Unbranded prompts tell you whether AI assistants recommend you to people who don't know you yet; branded prompts tell you whether those assistants get you right once someone asks about you by name. They answer different questions, respond to different fixes and move at different speeds, so mixing them into one score hides most of what your tracking can tell you. This guide covers the three segments worth separating, how to read them together and what each pattern says to do next.
The three prompt segments
Every prompt in a tracking set belongs to one of three segments, and each one answers a distinct question about your brand.
| Segment | Example | Question it answers | What mostly drives it |
|---|---|---|---|
| Unbranded | "Best scheduling software for a physical therapy clinic" | Do we make the shortlist? | Third-party consensus, use-case clarity |
| Branded | "[Brand] pricing for 10 users" / "Is [Brand] HIPAA compliant?" | Are we described accurately and favorably? | Your own pages, review profiles, press |
| Competitor-branded | "[Competitor] vs [Brand]" / "Alternatives to [Competitor]" | Do we show up in a rival's conversation? | Comparison pages, roundups, switching stories |
If you haven't built a set yet, start with how to build a prompt set worth tracking, which covers the overall balance. This post is about interpreting what comes back.
What unbranded prompts tell you
Unbranded prompts are the closest thing AEO has to a market-share measure: they show whether an assistant thinks of you when a buyer describes a need without naming anyone. A strong unbranded mention rate means the sources an assistant draws on associate you with the category and with specific situations.
Read unbranded results along three dimensions:
- Presence: are you named at all, and how often across repeated runs?
- Position: are you first, in the middle or in a "you could also consider" tail?
- Framing: which use case or segment does the assistant attach to you ("best for small clinics," "a budget option")?
Framing is the most overlooked: if assistants call you the budget choice and you sell to enterprises, the problem is positioning, not visibility. Calculating your share against named competitors is covered in how to measure AI share of voice.
What branded prompts tell you
Branded prompts show what a buyer hears at the moment they check you out, which usually comes after a referral, an ad, a sales email or an unbranded answer. Being mentioned isn't the question here (you almost always are); quality is.
Score branded answers on:
- Accuracy: pricing, features, integrations, locations, leadership and compliance claims
- Completeness: does the answer include your main differentiator or skip it?
- Sentiment: neutral, positive or hedged with old complaints
- Recency: does it reflect the current product, or one from two years ago?
- Sources: which pages the assistant cited for each claim
Useful branded prompt types include facts ("Who owns [Brand]?"), pricing, reputation ("[Brand] reviews," "Is [Brand] legit?"), fit ("Is [Brand] good for a team of 50?") and how-to questions from existing customers. Errors found here feed straight into finding and fixing AI hallucinations about your brand, and recurring negative framing deserves its own sentiment tracking.
What competitor-branded prompts tell you
Competitor-branded prompts reveal whether you are part of the conversation buyers have about the market leader, which is often where switching decisions happen. "Alternatives to [Leader]" and "[Leader] vs [Challenger]" prompts carry high intent and are usually less contested than broad category prompts.
Track two things: whether you appear in alternatives answers at all, and how you're framed in head-to-head comparisons (who "wins" each dimension, and whether the reasons given are accurate). Weak results here usually trace back to missing or vague comparison content. See comparison pages that win AI recommendations and "Alternatives to X" pages for AI search.
Reading the segments together
The combination of results across segments is more diagnostic than any single number, because each pattern points to a different kind of fix.
| Unbranded | Branded | Likely diagnosis | Where to act first |
|---|---|---|---|
| Strong | Strong | Healthy; defend and extend | New use cases, new platforms, freshness |
| Weak | Strong | Assistants know you but don't associate you with the category | Reviews, roundups, communities, PR tied to use cases |
| Strong | Weak | You get recommended, then described poorly | Fact pages, pricing clarity, correcting outdated sources |
| Weak | Weak | Thin footprint overall | Entity basics first, then third-party presence |
The second row is the most common pattern for young or niche brands. A clear website is enough for an assistant to answer "What is [Brand]?" but rarely enough to earn a spot on "best X for Y," which tends to reflect what many independent sources say. How AI assistants decide which brands to recommend explains why.
Use competitor-branded results as a tiebreaker: absent from unbranded prompts but present in alternatives answers means assistants see you as a challenger to a specific leader, a position you can build on.
Example: reading a split scorecard
The numbers below are illustrative. A mid-sized HR software company tracks 120 prompts across four assistants, five runs each, and sees:
- Unbranded (80 prompts): mentioned in 18% of runs, usually in the tail of the list, framed as "good for startups"
- Branded (25 prompts): 92% accurate on facts, but pricing answers cite a 2024 review site listing with old plan names
- Competitor-branded (15 prompts): named in 40% of alternatives answers for one leader, absent for the other two
A blended "mention rate" for this company would look respectable and say nothing. Split apart, it suggests three actions: update the pricing page and ask the review site to refresh its listing (branded, likely a quick win); publish comparison pages for the two leaders where it's absent (competitor-branded); and run a review and roundup program aimed at its mid-market use case, since the "startups" framing is holding back unbranded visibility.
How to classify prompts cleanly
Classification rules decide whether your segments mean anything, so write them down before tagging. Common decisions:
- Count product names, sub-brands, founder names and your domain as branded
- Count common misspellings of your name as branded
- Put prompts that name you and a competitor in competitor-branded, tagged "head-to-head"
- Treat ecosystem prompts ("best Shopify app for subscriptions") as unbranded, tagged with the platform
- Run each prompt in a fresh session, because earlier turns that mention your brand turn an unbranded question into a leading one (see optimizing for multi-turn conversations)
Run tracking in clean, signed-out or memory-free sessions where possible. Personalization can make a branded history leak into unbranded results, which flatters the numbers.
Reporting and targets
Report each segment on its own line with its own target, because they move at different speeds. Branded accuracy can improve within weeks of fixing a pricing page if assistants retrieve it live. Unbranded share of voice usually takes months, since it depends on reviews, roundups and coverage you have to earn.
Reasonable target shapes (set the actual numbers from your baseline, as covered in setting AEO targets):
- Unbranded: steady growth in mention rate and position against a fixed competitor set
- Branded: accuracy close to 100% on core facts, with no recurring negative claim left unaddressed
- Competitor-branded: present in alternatives answers for every leader you compete with
Common mistakes
- Overweighting branded prompts. They're easy to win, so a set full of them produces flattering numbers that don't predict pipeline.
- Ignoring branded accuracy because "we're mentioned." A mention with the wrong price can lose a deal you never knew about.
- Changing the prompt mix between periods. Shifting the ratio of branded to unbranded prompts changes the blended score even when nothing else moved.
The bottom line
Branded and unbranded prompts measure two different things: whether you're described correctly and whether you're recommended at all. Keep them, and competitor-branded prompts, in separate segments with separate targets, read the pattern across them to choose the fix, and resist any single blended score. AEO-Rank tags and reports prompts by segment across ChatGPT, Perplexity, Gemini, Claude and Grok, so each number answers the question it's meant to.
