← Back to home

Sample report: our own brand, unedited

This is the real output of one weekly run on brandscope.dev — 2026-11-02, five core answer surfaces plus DeepSeek. We publish our own numbers, including the bad ones, because a vendor that only shows a polished sample tells you nothing about how it behaves on a hard week.

Client reports look the same, except the brand is theirs. Names of other clients are never published.

Unbranded discovery

0%

0 of 27 answers

Branded mention rate

94%

17 of 18 answers

Answers judged accurate

38%

8 of 21 judged

AI Visibility Score

41

Discovery-led, so at its floor

What this run says

Three findings, in the order they matter — and one of them is the reason our own score is 41 rather than 90-something.

1. Buyers who don't know us never see us

Of 27 answers to questions that never contained the word “BrandScope” — “best AI mention monitoring tool for agencies”, “top AI visibility tools in 2026” — AI named us zero times. Not once, across ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews.

This is the honest state of a young brand in a crowded category. It is also the number we tell clients to optimise, because it is the one that predicts revenue.

2. When someone names us, the answer is good

17 of 18 branded answers mentioned us, and every one of those mentions was in the first position. The model knows what we do and recommends us when asked directly. That is recall, not discovery — and blending the two into one number is exactly how tools report a healthy 87% while discovery is zero.

3. Most descriptions are only partly right

Of 21 judged mentions, 8 were fully accurate (38%), the rest partial or misleading. The recurring error: describing us as a generic “AI mention tracker” while omitting the citation verification — our actual differentiator. That is a positioning problem on our side, not a model problem, and it is fixable with a clearer definition sentence on the site.

Citation ownership

Every cited URL is classified by who owns it. The distinction matters: a mention that cites your own domain, a mention that cites someone else writing about you, and a mention that cites a different company with the same name are three different problems — and only one of them is your win. On our own brand we have caught the third case, where an answer about “BrandScope” was citing a same-name competitor's pricing page.

Cited domainCountClassification
brandscope.dev5Own — you are the only thing feeding the answer
g2.com1Third party — someone else is explaining you
otterly.ai · peec.ai2Third party — competitor pages being cited in the same answer
deepsmith.ai1Third party — needs checking whether it is about us

Only answers that mentioned us carry a citation, so this table is a subset of the 17 branded mentions — not the full answer set. The competitor domains are what makes the “top tools” questions hard: the answer is assembled from those pages, and our domain is not among them.

Per-engine breakdown

The engines fail differently, which is why we report per surface rather than one average. This run is unusually uniform — and that uniformity is itself the finding:

EngineUnbrandedBranded
ChatGPT0 / 64 / 4
Claude0 / 64 / 4
Gemini0 / 64 / 4
Perplexity0 / 64 / 4
Google AI Overviews0 / 31 / 2
DeepSeek (not scored)0 / 64 / 4

0 of 6 on every single engine is not a sampling artefact — it is the same answer everywhere. When the model does not know a brand exists in a category, all five surfaces agree about it. Compare that to the branded column, where four out of four is also consistent. Consistency in both directions is what tells you this is a real measurement rather than noise.

What we would do next (and what we cannot promise)

The fixes below come from this run's data. They are industry practice, not proven causal levers — nobody in this category, including us, has published a stable cross-platform result showing a specific action produced a specific lift.

  1. Get into the list-shaped answers. “Top tools in 2026” and “best X for Y” are assembled from third-party roundups and directories. Our own pages barely enter them.
  2. Fix the definition sentence. 13 of 21 descriptions were partly wrong; a sharper one-line definition is the cheapest fix available.
  3. Keep the branded strength but stop trusting it. 94% branded looks great and predicts nothing about growth.

Want to see this for your own brand? The free check runs one branded question in about 30 seconds — it's on the homepage. The full report (everything above, for your brand) is the $49 one-off audit or a $39/mo subscription.

Run 2276211f, finished 2026-11-02. Method and known limits: how it works. Last updated 17 November 2026.