Do AI vendor pairings survive a change in buyer intent?
Across a full study window, most co-mentioned vendor pairs held together regardless of how the question was framed, but a substantial minority only appeared together when the question sounded like a purchase.
61.9%
When an AI assistant is asked to name vendors in a category, it often names more than one at a time, and it tends to name the same handful together repeatedly. This study asked a narrower question than "which vendors get named": does a given pairing of two vendors survive when the buyer's stated intent changes? A pairing might show up when someone asks a broad, informational question about a category, or it might only show up when the question is phrased the way a buyer close to a purchase would phrase it, for instance asking for a comparison or a recommendation to act on.
The unit of analysis here is what the study calls a pair-by-group observation: one specific two-vendor pairing, checked within one specific group of competing vendors, under two different query framings. There were 1,390 such observations across 43 groups of vendors, built from 1,390 underlying rows with 0 excluded, meaning the full dataset was usable. Each observation was classed into exactly one of two outcomes: the pair clustered on both intents, meaning it appeared together whether the question was informational or commercial, or the pair clustered on commercial intent only, meaning the co-mention only showed up under the buying-oriented framing.
Over the measurement window of 2026-05-01 to 2026-09-15, 61.9% of pairs held together across both framings. The remaining 38.1% appeared together only when the query carried commercial intent. That is a real majority holding stable, but not an overwhelming one. A little over one in three pairings that an AI assistant puts side by side when a buyer is closer to purchasing do not show up together at all when the same buyer is just researching the category.
Why would this happen? The most direct explanation is that the underlying language the AI model was trained on and retrieves from treats "best X" and "X vs Y" content differently from "what is X" or "how does X work" content. Comparison pages, review roundups, and buyer's-guide content naturally pair vendors that compete head-to-head for the same purchase decision, which is exactly the commercial-intent framing. Informational content is more likely to discuss a vendor on its own, or paired with whatever else is topically adjacent, which need not be a commercial competitor at all. So a pair that only clusters under commercial intent is not necessarily a weak or accidental pairing, it may simply be a pairing that lives in comparison content and nowhere else.
An alternative explanation worth naming: this could be an artifact of query phrasing rather than a genuine shift in the assistant's underlying model of the category. If the two intent framings used in this study differ in ways beyond just informational versus commercial, for instance in specificity or length, some of the apparent split could reflect that rather than intent itself. The study design classed pairs by outcome, not by the wording of the queries themselves, so this page cannot rule that out. What would distinguish the two explanations is testing multiple phrasings within each intent category and checking whether the both-intents rate holds steady, which is outside the scope of what was measured here.
The practical takeaway for a company tracking its own co-mentions is that a single snapshot, taken under one query framing, is not enough to know whether a pairing is durable. A vendor that shows up paired with a competitor only under "best X" style prompts, and never under plainer informational ones, is being positioned by the assistant as a comparison-shopping alternative, not as a generally associated name in the category. Whether that is good or bad depends on what the vendor wants: comparison-page visibility is arguably more valuable for late-stage buyers, but it means the vendor has less presence earlier in the research process, when the informational framing dominates. The overall split, roughly three pairs holding steady for every two that don't, is the headline number, but as the next finding shows, that average badly understates how differently this plays out depending on how tightly a group of vendors is already bound together.
Other findings
Does the strength of a vendor grouping predict whether its pairings survive a change in buyer intent?
Why do tightly grouped vendor pairs hold together almost universally?
Pairs from tightly grouped vendor sets held together across intents at a rate far above the overall average, while loosely grouped pairs split almost evenly, showing that grouping tightness is the main driver of a pairing's durability.
Is the commercial-intent-only clustering pattern consistent across vendor groups, or does it depend heavily on which group you look at?
How much does a commercial-only pairing rate vary from one vendor group to another?
The share of pairs that cluster only under commercial intent ranges from zero to the full group across the 28 groups where this was measured, meaning the overall average describes no single typical group well.
What exactly counts as a unit in this study, and how much of the raw data was excluded or adjusted before reporting?
How was the vendor-pairing dataset for this study built?
The dataset covers 1,390 pair-by-group observations across 43 vendor groups with no rows excluded, though three mitigations were applied during construction, which matters for how much weight to put on the headline percentages.
Want to know how AI answers describe you?
We run the same measurement on your category. Fifteen minutes with founder Omar Jenblat, your own numbers, no deck.
- Your category measured the same way
- Your own numbers, not a sample deck
- Fifteen minutes, no obligation
