Guide · 3 min read

Why AI visibility changes from week to week

Movement in the Index can be noise, a change in measurement conditions or a genuine shift. This guide explains how to tell which.

Generative answers vary

Ask an AI assistant the same question twice and you may get two different lists. That is a normal property of generative models, not a fault in measurement. It means that any single week's reading for a brand contains some random variation.

The Index reduces the effect by publishing visibility over a rolling window of several weeks rather than a single run. Each new week enters the window and the oldest leaves. A brand's figure moves only as far as the new week differs from the one it replaces, which smooths out much of the noise.

Even so, small movements happen constantly. A brand moving a place or two, or a fraction of a percentage point, in a single week is usually best read as part of that variation rather than as news.

Changes in measurement conditions

Sometimes the conditions themselves change. An assistant may move to a new default model, the Index may adopt that model after an overlap period, or a category's question panel may be revised from a stated week. Each of these is logged and shown on history charts and on the methodology page.

When a movement coincides with one of these markers, it should be read with that context. The Index does not claim that a model change caused a particular brand's rise or fall, but it makes sure the reader can see that the comparison spans different conditions.

Categories can also be held when a week's collection fails the quality guards. A held category carries forward its last good results with a notice, and no weekly changes are reported for it until a new edition passes.

Genuine shifts

Some movements are sustained. A brand that climbs steadily over several weeks, across more than one assistant, is more likely to reflect a real change in how it is discussed than a brand that jumps once and falls back.

With web search switched on, assistants often draw on current online material. Product launches, pricing news, prominent reviews and coverage of service problems can all change what is written about a category. The Index records the outcome in answers but does not attribute it to any specific cause.

How to judge a movement

A useful checklist: is the change larger than the brand's usual week-to-week range? Has it persisted for more than one edition? Is it visible on more than one assistant? Does a model-change marker or a held week sit in the same period? Is the answer count behind the figure large enough to trust?

Brand pages show the history chart and per-assistant figures together for exactly this reason. The weekly archive lists entrants, exits and movers for each published edition, comparing ranks with the preceding stored published week.

Movers are reported only among brands with meaningful weekly visibility, so that a small brand moving from rank 140 to rank 90 on a handful of answers does not crowd out more substantial changes.

What the Index will not do

It will not invent a change where data is missing. If a week is absent or held, the history shows a gap rather than a zero. If a brand was not ranked in the previous edition, it is marked as a new entrant rather than given a movement figure.

Corrections follow the same principle. If a published week is found to be flawed, it is withdrawn and replaced with a corrected edition, and the replacement is recorded. Movement figures always compare published editions as they currently stand.

Live evidence · In the week to 13 November 2026 (week 47, published 16 November)

This week's biggest movers

The largest recorded rank movement in each published category compared with the previous published week.

This week's biggest movers
CategoryBrandMovementVisibility
Banking and current accountsmoney.co.ukDown 14 places5.4%
Investing and wealthNvidiaUp 16 places5.3%
Mortgages and personal financeL&C MortgagesDown 4 places17.3%
Business finance and paymentsUltimate FinanceUp 11 places5.3%
Cars and motoringParkersUp 16 places6.4%
Fashion and clothingThe IndependentUp 11 places5.3%

Full definitions and limitations are in the methodology.