Branded prompts and unbranded prompts behave like two different channels in AI search. Branded queries surface owned pages, support docs, and reviews; unbranded queries surface category lists, comparison posts, and third-party authorities. Reporting them under a single visibility score hides what is working. Split the measurement, split the source map, and split the content investment.
Why branded and unbranded prompts now behave like two channels
“Is Acme worth buying?” and “What’s the best CRM?” tell you different things about your AI visibility. Unbranded prompts behave like a discovery channel: the model synthesizes an answer from category authorities, and your brand is one of several names that may or may not appear. Branded prompts behave like an owned-media channel: the question already names you, so the model retrieves your site, reviews about you, and stories told about you.
Parse tracks AI visibility across ChatGPT and Google AI Mode. Looking across those answers, the consistent pattern is that a brand can be strong on branded prompts and invisible on unbranded ones, or the reverse. Reporting them together masks both failures.
What AI Overviews and ChatGPT do with each prompt type
Unbranded prompts dominate the AI answer surface. Semrush's analysis of 200,000 AI Overviews found the sample skewed heavily toward informational queries, and Ahrefs reported that AI Overviews appeared on roughly one in five desktop SERPs in its September 2025 dataset of 146 million SERPs. Most of those are unbranded category questions, the kind where the user has not decided who to buy from.
Branded prompts behave differently. When the query names you, the model has a stable retrieval anchor: your domain, your knowledge-panel data, your Wikipedia or Wikidata entry, your G2 or Capterra page. Citation sets compress around that anchor. Pew Research Center found that Google users clicked an external result on 8% of searches with an AI summary versus 15% without, but the click bias is largest on informational queries, not branded ones. People still click when they were going to your site anyway.
The operational read is simple: unbranded prompts are how AI introduces you, branded prompts are how AI describes you.
Discovery channel. Category authorities, comparison content, and community sources earn the citation. You compete to appear at all.
Owned-media channel. The model retrieves your site, reviews, and entity data. You compete on what the answer says about you.
Comparison and alternative queries. Citations split between you, competitors, and review platforms. Share of voice is the metric that matters.
If you want to know when AI changes its answer about your brand, start with a free brand check. It takes a minute.
Which sources AI cites for branded vs unbranded prompts
The citation set diverges sharply by prompt type. On unbranded category prompts, the recurring high-trust sources are large editorial brands (G2, Capterra, Forbes Advisor), specialized blogs (Search Engine Land, Search Engine Journal), community sources (Reddit, Stack Overflow, Quora), and a small set of category-defining tools and vendors. Parse ranks the source domains AI leans on most in the sources behind AI answers. Semrush's AI Overviews dataset and Conductor's AEO and GEO benchmarks both show the same pattern across information-heavy verticals.
On branded prompts, the citation set narrows. The model typically pulls from the brand's own site, the brand's documentation, review platforms with detailed profile pages, and a few editorial reviews. Wikipedia is overrepresented for established companies; Wikidata is increasingly load-bearing for entity disambiguation. BrightEdge found that ChatGPT and Google AI disagreed on brand recommendations 62% of the time, which is also a signal that the source mixes feeding each model differ, especially on category prompts that have many plausible answers.
You can map the same pattern in your own data. Pull ten branded and ten unbranded prompts in your category. Record the cited URLs by domain. The two lists almost never look the same.
How to split your measurement so the funnel is legible
Stop reporting one AI visibility score. Build two scoreboards that sit beside each other.
The unbranded scoreboard tracks discovery: how many category and buyer-intent prompts mention you, the share of mentions versus named competitors, and the share of citations that go to a third-party source (G2, Reddit, an editorial review) versus an owned page. This is your "are we even in the consideration set" metric.
The branded scoreboard tracks defense: presence on every prompt that names you, the citation share for your own domain, sentiment of the answer, and the presence of competitor names inside answers about you. This is your "what does the model say when it has already heard of us" metric.
- Unbranded prompts are a discovery channel; branded prompts are an owned-media channel.
- Citation sources diverge sharply between the two prompt types; one dashboard hides that divergence.
- Measure discovery as share of mentions and citation share; measure defense as presence, sentiment, and competitor mentions in branded answers.
- Invest unbranded budget in earned media, community, and review platforms. Invest branded budget in owned content, documentation, and entity data.
- Build the prompt set, the source map, and the executive report around the split.
For tactical depth on building the underlying prompt list, see which prompts should you track. For the executive view, see how to report AI visibility to your CEO.
What to invest in for each side of the split
The investment decision flows from the source map. On unbranded prompts, the citations point to sources you mostly do not own. That is where earned media, community presence, review platform optimization, and authoritative third-party content matter. Adobe's holiday season analysis reported AI-referred traffic to retail grew 4,700% year over year and converted at a higher rate than other channels, but only on the prompts where the brand appeared in the answer set at all. Discovery investment is what gets you into that set.
On branded prompts, the citation control point is your own surface area. Owned content depth, FAQ schema, current and complete documentation, accurate knowledge-panel data, and Wikidata or Wikipedia presence determine what the model says about you. The marginal return per dollar is high because branded prompts have far fewer competing sources to displace.
A practical split that fits most mid-market budgets: 60 to 70% of new spend into discovery (unbranded) when you are starting from low share of mentions, 50/50 once you reach reasonable category presence, and a defensive tilt toward branded investment once the funnel is healthy. The wrong move is to spread evenly without checking which scoreboard you are actually losing on.
Branded and unbranded prompts answer different questions, recruit different sources, and respond to different work. Treating them as one metric tells the executive team a story the data does not actually support. Split the dashboard, split the source map, and the funnel becomes legible again.