How often does the top-mentioned company actually change from one week to the next

A company that appears first in an AI engine's answer this week wants to know one thing above all: will it still be first next week. This study answers that directly by tracking 480 separate weekly series, one for each combination of 160 prompts and 3 engines, across a run window from 2026-04-27 to 2026-08-03, with a median of 13 weeks of observations per series.

Across all adjacent week pairs in all series, there were 5905 total transitions observed. Of those, 3328 showed a different top-mentioned company than the prior week. That is a leader change rate of 56.4%, meaning the complementary figure, the rate at which the same company held first place across consecutive weeks, is 43.6%.

That number is close to a coin flip. It means that knowing a company was first this week gives you barely more than even odds of correctly predicting it will be first again next week, averaged across the whole cohort and all three engines. This is not a claim that any individual company's position is random. Some companies may hold a durable lead on a specific prompt and engine while the aggregate churns elsewhere; the 56.4% figure is a population-level average, not a per-company guarantee. A company evaluating its own results should look at its specific prompt-engine series rather than assume the average applies uniformly.

Why the rate might be this high

Three explanations are consistent with the data, and the data collected here cannot fully separate them:

1. Retrieval drift: the content each engine draws on when generating an answer changes week to week as the web changes, so the ranking genuinely does move.
2. Generation-side sampling: language models sample from a distribution over outputs rather than deterministically repeating the same answer, so identical retrieval could still produce a different named company.
3. Prompt-level sensitivity: small variations in how a recurring prompt is phrased or contextualized when re-issued could shift which company gets named first, even without any real change in the underlying market.

Distinguishing these three would require sending the identical prompt multiple times within a single week, to isolate pure sampling variance, and comparing that within-week variance to the week-to-week variance reported here. That comparison was not part of this study's design and is a clear direction for follow-up work.

What this rate does not mean

It does not mean the engines are unreliable at surfacing plausible companies. Appearance rates across prompt types are consistently high, 100% for the 4355 transactional prompts and 100% for the 823 commercial-investigation prompts in this dataset. The instability is concentrated in which single company gets the top slot, not in whether a reasonable set of companies gets mentioned at all.

What follows practically

For a business tracking its own AI visibility, the immediate implication is to stop treating any single week's top-position result as a fixed fact. A drop from first place to third place in one week's read, given a background leader-change rate of 56.4%, is close to what would happen by default even if nothing about the company's actual standing changed. The more informative signal is a trend sustained across several consecutive weekly reads, not a single comparison.