audit runs the same buyer-intent prompt set across the AI engines that answer category questions for your buyers:

• ChatGPT
• Claude
• Google Gemini
• Perplexity
• Grok
• Microsoft Copilot

We also capture Google's AI surfaces — AI Overviews, and AI Mode where the locale is supported — through direct SERP capture at audit time.

Every engine gets the same question set, the same region, and the same prompt version, so the results are comparable to each other and to your last audit. We capture classical search context alongside the AI answers, because AI engines lean on classical search signals: a prompt where your domain is invisible on the first results page is a prompt where an AI Overview is unlikely to cite you, regardless of any other GEO work.

For every engine, on every prompt, the audit record holds:

• The full rendered answer text, byte-for-byte. No paraphrasing, no summarising, no cleanup.
• The full citation list — every URL the engine surfaced, in the order it surfaced them, with whatever per-citation metadata the engine exposed.
• Response metadata — the model identifier where the engine exposes one, response timing, and the region and prompt version the run was pinned to.
• The reason an engine returned nothing, when that happens: an empty answer, a refusal, an unsupported locale, or a transient upstream error. A missing answer is recorded as a miss with a reason, never dropped from the average.
• The exact passage where each brand mention occurs, so any sentiment claim can be read back to the sentence that produced it.

This verbatim layer is the foundation of every customer-facing claim downstream. If the audit says "ChatGPT cited you on this prompt," you can open the exact response and check. There is no opaque step between engine output and what you see.

We also do not re-run a prompt until the answer looks better. The first usable response is the response we score.

The methodology principle is: capture verbatim, abstract sparingly, and always link the abstraction back to the capture.

• Verbatim: every engine answer, every cited URL, every reason an engine returned nothing, every brand-mention passage, every classical search snapshot.
• Lightly abstracted, with verbatim links: whether your brand was present in an answer, whether a cited URL is yours, a tracked competitor's, or a third party's, and the sentiment of each mention.
• Presented as conclusions: recommendations, competitor profiles, and the audit's overall scoring.

The abstraction layers are what make the audit usable. The verbatim layer is what makes it auditable. You can always drill from any summary metric down to the engine output that produced it.

Scoring follows from the same principle. Citations count for more than mentions, because attribution is the durable visibility signal. Engines that skip your brand count as misses rather than quietly dropping out of the average. And a recommendation only ships if it traces back to captured answer text or source evidence.

To set expectations:

• We do not edit, paraphrase, or "clean up" engine output. What the engine said is what we captured; if it contained errors, those errors are visible in the audit.
• We do not estimate inclusion in engines we cannot directly query. If an engine is regional-restricted or behind a closed beta we cannot reach, we mark its data as unavailable for that prompt rather than guess.
• We do not blend inputs from disagreeing engines into a single number without flagging the disagreement. Disagreement is signal; collapsing it is data loss.
• We do not recommend actions outside the audit's evidence base. Every recommendation links to the engine and prompt evidence that motivated it.
• We do not pad the recommendation list. Brand teams read from the top — a list of fifty recommendations gets the top five implemented and the rest ignored. We aim for fewer, higher-conviction actions grounded in specific evidence.

Every audit is stamped with the methodology version in force when it ran. When we change what the audit measures, the version increments, and audits run under an earlier version stay readable — they are simply versioned to the methodology that captured them.

This is operational hygiene that matters when the same prompt is re-audited months apart. A drop in a brand's score from one audit to the next can be a real visibility loss, or it can be a methodology refinement. Versioning lets you tell the difference.

The most useful single move when reading an audit is to drill from the summary view straight to the verbatim engine output for two or three high-stakes prompts. The summary numbers tell you whether you are winning or losing the prompt; the verbatim output tells you why. Almost every actionable insight surfaces from reading the actual engine responses with brand and competitor names highlighted.

Sentiment is the signal brand teams find most surprising. Your brand can be mentioned in ninety percent of AI answers on a category and the mentions can be neutral-tilting-negative, in which case the visibility win is hollow. Because we keep the exact sentence the engine produced, that claim is checkable rather than asserted.

All of it exists so the data you read in the audit UI is data you can defend to a CMO, a CEO, or a board. That is the methodology: intentionally boring in what it promises, and intentionally rich in what it keeps.