Branded vs unbranded questions: where AI visibility is won
Unbranded questions, the ones that name a need rather than a company, are where AI visibility is won. The assistant has to decide which brands to name, and a buyer who has not heard of you can only discover you there. Branded questions mostly show whether AI describes you correctly once someone already knows your name.
Both matter, but they answer different questions. Mixing them into one score is the most common way a GEO report flatters a brand. This guide explains the difference, shows what South African search data says about unbranded questions, and gives a practical way to split your question set.
In short
- An unbranded question names a need ("best funeral cover in South Africa"); a branded one names a company ("Kestrel Insurance claims process").
- On unbranded questions the AI chooses the brands. That is where share of voice is won or lost.
- Branded questions test accuracy: products, prices, locations and reputation.
- Weight your question set heavily towards unbranded questions, and report the two groups separately.
- Do not drop questions because keyword tools show no volume. In our South African data, every such search still showed an AI Overview.
What is the difference between branded and unbranded questions?
The test is simple: does the question contain your brand name, or a competitor's? The table uses the site's fictional sample brands.
| Question type | Example | What it tells you |
|---|---|---|
| Unbranded category | "Which insurer has the best car insurance in South Africa?" | Whether you are discovered at all |
| Unbranded problem | "How do I lower my car insurance premium?" | Whether you are part of the research stage |
| Unbranded local | "Best car insurance broker in Durban" | Whether you are found locally |
| Branded | "Is Kestrel Insurance any good?" | Reputation and sentiment |
| Branded facts | "What does Kestrel Insurance cover on a comprehensive policy?" | Whether the facts are right |
| Comparison naming you | "Kestrel Insurance vs Competitor A" | How you are positioned against a rival |
| Competitor-branded | "Is Competitor A reliable?" | Whether you are offered as an alternative |
Why is unbranded search where AI visibility is won?
Because that is where the choice happens. Ask an assistant about Kestrel Insurance by name and the answer will talk about Kestrel Insurance. Ask it which insurer to use and it names a handful of brands and leaves the rest out. Being one of the handful is the whole contest.
South African search data shows how many of these choices are already being made by AI. In our analysis of 96 Google South Africa searches on 11 October 2026, 95 of the 96 named no company or organisation: they were category questions such as "best bank account in south africa" and "best inverter for home south africa". 92 of them showed an AI Overview, including 58 of the 59 list-style searches that began with "best", "cheapest", "most" or "highest".
The answers draw on many sources. The 86 Overviews that listed sources cited 10.9 on average, from 412 different domains in total. Providers' own sites earn a place alongside comparison sites and publishers. In the 8 insurance searches, for example, the comparison site hippo.co.za was cited in 7 of the Overviews and the insurer site outsurance.co.za in 6. On unbranded questions, a brand's own pages can be part of the answer, but only if they answer the question.
What do branded questions tell you?
They show whether AI gets you right. A branded question surfaces outdated prices, discontinued products, wrong branch locations and confusion with similarly named businesses.
Confusion happens even on the simplest branded search. On 11 October 2026, a Google search for our own name, "citedai", ranked a similarly named company (cited-ai.com) first and our site second, and Google's AI Overview cited both. A buyer searching for one could easily read about the other. If that can happen on a search for your exact name, it can happen in an assistant's answer. Our guide on how to stop AI confusing your brand covers the fixes.
So keep branded questions, but treat them as an accuracy audit. Count how often the facts are right, not how often you are named.
How should you split a question set?
Most of the set should be unbranded. For example, a 40-question set for a fictional insurer might look like this:
- 14 unbranded category questions: "best", "cheapest" and "which" questions for each product line.
- 8 unbranded problem questions: the how and why questions buyers ask while researching.
- 6 unbranded local questions: the same needs, with the cities and provinces you serve.
- 6 branded questions: reputation and key facts, such as cover, prices in rand and claims.
- 3 comparison questions that name you against your closest rivals.
- 3 competitor-branded questions, to see whether you are offered as an alternative.
Groups 1 to 3 (28 questions) feed share of voice, presence and first position. Groups 4 to 6 are reported on their own. Keep the set fixed from month to month so changes are real. Our guide to building a GEO question set goes step by step.
Why keep questions that keyword tools say nobody searches?
Because the questions people put to AI are often longer and more specific than the ones keyword tools measure. Google itself says people ask longer and more specific questions, and more follow-up questions, in its AI search experiences.
Our data agrees. Of the 96 South African searches, 14 had no measurable monthly volume in Google Ads, and all 14 showed an AI Overview. The 31 searches with fewer than 100 a month all showed one too, against 82.4% of searches with 1,000 or more. Low volume does not mean low importance: these are often the precise, high-intent questions a buyer asks just before choosing.
How should you report branded and unbranded results?
- Unbranded: presence, first position and AI share of voice, per engine, against named competitors.
- Branded: the share of answers with correct facts, and a list of errors to fix.
- Comparison and competitor-branded: how often you are recommended, and how you are described against the rival.
Never fold branded answers into share of voice. A brand named in every answer to its own name has proved only that the assistant read the question.
Frequently asked questions
What is an unbranded question?
An unbranded question names a need, a product type or a problem but no company, such as 'best car insurance in South Africa'. A branded question names a company, such as 'is Kestrel Insurance any good?'.
Why do unbranded questions matter more for AI visibility?
Because the AI has to choose which brands to name. On a branded question your brand is named almost by definition, so the answer shows little about whether buyers who do not know you yet will hear of you.
Should branded questions be measured at all?
Yes, but separately. Branded questions show whether AI describes you accurately: your products, prices, locations and reputation. Report them as an accuracy check, not as share of voice.
Does a comparison question that names my brand count as branded?
Yes. A question such as 'Meridian Bank vs Competitor A' names your brand, so you will be in the answer. Track comparison questions as their own group: they show how you are positioned against a rival, not whether you are discovered.
Should I drop questions that keyword tools say nobody searches?
No. In our analysis of 96 Google South Africa searches on 11 October 2026, all 14 searches with no measurable Google Ads volume still showed an AI Overview. Keyword tools are a poor guide to the longer questions people put to AI.