AI Search Visibility Tracking for Branded and Unbranded Mentions

Two kinds of AI visibility, two different jobs

An unbranded mention happens when a buyer asks a category question without naming you, such as 'which payroll software suits a 50-person company' or 'best regulated broker in Bahrain', and the engine names your brand anyway. That is discovery: AI introducing you to someone who did not know to look for you.

A branded mention happens when the buyer already has your name and asks about you: 'is Vendor X good for enterprises', 'Broker X fees', 'Vendor X vs Vendor Y'. The engine is no longer deciding whether to include you; it is deciding what to say. Here the risk is not absence but inaccuracy: an outdated price, a wrong licence, an old complaint presented as current.

Both matter, but for different reasons, and they respond to different work. That is why AI search visibility tracking for branded and unbranded mentions should keep them as two panels with two sets of metrics.

Why blended numbers mislead

If you mix branded and unbranded prompts in one panel, your coverage figure is inflated. Engines almost always mention a brand when the prompt names it, so branded prompts push the blended number up regardless of how you perform on discovery. A team can report 'present in 60% of AI answers' while being absent from nearly every category prompt that matters for new pipeline.

The reverse problem also occurs. A blended view can hide a reputation issue on branded prompts, such as an engine repeating a resolved complaint, because the coverage number looks healthy. Splitting the panels removes both distortions and makes each metric mean what it says.

What to track for each prompt type

For unbranded prompts, the core metrics are brand coverage (share of runs that name you), share of voice against named competitors, average position in the answer, and citation share for your own domain. Tag prompts by funnel stage so you can see whether you appear in early education questions, mid-funnel shortlists or late comparisons.

For branded prompts, track accuracy against a fact sheet (prices, products, markets, licences, leadership), sentiment, the sources cited when engines describe you, and which competitors engines bring into answers about you. A branded answer that says 'Vendor X is fine, but most teams choose Vendor Y' is a mention, but not a good one.

• Unbranded: are we on the shortlist, how high, and against whom?
• Branded: is the description correct, fair and sourced from pages we can influence?

Worked example: a forex broker's two panels

A broker active in the UAE and Saudi Arabia builds an unbranded panel of 60 prompts, such as 'best forex broker for beginners in the UAE', 'brokers with Islamic accounts and low gold spreads' and 'which brokers offer copy trading in Saudi Arabia', in English and Arabic. The branded panel has 30 prompts: 'is Broker X regulated', 'Broker X minimum deposit', 'Broker X withdrawal problems', 'Broker X vs Broker Y'.

The first baseline shows decent unbranded coverage in English, almost none in Arabic, and a branded accuracy problem: two engines cite an old review that lists a higher minimum deposit and a platform the broker has dropped. The two panels lead to two work streams. Arabic content and regional placements address discovery. A dated fact page, a corrected review listing and a clear licence page with the entity name and local regulator address the branded answers. Each stream is judged on its own panel.

Choosing prompts for each panel

Good unbranded prompts come from how buyers describe their problem before they know vendors: sales discovery notes, search queries in Search Console, community questions. Include context such as company size, location and constraints. Good branded prompts come from the questions prospects ask once they know you: demo questions, objections, review-site complaints and 'vs' searches.

Keep both panels stable over time so trends remain valid. Add new prompts as new lines rather than editing old ones. Review the mix quarterly against what sales hears, because a new competitor or a new product line changes both panels.

For branded prompts, keep a short fact sheet beside the panel: current prices, account or plan types, markets served, licence or certification details, leadership names and key product limits. Every branded answer is scored against it. The fact sheet also serves the fixing work, because it defines exactly what your own pages and third-party listings should say. When a fact changes, update the sheet first, then the pages, then watch the branded panel to see how quickly engines pick up the change.

Reporting and what Suggesting.ai does

Report the two panels side by side. Discovery: coverage, share of voice and position by engine and market, with AI referral sessions from unbranded landing pages. Reputation: accuracy issues open and closed, sentiment trend and citation sources for branded answers. Leadership can then see whether AI is introducing you to new buyers and whether it describes you correctly to buyers who already know you.

Suggesting.ai builds both panels in its free 48-hour brand and AI presence audit and keeps them on independent AI-search tracking with repeated runs. Organic GEO work closes unbranded gaps through content and third-party corroboration, and fixes branded inaccuracies at their source. Where the category is eligible, ChatGPT Ads can add presence on unbranded prompts; for financial services outside the US, OpenAI's ad policy generally prohibits them, and the audit confirms eligibility. The goal is simple: AI suggesting you to new buyers, and describing you correctly to existing ones.