The short version
- A branded prompt (one that names you in the question) makes the model repeat your name almost every time. It returns near-100% visibility by construction, so it measures nothing.
- Mixed into a tracked set, branded prompts inflate your headline visibility by 15 to 20 points and hide the unbranded, buyer-intent prompts where a mention has to be earned.
- AI answers are also non-deterministic. Presence should be a rate with a sample size and a confidence interval over a window, not a yes/no from one scan.
- Branded prompts are a poor visibility metric but a great sentiment and accuracy signal. llemmy keeps them out of the headline score and routes them into a sentiment view tied back to cited sources.
If you use any AI visibility tool to monitor branded prompts, there is a good chance your reporting is quietly broken. When you ask an AI engine a question that already contains your brand name, your brand shows up in the answer almost 100% of the time. You are not measuring visibility. You are measuring whether the model can repeat a word you handed it.
At llemmy we made a deliberate call to handle this differently. Here is the thinking, and it applies no matter which tool you use.
The headline llemmy builds is made of unbranded, buyer-intent prompts, each number carrying its sample size and 95% confidence interval. Branded prompts never touch this score.
What counts as a branded prompt
Anything that names you in the question. "What is [brand]?", "Is [brand] any good?", "[brand] vs [competitor]", "[brand] pricing." The model already has the answer in the prompt, so of course it mentions you. A branded prompt that returns 100% visibility is not a finding. It is a tautology.
Prompts · mention rate
Sample tracked set, rolling 30-day window
In llemmy, branded prompts (greyed) never enter the headline visibility score. The number you act on is built from earned, buyer-intent prompts.
Why this quietly breaks your reporting
It inflates your headline. Mix a handful of branded prompts into your tracked set and your visibility number jumps 15 to 20 points for no real reason. The dashboard looks great. Nothing changed in the real world.
It hides the prompts that matter. The questions that actually decide whether you win are the unbranded ones: "best [category] tool", "alternatives to [competitor]", "who should I use for [job]". That is where the model picks winners and you have to earn the mention.
It turns agency reporting into a vanity exercise. Showing a client "you are 98% visible in AI" off branded prompts is a number that can only go down. It sets a false baseline and buries the exact gap you are being paid to close.
A second, separate problem: non-determinism
Even on a good unbranded prompt, a single scan is a coin flip. Run the same prompt three times and you can get three different answers. So "did we appear" should never be a yes/no from one run. It should be a rate with a sample size and a confidence interval, measured over a rolling window, so one noisy day does not move your headline or fire a false alarm. If a tool shows you a bare "appeared / did not appear", be skeptical.
So are branded prompts useless? No
You are just using them for the wrong thing. Branded prompts are a bad visibility metric but a great reputation and accuracy signal. They answer two questions presence tracking cannot.
Sentiment: how does the AI describe you when it does talk about you? Positive or negative, the actual language, and whether it is improving or sliding. Real sentiment means judging the meaning of the answer, not counting keywords, so "not the cheapest, but worth it" reads as positive, and a genuine compliment with no trigger words still counts.
Brand sentiment · how AI talks about you
142 branded answers analyzed · last 30 days
Judged on meaning, not trigger words. A backhanded compliment reads positive; a complaint with no negative keywords still counts.
Accuracy: does the AI actually know you correctly, or is it confidently wrong about your pricing, your features, or who you serve? Branded prompts are where you catch the model hallucinating about your own brand before a buyer does.
And the part most tools skip: tie that back to the sources the AI cited to form the opinion. If an engine keeps calling you "overpriced", you want the URL that taught it that.
Sources shaping the sentiment
Cited by the engines when they describe your brand
When an engine calls you "overpriced", trace it to the page that taught it that, then go fix the source.
What to actually do (works in any tool)
- Classify prompt intent and keep branded prompts out of your visibility KPI. Build the headline on unbranded, buyer-intent prompts where the mention has to be earned.
- Report presence as a rate with a sample size and a confidence interval over a window, not a single run.
- Point your branded prompts at sentiment and factual accuracy, and trace both back to cited sources.
- Watch the gap, not the vanity number: the queries where you rank on Google but get left out of the AI answer.
How llemmy handles it
We made the deliberate call to exclude branded prompts from the headline visibility score and route them into a sentiment and accuracy view instead, because counting them was dishonest math. Branded prompts still run, but they feed how AI describes you and whether it gets you right, traced back to the sources behind the opinion.
None of this requires our tool. If you take one thing from this post, open your current dashboard and check whether branded prompts are padding your visibility number. If they are, your reporting is telling you the story you want to hear instead of the one you need.
Want to see the honest version of your own numbers? Run a free GEO audit, or start free and watch llemmy split branded from unbranded automatically.
FAQ
What is a branded prompt in AI visibility tracking?
A branded prompt is any tracked question that names your brand, like "what is [brand]", "is [brand] any good", "[brand] vs [competitor]" or "[brand] pricing". Because the brand name is already in the question, the model repeats it in the answer almost every time, so these prompts return near-100% visibility by construction.
Why are branded prompts a bad AI visibility metric?
They return a near-100% mention rate no matter how the market actually sees you, so they inflate your headline number and hide the unbranded, buyer-intent prompts where a mention has to be earned. Mixing them into a tracked set can lift a visibility score by 15 to 20 points for no real-world reason.
What should branded prompts be used for instead?
Treat them as a reputation and accuracy signal, not a visibility metric. Use them to measure how an AI describes you, whether it states your pricing, features and audience correctly, and which cited sources shaped that opinion. Keep them out of the headline visibility KPI.
How does llemmy handle branded prompts?
llemmy classifies prompt intent and excludes branded prompts from the headline visibility score, building the headline on unbranded, buyer-intent prompts. Branded prompts are routed into a sentiment and accuracy view that judges the meaning of each answer and traces it back to the sources the AI cited.
Why is a single AI visibility scan unreliable?
AI answers are non-deterministic, so running the same prompt several times can return different answers. Presence should be reported as a rate with a sample size and a confidence interval over a rolling window, not a yes/no from one run.
By the llemmy team, June 2026. Related reading: The State of AI Search & GEO in 2026 and how llemmy compares to other GEO tools.