A team runs two AI tests five minutes apart. In the first, ChatGPT describes the company accurately when given its name. In the second, the same tool is asked for providers serving that company’s market and use case. The company disappears completely.
For marketing and SEO teams in Dubai, the UAE and elsewhere, this is a useful distinction rather than a platform failure. Branded vs unbranded AI prompts test different parts of AI visibility. Treating both answers as one score makes it hard to see whether the problem is recognition, category fit, commercial discovery or something else.
One company can pass recall and fail discovery
A branded prompt includes the business name, product name or a uniquely identifying term. It asks whether an AI assistant can recognise, describe or compare a known entity. An unbranded prompt omits that identifier and asks the system to find options based on a category, problem, location, buyer type or requirement.
Branded prompts measure recall and description. Unbranded prompts measure whether a business is surfaced during category discovery, comparison and recommendation. Each set should be scored separately because an accurate named answer does not show that the business will appear in buyer research, while one weak discovery answer does not automatically mean the brand information is wrong.
The distinction is particularly important before rebuilding pages, rewriting every service description or paying for an AI visibility programme. First establish which commercial decision each prompt is meant to inform.
Prompt families and the decision behind each one
| Prompt family | Buyer or business decision tested | A positive result can indicate | It cannot prove |
|---|---|---|---|
| Brand recall | Does the system know the name? | Entity recognition and basic factual retrieval | Unbranded market discovery |
| Brand description | Is the company described accurately? | Clarity of core services, location or positioning | That buyers will be referred to it |
| Brand comparison | Can the business be distinguished from alternatives? | Clear differentiators and category associations | Consistent preference over competitors |
| Category discovery | Which providers appear without a name supplied? | Potential visibility for broad buyer research | Suitability for every use case |
| Location discovery | Who is surfaced for a market or city? | Local entity and service relevance | National or international prominence |
| Problem-led prompts | Who can solve a stated operational need? | Answer-ready relevance to buyer intent | General brand awareness |
The existing ChatGPT brand-visibility test for discovery rather than recall is a useful starting point for named prompts. Its value is in setting a controlled branded foundation, then resisting the temptation to call that foundation a complete market-visibility result.
Score recall and discovery on separate sheets
Use the same basic discipline for both sets: record the prompt, platform, date, market setting where relevant, answer format and the exact scoring rule. AI outputs can change with wording, product changes and platform behaviour. A clean record will not remove variation, but it stops the team mistaking prompt drift for performance.
What a branded scorecard should inspect
- Whether the business is recognised without confusion with a similarly named company.
- Whether the primary category, core service and operating location are correct.
- Whether the answer repeats a stale service name, outdated market or unsupported claim.
- Whether a comparison prompt identifies a meaningful differentiator rather than generic praise.
What an unbranded scorecard should inspect
- Whether the business is included at all in a realistic set of options.
- Whether inclusion occurs for the right category, buyer type and use case.
- Whether the business is framed accurately when it appears.
- Which alternatives recur, and whether they are genuine commercial comparators.
- Whether the answer gives reasons, sources or caveats that expose an evidence gap worth investigating.
Do not compress these into one percentage. Report a recall score and a discovery score side by side, then add a short diagnostic note. If a broader view is needed, use a prompt portfolio method similar to an AI share of voice measurement checklist, where prompt coverage matters as much as an attractive single answer.
Mixed results are the useful part
Consider a software firm serving logistics operators. A branded prompt may correctly identify its platform, headquarters and product. Yet a prompt asking for software providers for fleet compliance reporting may not include it. That is not a contradiction. It suggests the system can retrieve the entity but does not strongly connect it to that unbranded buying problem.
The next inspection should be narrow. Check whether the relevant service or use-case page is indexable, present in the rendered HTML and described consistently. It is common to find the product name in Organisation structured data while the public service page uses a different label, with the useful use-case copy loaded only after JavaScript runs. That creates a weaker, less repeatable relationship between company, offering and buyer problem.
Conversely, if the firm appears in category discovery but receives a poor branded description, prioritise entity correction. Check the organisation name, founding facts, location references, product naming and structured-data relationships before changing category content. Discovery may be working despite messy recall, not because the mess is harmless.
A compact prompt portfolio that teams can maintain
Start with a manageable set rather than fifty variations that nobody reviews. For a specialist B2B company, use six to twelve prompts per market or priority service line, split across the families below.
- Two brand recall prompts using the company and product name.
- Two brand description or comparison prompts that test factual positioning.
- Two category discovery prompts using the service and buyer type.
- One or two location discovery prompts where geography genuinely affects buying.
- Two problem-led prompts using the operational issue a buyer would actually state.
Keep the prompts plain. A narrowly engineered phrase can produce an answer that looks good but does not resemble buyer research. Record reasonable variants separately rather than subtly changing the wording each month. The aim is controlled prompt tracking, not a small theatre production around a favourable result.
Questions teams ask about prompt sets
What counts as a branded AI prompt?
A prompt is branded when it includes a company name, product name, domain, founder name or another identifier that points directly to one business. Examples include asking what a named company does, whether it serves Dubai clients, or how it compares with a named competitor. These prompts are useful for testing recognition, factual accuracy and positioning.
How many unbranded prompts are enough?
Begin with six to twelve prompts covering the buyer decisions that matter most: category, use case, location and problem. Add prompts only when they represent a distinct research situation. Ten slight rewrites of the same category question create apparent coverage without testing much more. Review the set when services, markets or buyer language changes.
Should branded and unbranded scores be combined?
Usually, no. Keep recall and discovery separate because they answer different questions. A combined headline can be used for senior reporting only if the underlying split remains visible and the weighting is documented. Otherwise, strong brand recall can conceal weak commercial discovery, and a single poor discovery result can trigger unnecessary edits to accurate brand information.