Visibility Rate vs Rank: Can You Rank #1 in ChatGPT?

Is there a stable rank in AI answers?

No — not in the way "rank #1 on Google" means something. A traditional search results page has a genuinely stable order: the same query, asked twice in short succession, returns the same ten links in the same sequence, barring an algorithm change. That stability is what makes "we rank #3 for this keyword" a meaningful, trackable fact.

ChatGPT and other generative engines don't have that property. The same prompt, asked twice, assembles a new answer each time — a new selection of which sources to pull from, a new synthesis, a new order. There's no fixed list sitting behind the answer waiting to be revealed consistently; each run is closer to a fresh draft than a lookup.

What does the evidence on repeat runs say?

A SparkToro and Gumshoe.ai study (600 volunteers, 12 prompts, roughly 2,961 runs across ChatGPT, Claude and Google's AI, November–December 2025) is the clearest public test of exactly this question. It found under a 1-in-100 chance of an identical brand list appearing twice, and under a 1-in-1,000 chance of that identical list appearing in the same order twice. We cover the full study, including its disclosed conflicts of interest, in how many runs you actually need — the relevant finding here is narrower: order was the least stable thing measured.

What did hold up: appearance frequency. Certain brand names recurred in 60% to 90% of responses for a given category and intent, even as the exact ranking shifted under them. That's the core asymmetry this whole argument rests on — presence is a real, measurable, relatively stable signal; position on any single run is close to noise.

What should replace rank as the target?

Visibility rate: how often a brand is named across a meaningful number of repeated runs, stated as a rate with its run count attached, not a single ordinal position. "Named in 7 of 10 runs" is a defensible, honest claim. "We're #2 in ChatGPT" from one check is not, because the evidence above says that specific ranking has a real chance of being different on the next identical question.

Share of voice — a brand's slice of every mention across a fixed prompt set, covered in how to calculate it — is the natural second metric once appearance is tracked properly: it answers "how do we compare to named competitors" without leaning on an unstable ordering. This post's thesis is the same one argued at greater length in presence, not position, including the strongest case for the other side.

What is the best case for tracking position anyway?

It's not nothing, and a fair version of this argument should say so. Within a single answer, a brand named first probably does get more attention from the reader than one named fifth or buried in a longer list — attention within one run is a real thing, even if that specific ordering is unlikely to repeat on the next run. If a team is specifically trying to understand "when we do appear, are we usually leading the list or trailing it," averaging position across many runs (not reading one run's position as fact) is a legitimate, if secondary, thing to track.

The mistake isn't tracking position at all — it's reporting a single run's position as if it were a stable ranking, the way a Google rank check behaves. Averaged across enough runs, position becomes a soft signal worth a footnote. From one run, it's closer to a guess that happened to land somewhere.

What would change our mind?

A study showing that repeated runs of an identical prompt, on any major generative engine, return brands in a meaningfully stable order — not just a stable set of brands, but a stable sequence. That's a specific, falsifiable claim, and nothing published so far supports it; the one large public study on this (SparkToro and Gumshoe.ai) found the opposite. If an engine vendor or an independent researcher publishes evidence of order stability at scale, that would undercut this post's thesis directly, and we'd update it.

It's also worth naming our own incentive here plainly: ContextOwl reports visibility and share of voice, not a ranked position, so this argument is convenient for our own product framing. That doesn't make the underlying repeat-run data wrong — the study behind it wasn't run by us and its authors have their own, different commercial interest (selling AI tracking tools) — but it's a reason to check the source yourself rather than take our framing of it on faith.

FAQ

Is there any sense in which "rank #1 in ChatGPT" is meaningful? Only within a single specific run, as "named first in this particular answer" — not as a stable, repeatable fact the way a Google rank is. Treat any ranking claim with a run count attached, or be skeptical of it.

If position is unreliable, why do some vendors sell an "AI rank" product? It's an easy number to sell because it mimics a format buyers already understand from Google rank tracking. The evidence on repeat-run variance says that familiarity doesn't make it an accurate measurement.

Does visibility rate have the same instability problem as rank? Less of one. The same study found appearance frequency held up far better than order across repeated runs — it's not perfectly stable either, but it's the more defensible of the two to report.

How many runs does it take to trust a visibility rate? More runs narrow the uncertainty; there's no single correct number. See how many runs you actually need for the tradeoffs.

What to do next

1. If your current reporting includes an "AI rank" number from a single run, add the run count next to it — or drop it until you have one.
2. Start tracking visibility rate (named in X of Y runs) as the primary number for any AI-answer metric you report.
3. Keep position as a secondary, averaged-across-runs signal at most — never a single run's finding.

The AI Citation Snapshot shows one live answer and whether you're named — useful context for a single run, not evidence of a stable rank.