Short answer, as of October 2026: branded prompts ("is Acme Plumbing any good?") measure whether AI assistants describe your business accurately, and unbranded prompts ("who should I call for a burst pipe in Leeds?") measure whether they recommend you at all — two different questions that need separate prompt lists, separate targets and separate lines in a report. Combine them and you get one comfortable number that hides the problem you most need to see.
The distinction sounds obvious. In practice most first prompt sets get it wrong in one of two ways. Some are nearly all branded, because those are the questions the business owner thinks to ask, and the tracker reports a visibility rate near 100 per cent. Others ignore branded prompts entirely, and nobody notices for months that an assistant has been quoting a price from two years ago.
What each type measures
Branded prompts: accuracy
When a prompt contains your name, the assistant is not deciding whether to mention you. You have already been mentioned. What is being tested is the content of the answer:
- Does it describe what you do correctly?
- Are the prices, hours, locations and service area right?
- Does it attribute a competitor's product, or a different company with a similar name, to you?
- Is the tone neutral, positive or cautious, and which sources is that tone drawn from?
- Does it say it has no information?
The last one matters more than it looks. "I don't have reliable information about that business" in reply to a branded prompt means the assistant found too little to work with. That is an entity problem, and the fix is in entity consistency and the AI knowledge graph.
Unbranded prompts: discovery
When a prompt describes a need without naming anyone, the assistant chooses which businesses to bring up. Whether you are among them is the central question of AI visibility, because it is the only path by which someone who has never heard of you finds you through an assistant.
Unbranded results are far noisier. The same question asked five times can produce five different shortlists. That is why they need repeated runs and why they should consume most of a prompt budget.
Comparison prompts: positioning
A third group sits between the two: prompts that name you and a competitor. "Acme or Borealis for a small office?" Your name is supplied, so it is technically branded, but the answer reveals something neither of the other types does — which attributes the assistant associates with each of you, and who it leans toward for which kind of buyer. Track these as their own short list.
How the three compare
| Branded | Unbranded | Comparison | |
|---|---|---|---|
| Example | "What does Acme Plumbing charge for a call-out?" | "How much does an emergency plumber cost in Leeds?" | "Acme Plumbing or Borealis for a boiler repair?" |
| Question it answers | Is what they say about us correct? | Do they bring us up? | How are we positioned against a rival? |
| Typical mention rate | Near 100% by construction | Anywhere from zero to most runs | Near 100% by construction |
| What to measure | Factual accuracy, sentiment, sources | Mention rate across runs, share of voice | Which brand is favoured, and on what grounds |
| Variability | Low | High | Moderate |
| Runs needed | One or two per period | Three to five per period | Two or three per period |
| Fix when it is bad | Correct the source facts | Publish the clearest answer; earn third-party corroboration | Publish an honest comparison of your own |
Why you must not add them together
Suppose a set of twenty prompts: ten branded, ten unbranded. The branded ten name you every time. The unbranded ten name you in two. A tracker that reports one combined figure says 60 per cent. The business reads that as "we are doing reasonably well." The truth is that people who do not already know the name are told about the business one time in five.
Now suppose the unbranded figure doubles to four in ten. The combined number moves from 60 to 70. A doubling of the thing you care about shows up as a modest uptick, and it is easy to miss.
Separate lines fix both problems. Report branded results as an accuracy check — "seven of eight answers correct; one quotes the old address" — and unbranded results as a mention rate by group. How to lay that out on a page is covered in the monthly AI visibility report.
Building the branded list
Keep it short and fixed. Five to eight prompts cover most businesses:
- What is it? "What does [name] do?"
- Is it trustworthy? "Is [name] legit?" or "[name] reviews."
- What does it cost? "How much does [name] charge for [main service]?"
- Where does it operate? "Does [name] serve [area]?" or "Where is [name] located?"
- How to reach it? "What are [name]'s opening hours?"
- Alternatives. "What are the alternatives to [name]?" — this shows who assistants consider your peers.
Run them monthly and read every answer. This is the one part of tracking where reading beats counting. A wrong fact found here is usually traceable to a specific source within a few minutes, and the repair process is in when AI answers are wrong about your business.
Two situations call for a longer branded list. After a rebrand, track the old name and the new one side by side until assistants stop answering under the old one, as described in AI visibility through a rebrand. And if your name is shared with another business or is a common word, add prompts that include your city or category to see whether assistants can tell you apart.
Building the unbranded list
This is where the thought goes. Write prompts the way a customer who has never heard of you would phrase the need, and cover several intents rather than twenty variations of one:
- Recommendation — "best [category] for [type of customer]"
- Cost — "how much does [service] cost"
- Problem — "[symptom], what should I do"
- How-to that leads to a provider — "can I [task] myself or do I need a professional"
- Local — "[service] near [place]" where location matters
Use real phrasings where you can. Sales call notes, support tickets, the questions on your enquiry forms and Search Console queries are better sources than a brainstorm. The full method, including how to size the set, is in how to build a prompt set.
One caution: do not write unbranded prompts that only you could answer. "Which plumber offers a 12-month workmanship guarantee and free boiler checks in LS6?" will name you, and it will also tell you nothing. A prompt is useful when a real stranger might type it.
Reading the results together
The two lists inform each other.
Branded accurate, unbranded absent. Assistants know who you are and do not think of you. This is the commonest pattern for an established business. The work is on category-level content and third-party mentions, not on facts.
Branded inaccurate, unbranded present. You are being recommended with wrong details. Urgent, because the recommendation is sending people to the wrong price or the wrong address. Fix the facts first.
Branded "no information", unbranded absent. You are effectively invisible. Start with the basics covered in GEO for a new site with no authority.
Both healthy. Move budget toward comparison prompts and toward unbranded prompts in adjacent categories you want to grow into.
Budgeting the two types
Trackers meter on prompt-checks: one prompt, in one engine, one time. Branded prompts are stable, so one or two runs a month is plenty. Unbranded prompts need three to five runs each to be readable. A sensible split puts about three quarters of the allowance on unbranded prompts.
As a worked example on DigiRank Expert's Starter plan, which includes 200 prompt-checks a month across six engines: six branded prompts run once (36 checks), plus nine unbranded prompts run three times (162 checks), uses 198. That is a complete small-business set. The Agency plan's 2,500 checks support the same structure across several clients. Plans are on the pricing page, and the tracker's Opportunities tab, described on the features page, lists the unbranded prompts where you are missing so the next piece of work is already chosen.
Frequently asked questions
What is the difference between a branded and an unbranded prompt? A branded prompt includes your business name, such as asking what a company does or whether it is any good. An unbranded prompt describes a need without naming anyone, such as asking who to hire or which tool to use. The first tests what assistants say about you. The second tests whether they bring you up at all.
Should branded and unbranded prompts be reported together? No. Branded prompts name you almost every time by design, so adding them to unbranded results inflates the rate and hides the discovery problem. Report them as two separate lines with different goals: accuracy for branded, mention rate for unbranded.
How many branded prompts do I need? A small fixed set is enough for most businesses: roughly five to eight covering what you do, pricing, location or coverage, reviews, and comparisons with one or two named competitors. Their purpose is to catch wrong facts, and a handful of questions run monthly does that.
Why does an assistant describe my business wrongly when asked by name? Usually because the sources it draws on disagree or are out of date: an old directory listing, a previous address, retired pricing, or a page on your own site that was never updated. Fix the source, then keep the branded prompt in your set to confirm the answer changes.
Do unbranded prompts matter if people already know my brand? Yes. Existing customers ask by name. New customers describe a problem. Unbranded prompts are how you find out whether assistants introduce you to people who have never heard of you, which is where growth comes from.
What is a comparison prompt, and which type is it? A comparison prompt names you and a competitor together, such as asking which of two businesses is better for a given need. It is branded, because your name is supplied, but it is worth tracking separately since it shows how assistants position you against a specific rival.
How should I split a limited prompt budget between the two types? Put most of it into unbranded prompts, because they vary more and need repeated runs to read. A common split is about three quarters unbranded and one quarter branded and comparison. Branded prompts are stable and can be run less often.