Branded prompts inflate your AI visibility score

Every prompt that names your brand makes your AI visibility look better than it is. Ask an assistant "is Acme a good CRM for agencies?" and it will mention Acme, almost every time. Averaged into your visibility number, that answer counts the same as an open question a buyer might ask before they have ever heard of you.

The fix is simple: keep branded prompts to a small share of the set, and report them separately. This piece explains why, shows how large the distortion gets, and sets out the thresholds the RateMyPrompts rubric uses.

Why do branded prompts inflate the number?

Because the prompt contains the answer. A visibility tracker counts how often your brand appears in the answers to your prompts. When the prompt names you, the assistant has to talk about you, so the mention is guaranteed by the question rather than earned in the answer.

That makes branded prompts a different measurement. They test reputation: whether assistants know you exist and describe you correctly. Unbranded prompts test discovery: whether assistants bring you up when a buyer describes a need. Both are worth knowing. Averaging them together produces a number that describes neither.

Tool vendors say the same. Profound's prompt-design guide recommends tracking "far more unbranded than branded prompts", with branded and competitor prompts a small percentage of the list. Rankscale describes branded visibility as the floor and non-branded as the ceiling, and tracks them as separate groups with different thresholds.

How much can a few branded prompts distort the result?

More than most teams expect, because the gap between the two rates is so large. Take an illustrative set of 100 prompts:

The figures are made up, but the proportions are realistic. The dashboard says 28.4%. The share of buyer questions where you actually appear is 2%. Thirty branded prompts made the headline figure fourteen times the discovery rate.

What we found in real prompt panels

We looked at the prompt panels behind 22 brands tracked on the Zebora platform: 8,912 prompts, of which 5,377 were labelled by style, joined to about 28,000 scored answers. These are panels built by marketers to track their brands, not a sample of how people use AI, so they tell us how measurement sets are built and what that does to the numbers.

At first, style looked decisive. "Direct query" prompts returned 4.25 times their brand's average mention rate, and every other style under-performed. Then we checked which prompts named the brand:

The styles with high mention rates were exactly the ones full of branded prompts. The apparent format effect was branded contamination with a format label on it: those styles showed mention rates 4 to 20 times higher than unbranded ones.

With every branded prompt removed, the rates collapsed to between 0.35% and 1.65% depending on style, and their 95% intervals overlapped almost entirely. That is what genuine unbranded visibility looked like for these brands, and it is the number that matters for growth.

What the rubric does with branded prompts

Branded contamination is one of the seven dimensions of the Prompt Fit Score, weighted at 15%, and it caps the buyer-relevance dimension too.

A set that is 40% branded cannot score well however good the rest of it is, because it is not measuring discovery.

Where branded prompts belong

Keep them, in their own segment. A short reputation segment of 5 to 15 prompts is useful for checking that assistants get your pricing, features and positioning right, and for spotting damaging errors. Report it as its own figure, with sentiment and accuracy, and never add it to the discovery number.

Treat prompts naming competitors the same way. "Acme vs Globex" is a legitimate comparison question, but it also primes both brands. Keep competitor-named prompts to around 20% of the set and report them as a comparison segment. For everything else, write unbranded prompts from real buyer language, as covered in how to build a prompt set from buyer language.

What to do

1. Count your branded share. Search the set for your brand name, product names and common misspellings.
2. Split the set into three segments: unbranded, own-brand and competitor-named.
3. Report the unbranded mention rate or share of voice as the headline. Put the branded figure next to it, never inside it.
4. Keep own-brand prompts to 10 to 15% at most, and competitor-named ones to about 20%.
5. Replace the branded prompts you remove with unbranded questions at the same journey stages, so coverage does not fall.
6. Check the result. The free RateMyPrompts scorecard shows your branded share, flags brand-heavy sets, and works out how many points trimming them is worth.