The subject is an anonymised composite of engagements in one category, not a single named client. The competitor names are redacted. The failure modes, the source concentration and the shape of the roadmap are what these audits actually surface. The individual figures should be read as illustrative of the method, not as a verifiable result for a specific company. We would rather say that plainly than let you assume otherwise.
Six of the forty-two prompts tested, chosen because each shows a different failure. Note the follow-up turn: single-prompt testing would have missed the most important finding in the audit.
“What are the best API observability tools for a mid-sized engineering team?”
Five products named. The subject was not among them. Three of the four cited sources were third-party; the only vendor domain cited belonged to a competitor whose documentation answered the question directly.
“Which observability tools work well with OpenTelemetry without vendor lock-in?”
Subject named fourth of five, described accurately but briefly. No link. The description came from a Reddit thread rather than from the subject's own site, which had a stronger page on the same topic that was not retrieved.
“[Subject] vs [Competitor A]: which is better for a Series B startup?”
Balanced comparison, subject's own pricing page cited. One material error: a pricing tier that was discontinued nine months earlier was presented as current, sourced from a G2 listing that had never been updated.
“What does [Subject] cost?”
Correct on the current published tiers, then repeated the discontinued tier from the same stale G2 listing as a third option. Brand accuracy failure, caused by a third-party source the subject controls the content of.
“Follow-up: which of those would you pick for a team of 30 engineers?”
Subject dropped out entirely at the narrowing turn. The retrieved set moved to team-size-specific content, which the subject had none of. This is the most common failure we see and it is invisible to single-prompt testing.
“Which API observability tools have the best onboarding for small teams?”
A YouTube walkthrough by an independent practitioner was cited second. The subject had no third-party video coverage at all, on a question their product genuinely wins.
Every metric below is defined in the published measurement standard, with its formula and what it excludes. Citation rate and brand mention rate are reported separately, never blended.
Declared and frozen before measurement: 12 category discovery, 14 comparison, 10 qualification, 6 follow-up turns. Each run 5 times across the window.
40 of 210 answers linked a page on the subject's domain. Concentrated almost entirely in branded comparison and pricing prompts.
71 of 210 answers named the subject. Reported separately from citation rate, never blended, because blending them would nearly double the headline number.
Against a competitor set of six, declared before measurement. The category leader took 31%.
20 of 71 answers naming the subject contained a material error. 17 of those 20 traced to one stale G2 listing.
Of all citations across the set, the subject's own domain accounted for 9%. The rest sat on sources the subject did not own and was not working on.
| Platform | Citation rate | What explains it |
|---|---|---|
| ChatGPT | 16% | Heaviest reliance on Reddit of any platform tested. Subject's absence from community threads cost most here. |
| Gemini | 24% | Best performance, tracking the subject's existing Google rankings. Mention rate far exceeded citation rate, as expected on this surface. |
| Claude | 27% | Strong where documentation answered the question. The subject's docs are public and good, which is why this is the best citation rate in the set. |
| Perplexity | 14% | Visible source lists made the concentration on Reddit and G2 directly observable, which is what directed the roadmap below. |
| Copilot | 9% | Traced to weak Bing indexation rather than anything AI-specific. 40% of the subject's pages were not in the Bing index at all. |
Ordered by damage done, not by effort. Note that the top two items are days and weeks of unglamorous work, and neither is content production.
One out-of-date third-party listing caused 17 of the 20 material accuracy failures. The subject controls this content and had not looked at it in two years.
40% of pages missing from the index Copilot grounds on. Unglamorous, cheap, and the entire Copilot problem.
Complete drop-out at the narrowing turn. The retrieved set moves to content the subject does not have, on questions their product wins.
Reddit accounted for 22% of all citations. Transparent factual correction first, then genuine practitioner participation.
A competitor's YouTube walkthrough is cited on a question the subject wins on the merits. There is no subject-related video in the retrieved set at all.
Illustrative composite. These are engagement patterns described by category, not named clients, and the figures show the shape of an outcome rather than a single verifiable result. Read them as an indication of what the work produces, not as a claim you could check. Where we publish a number about the firm itself, it carries its sample size, window and method on the page.
The free AI Visibility Audit produces this, on your category, against your competitor set, scored against the same published standard.