Why near-universal visibility does not mean a stable number one

Two findings from this study can look contradictory until they are placed side by side. On one hand, visibility is broadly common: 1596 of 1664 companies in the cohort, 95.9%, appeared somewhere in at least one engine's answers, and zero companies in the cohort were completely invisible across all three engines (0 of 1664, a rate of 0%). On the other hand, the identity of the single company named first is highly unstable, changing in 56.4% of week-to-week comparisons and holding steady for the entire window in only 85 of 480 series.

These are not contradictory. They describe two different layers of the same system, and understanding the difference is the single most useful thing a company evaluating its AI presence can take from this study.

Two separate questions, two separate answers

The first layer is presence: does an engine's answer ever mention this company at all, anywhere in the response, for a given prompt. This is the layer where the cohort does well almost universally, with 100% appearance rates on transactional prompts and similarly high rates across informational, navigational, and commercial-investigation prompt types.

The second layer is precedence: of the companies mentioned, which one is named first, the position most likely to be read, clicked, or acted on. This is the layer that turns over constantly. A company can be present in essentially every relevant answer and still rarely or never be the one named first, and which company holds first place this week is a poor predictor of which company will hold it next week.

Why the two layers behave so differently

Presence is a lower bar. An engine generating a helpful answer to a commercial question will typically list several plausible companies, because that is what makes the answer useful, comprehensive, and defensible as a recommendation. The threshold for being included in that list appears to be relatively easy to clear across this cohort, which is consistent with why 95.9% of companies clear it.

Precedence is a much narrower and more contested outcome. Only one company can be first. Small differences, in how content is retrieved that week, in how the model happens to sample its response, in what specific angle the prompt takes, can tip which of several roughly-equally-plausible companies gets named first, without any of the others dropping out of the answer altogether. That is consistent with a wide field of eligible companies rotating through the top slot, rather than a stable hierarchy with one dominant company and everyone else absent.

What this means for strategy

A company chasing first place should recognize it is competing in the narrower, noisier layer, where week-to-week outcomes are close to a coin flip and durable wins are rare (85 out of 480 series). A company worried about whether it appears in AI answers at all is asking a question this cohort has already mostly answered: presence is common. The more productive place to spend effort is making sure presence is maintained consistently across prompt types and engines, since that is the layer that is achievable and durable, rather than fixating on holding a first-place position that turns over for most companies most of the time.