AI Rank Tracking: 88% of Brand Mentions Land in the Top 5

What We Measured

Every tracked answer between 2026-07-06 and 2026-08-22 where the brand was named and its ordinal position in the answer could be read. That is 531 positioned mentions. Answers where the brand was absent are not in the denominator, so this describes where you land given that you were named at all.

The window starts in July because our engines moved to the current generation of web-search models around then. It matters: measured over all time instead, first place reads 52.9% rather than 35%. The wider figure is the flattering one and it describes retired models, so the narrower one is what this post uses.

The Shape of a Mention

The decay is steep and immediate. First place takes 35% of all mentions, second 21.1%, third 15.1%. By position six the share has fallen to low single digits and it never recovers.

This is not the familiar click-through curve from search, where the decay describes attention across a list everyone can see. Here the decay is mostly the list ending. An answer that names four products simply has no position five, so the tail is thin because it barely exists rather than because nobody reads it.

There Is No Page Two

Only 11.5% of mentions land past position five, and the furthest any brand appeared in this window was #20, once. Compare that to organic search, where ranking 40th is an ordinary state of affairs and improving to 15th is real progress worth reporting.

Two things follow. First, a rank-tracking dashboard built on the SEO model will spend most of its pixels on positions that almost never occur. Second, and more useful, the interesting event in AI search is binary. You entered the answer or you left it. Watching a brand move from third to second is watching noise; watching it disappear from the answer entirely is watching something happen.

Position is also subject to the same run-to-run movement as every other metric here. We measured that separately: a visibility score moves up to 15 points between runs with nothing changed. A one-place shift in a five-item list is well inside that.

Why This Is Not the SEO Curve

The distribution above looks superficially like the click-through curve every SEO knows, where the first result takes a third of the clicks and the tenth takes almost none. The resemblance is misleading, and acting on it is where teams go wrong.

In search, all ten results exist. Ranking tenth is a real, occupied position, and the decay describes how people behave once they see the page. Improving from tenth to fourth is a change in a stable world: the slots were always there, and you moved between them.

In an AI answer the slots are created fresh each time. A question that gets a four-item answer has no fifth place for anyone to hold. So the thin tail here is not mostly a story about attention, it is a story about supply. There is no queue of brands sitting patiently at position twelve waiting for their turn to move up.

That changes what an improvement even looks like. In search you climb. In an AI answer you are either included in a list that gets rebuilt on every query, or you are not, and the work that changes the outcome is the work that changes what the model finds when it rebuilds it. Chasing a place is chasing something that does not persist between runs.

It also explains a frustration people report constantly: their position bounces without explanation. It is not a mystery and it is usually not a change in standing. A shorter answer this week shuffles everyone, and a brand that was fourth is now second without having done anything. Reading that as progress, and repeating the tactic that supposedly caused it, is how teams end up optimising for randomness.

Does the Engine Change It?

Barely, and the honest read is that the differences here are not findings. Gemini puts a named brand first 40.1% of the time against Perplexity's 30.4%, a 9.7-point spread on a few hundred mentions each.

We are pointing that out rather than dressing it up because the shape is what generalises. All three engines produce a short list with a heavy first position, and none of them produces anything resembling ten ranked results. If you are choosing where to spend effort, the engine you care about should be decided by where your buyers ask, not by a few points of first-place share.

Reporting Position Without Overclaiming

Position still belongs in a report. It is the difference between being the recommendation and being the also-ran in someone else's paragraph, and clients ask about it. The problem is only what you claim when it changes.

Three rules keep it honest. Report position as a distribution rather than an average, because an average of a short list hides the only split that matters: how often you were first. Report it per prompt, not per account, so a gain on one question cannot be washed out by a loss on another. And pair it with presence, because a brand that moved from fourth to second while dropping out of two answers entirely has got worse, not better, and the position number alone will say the opposite.

The trap worth naming is the average. It is the natural thing to put on a slide, it moves every single cycle, and on a five-item list it moves for reasons that have nothing to do with the brand. Anyone reporting an average position monthly is committing to explaining a number that is mostly the length of this month's answers.

What to Track Instead

Position is worth recording. It is just a poor headline. Four things carry more information per unit of attention:

• 01Presence, per prompt. Whether you are in the answer at all, prompt by prompt. That is the binary the data says actually moves.
• 02Share of voice. On a short list, being one of three names is a different situation from being one of eight, and a bare position cannot tell them apart.
• 03Which prompts you lost. A prompt you have been absent from in two consecutive runs is a work item. A prompt where you slipped one place is not.
• 04The pages behind the answer. The citations are the lever you can act on; ranked listicles are the format most likely to put you in a position in the first place.

None of this means position is meaningless. First place is genuinely worth more than fourth, and it is a third of all mentions. It means the metric has two useful states and a long tail of noise, and a tracking setup should be built around the two states.