Profound ran 221 real ChatGPT shopping tasks with 56 participants in 2026, and the brands people actually picked turned up in the model's answers nearly twice as often as the ones they rejected, 24% appearance against 11%, with a 0.57 correlation between visibility and the final choice. None of those prompts named a brand. They were plain category questions, the kind no branded-term tracking ever captures. Most AI Share of Voice dashboards still default to exactly that branded view, checking whether an engine recognizes a company's name while skipping the unbranded queries where the buying decision actually gets made.
What branded AI Share of Voice actually measures
Branded AI Share of Voice tracks prompts that already contain a company's name, phrases like “is Asana good for a five-person team” or “HubSpot pricing for startups.” It tests recognition, whether an engine knows a brand exists and describes it correctly, usually scored as a simple mention rate across a templated list of queries, the same methodology we laid out in detail previously. Rankscale's own prompt-tagging framework calls branded visibility the floor: fail it, and nothing else about a brand's AI presence compounds. Pass it, and all that's proven is that the model has heard of the company, not that it would ever suggest it to someone who hasn't.
What non-branded AI Share of Voice actually measures
Non-branded AI Share of Voice tracks the opposite: prompts that never mention a brand at all, like “best CRM for real estate investors” or “what should I use to track my brand across ChatGPT.” Here the engine has to surface a company unprompted, competing against every rival it considers relevant to that exact question. Rankscale frames this as the ceiling, the only place real pipeline growth comes from, because a buyer who already knows to type a company's name has usually made up their mind already. It's the split we gestured at when we first defined AI Share of Voice, without yet separating the two prompt types out.
What the Profound shopping study actually found
Profound built the clearest evidence for that ceiling effect with a July 2026 study run alongside growth consultant Kevin Indig and Eric Van Buskirk of Clickstream Solutions. Fifty-six participants completed 221 real ChatGPT shopping tasks, and researchers recorded screens, transcripts, and think-aloud commentary alongside the model's actual citations. Brands shoppers ended up choosing appeared in 24% of ChatGPT's answers, against 11% for brands they passed over, with a correlation of 0.57 between visibility and the final pick.
Visibility and choice correlate strongly at 0.57.
That's a strong relationship for consumer research, and it held even though 88% of the shopping tasks involved a brand the participant had never heard of going in. The study's authors found that unfamiliarity almost never cost a brand the sale once a product was presented clearly inside the chat. Shoppers clicked through to an outside website in just 7.2% of tasks, and comparison grids inside the chat window captured attention in 35.9% of them. None of that shows up in a branded tracking report, because none of it starts with a brand name.
Why most AI Share of Voice tools still default to branded tracking
Branded prompts are mechanically easy to generate: take the company name, template it against twenty question formats, and the list is done. Non-branded prompts require real research, pulling the category language actual buyers use from support tickets, sales call transcripts, and review sites, then testing which of those phrasings an engine treats as relevant to a brand at all. That's closer to keyword research than brand monitoring, which is probably why the citation gap keeps surprising marketing teams long after launch. We measured the scale of that problem in our look at Semrush's 126-million-prompt study: a brand can be mentioned constantly on branded prompts and still miss nearly every category question that would have put it in front of a stranger.
The honest limits of AI Share of Voice as a metric
None of this makes AI Share of Voice a clean number, branded or not. Search Engine Land's Dan Taylor argued in June 2026 that the metric rests on what he called a hidden denominator: the universe of possible AI prompts is effectively infinite, so any vendor's sample, however large, amounts to a closed sandbox being presented as the open web. He pointed to OpenAI's move to GPT-5.0 in September 2025 as proof the number can move for reasons that have nothing to do with a brand's actual standing, since the platform-wide volume of outbound citations dropped and visibility scores fell with it, the same kind of swing we tracked when ChatGPT's own prompt share moved by twenty points within a single year. His proposed fixes, share of mentions, share of recommendations inside comparison answers, and share of narrative, the tone a brand gets described in, don't resolve the branded-versus-non-branded split either. Whichever metric a team picks, it still needs to be read against both prompt types separately, not blended into one score.
Does the shopping-study pattern hold for B2B buying too?
Profound's data comes from consumer shopping, not B2B software evaluation, and the two don't automatically transfer. The structural reason non-branded visibility matters doesn't depend on the product category, though: a buyer researching project management tools, payroll software, or a CRM is asking the same kind of unprompted question Profound's shoppers asked about kitchen gadgets, and an engine still has to pick which vendors to name without being told one. B2B research adds a wrinkle retail shopping doesn't have, since longer consideration cycles mean the same buyer may run a non-branded prompt weeks before ever typing a branded one, so a tool that only tracks branded prompts never sees the moment that mattered. Until a Profound-scale study runs specifically on B2B software queries, the honest read is that the mechanism almost certainly applies, even though the exact 24-versus-11 split is a shopping number, not a SaaS one.
How to split branded and non-branded tracking without starting over
Most teams already run branded AI Share of Voice, so the fastest fix is layering non-branded prompts on top rather than replacing anything. Pull the actual phrases prospects use before they know a company's name exists, mostly from support tickets, win-loss interviews, and review sites, the same sourcing detailed in our guide to increasing AI Share of Voice. Score the two prompt sets separately in every report leadership sees, since averaging them hides exactly the gap this piece is about. Re-run both sets after any major model release, because Taylor's GPT-5.0 example shows scores can shift for reasons that have nothing to do with content changes. And when a competitor shows up in a non-branded answer instead of a brand's own name, treat it the way we described in our competitive AI Share of Voice analysis: as a gap to close, not a one-off loss to shrug off.
Frequently Asked Questions
What's the difference between branded and non-branded AI Share of Voice?
Branded AI Share of Voice tracks prompts that already name a company, like “[brand] pricing.” Non-branded AI Share of Voice tracks category prompts that never mention a brand, like “best project management tool for agencies.” Branded measures recognition; non-branded measures whether an engine recommends a brand to someone who didn't ask for it by name.
Why does non-branded AI Share of Voice matter more for pipeline?
Profound's 2026 shopping study found chosen brands appeared in 24% of ChatGPT's answers versus 11% for rejected ones, all on prompts that never named a brand. A buyer who already types a company's name has usually decided; the non-branded moment is where that decision actually gets made.
What did Profound's shopping study actually measure?
Profound, with Kevin Indig and Clickstream Solutions, ran 221 real ChatGPT shopping tasks across 56 participants in 2026, recording screens and transcripts alongside the model's citations. It found a 0.57 correlation between how often a brand appeared in answers and whether shoppers picked it, and that unfamiliarity with a brand rarely hurt its chances.
Is AI Share of Voice a reliable metric at all?
Partially. Search Engine Land's Dan Taylor argued in June 2026 that the metric rests on a hidden denominator, since the universe of possible AI prompts is effectively infinite and no vendor's sample fully represents it. It's most useful when branded and non-branded prompts are scored and reported separately, not blended into one number.
How do I start tracking non-branded AI Share of Voice?
Pull the category phrases real buyers use before they know your name, mostly from support tickets, win-loss interviews, and review sites. Score those prompts separately from branded ones in every report, and re-test both sets after major model releases, since scores can shift for reasons unrelated to your content.
Does the shopping-study pattern apply to B2B software buying?
Likely yes in mechanism, though the exact numbers are unproven outside retail. A B2B buyer asking a category question without naming a vendor faces the same unprompted-recommendation problem Profound measured in shopping, just with a longer research cycle before any branded prompt appears.