- A consideration set is the small group of named competitors a model actually surfaces together for a comparison query. Whether you're in it at all is a different, earlier question than whether you win it.
- Consideration-set stability isn't one number. A 14,000-API-call Conductor study found brand overlap between repeated runs ranging from 40% at Purchase intent to 63% at Comparison intent. Tracking has to be segmented by intent, or the blended average hides exactly the distinction that matters.
- The mechanism for tracking this extends the fixed-prompt-set method already covered for general AI visibility. The change is tabulating named-competitor co-occurrence, not just your own mention frequency.
- Appearing in the set and winning it are different problems that respond to different levers. Extractability work can help you appear; earned off-site proof is what tends to move you to the top once you're there.
Mention vs. citation vs. recommendation already separates recommendation from the other two outcomes. This piece is the practical measurement layer on top of it: how to actually track whether you're even in the competitive set a model considers, since that's the leading indicator. A brand that never appears in the comparison at all has no chance of winning it, regardless of how good the underlying product is.
What is a consideration set, and why does appearing matter before winning does?
A consideration set is the small group of named alternatives a model surfaces together for a comparison or best-of query: "X vs. Y," "alternatives to X," "best tool for [use case]." It's a narrower, more specific thing than a general mention count. A brand can have a healthy citation rate on factual queries (pricing pages, documentation, how-to content) and still be structurally absent from the handful of prompts that actually compare it against the competitors a buyer is choosing between.
That's a different failure mode from losing a comparison, and it's the one a raw AI-mention tracker won't surface on its own, because it's measuring "were we mentioned at all," not "were we mentioned in the same breath as the companies a buyer is actually deciding between."
Why does this have to be tracked by intent, not as one blended number?
Being cited vs. being chosen already covers Conductor's 14,000-API-call study in the context of recommendation stability generally, and the same data is the direct argument for why consideration-set tracking specifically can't be one blended metric. Their finding, across 10 industries, 7 intent types, and 4 LLMs: brand overlap between repeated runs of the identical prompt ranged from 40% at Purchase intent up to 63% at Comparison intent, with lead-brand retention ranging from 59% to 91% across the same span.
Averaging those into one "consideration-set stability" number would hide the actual finding: a brand's set membership can be quite stable on Comparison-style queries while being nearly a coin flip on Purchase-style ones. A tracking setup that reports one blended percentage is reporting an average of two genuinely different realities, and can't tell you which one you're actually looking at.
How do you actually build this?
How to build your own AI visibility tracking already covers the mechanics: a fixed prompt set, run repeatedly per platform, logged with timestamps, read as a trend rather than a single check. Consideration-set tracking is the same infrastructure with two specific changes:
- The prompt set itself needs comparison and best-of phrasing, segmented by intent category, not general "tell me about [brand]" prompts, which measure mention frequency, not set membership.
- The thing being tabulated changes. Instead of counting your own appearance rate, log every named competitor that appears alongside you (or instead of you) on each run, for each prompt. Over enough runs, this produces a real co-occurrence pattern: which competitors you're actually being compared against, and how often you're present versus absent from that comparison, split by intent category rather than blended into one number.
The same reliability standard applies here as anywhere else on this site: a single run of a comparison prompt is one roll, not a reading. The AI visibility KPIs discipline (enough runs per prompt, tracked as a trend over weeks) applies to consideration-set membership exactly as much as it applies to citation frequency.
What moves you into the set, versus what moves you to the top?
These are genuinely different levers, and conflating them wastes effort. Getting into the set at all is closer to a content-extractability and entity-consistency problem: a model has to be able to resolve who you are and find content that plausibly answers the comparison query in the first place. Winning the comparison once you're in it is closer to an off-site, earned-proof problem: reviews, independent comparisons, and press are what a model weighs when deciding who to name first or most favorably among several plausible options, the same mechanism why earned media outperforms owned content already covers for citations generally.
Tracking consideration-set membership separately from consideration-set rank is what makes it possible to tell which of the two problems you actually have, instead of applying a generic "more content" fix to what might actually be a review-volume problem, or vice versa.
Frequently asked questions
What is a "consideration set" in AI search, and why track it separately from citations?
It's the small group of named competitors a model actually surfaces together for a comparison or best-of query, the AI-search equivalent of the shortlist a buyer builds before deciding. It's a different, earlier failure mode than losing a comparison: a brand can have a perfectly good citation rate for factual queries and still never appear at all in the specific queries that actually compare it to competitors, which a raw mention count won't surface.
How do you actually track whether you're in a competitor's consideration set?
The same fixed-prompt-set methodology already covered for general AI visibility tracking, with one change: instead of counting how often you're mentioned, tabulate which named competitors appear alongside you (or instead of you) on comparison and best-of style prompts specifically, run repeatedly per the same reliability standard, enough runs per prompt to see a stable pattern, not a single check.
Why would tracking need to be segmented by query intent specifically?
Because consideration-set stability varies enormously by intent, not by brand. A Conductor study of 14,000 API calls across 10 industries and 7 intent types found brand overlap between repeated runs ranging from 40% at Purchase intent to 63% at Comparison intent. Treating those as one blended number would average away the exact distinction that matters: whether you're stable where it counts most, at the moment closest to a buying decision.
What's a realistic cadence and setup for this?
The same cadence as any other fixed-prompt-set tracking, weekly, not a one-off, but the prompt set itself needs comparison and best-of phrasing specifically ("X vs Y," "alternatives to X," "best tool for [use case]"), segmented by the intent categories that matter for the buying journey, rather than one general mention-tracking prompt list.
Sources referenced
- Conductor: AI Brand Recommendation Study — Why Intent Type Predicts AI Output Consistency (14,000 API calls, 10 industries, 7 intent types, 4 LLMs)
This piece is the measurement layer on top of mention vs. citation vs. recommendation and being cited vs. being chosen, extending the tracking mechanics in how to build your own AI visibility tracking.
