What each verdict means
How to read Verified, Probably helped, No effect found, Went down and Not proven, the range under the number, and a line such as “98% sure this fix helped”.
The five verdicts
Each finished change gets one of five verdicts. The rules are fixed in advance and are the same for every business. A “clean comparison” means there was something to compare with, your answers were not already drifting in the weeks before, and no other change to the same answers went live in the same weeks.
Verified: assistants named you more often after the change, and the whole 95% range is above zero, with a clean comparison. This is verified lift.
Probably helped: the evidence leans towards a rise but is not strong enough for Verified. For example, the range is above zero but there was no clean comparison.
No effect found: the whole 95% range is within 10 points either side of zero, with a clean comparison. The change made little or no difference to what assistants say.
Went down: the whole 95% range is below zero, with a clean comparison. Assistants named you less often after the change.
Not proven: anything else. Usually the range is too wide to call, or there were fewer than 20 answers before or after the change on the questions and assistants it was aimed at.
On one business, most changes end as Not proven unless the effect is large. Not proven means “we cannot tell”, not “it failed”.
Read the number and the range
Open Proof and find a finished change, one whose pill shows a verdict. Then:
Read the large number, for example “+18 points”. It is how much more your answers moved than the comparison answers moved in the same weeks, in percentage points of the share of answers that name or cite you.
The number (1), its range (2), the how-sure line (3), the two tracks (4) and the evidence (5). Demo business, captured October 23, 2026. Read the line under it, for example “How often AI names you · 95% range +11 to +26”. The true effect is very likely somewhere in that range. Fewer answers make the range wider.
Read the how-sure line, for example “Over 99% sure this fix helped”.
Compare the tracks Your answers and Comparison answers. The grey dot is the share of answers that named or cited you before the change and the blue dot is the share after, on a track from 0 to 100%. If there is no Comparison answers track, there was nothing to compare with, and the number is just your own before and after.
Read the points underneath. They say what the change was compared with and anything that weakens the result.
In the example, assistants named the business in 28% of answers before the change and 45% after, while the comparison answers stayed at about 40%. The change gets the credit for the rise. 45 minus 28 is about 17 points; LegitGeo works it out from the exact shares, which gives +18.
What the how-sure line means
The how-sure line restates the 95% range as one number: given the answers collected, how likely the effect is to be on that side. It appears only on a finished verdict with a clean comparison.
Find the verdict pill, here No effect found.
The verdict (1) and its number, range and how-sure line (2). Demo business, captured October 23, 2026. Read the how-sure line beside the number. Its wording depends on the verdict.
“98% sure this fix helped”: the chance that the change made AI name you more often. Below 50% it reads, for example, “30% chance this fix helped”.
“97% sure this change made it worse”: shown with Went down.
“Over 99% sure the effect is small, within 10 points either way”: shown with No effect found.
A Verified result is always over 97.5% sure. The line never says 100% or 0%: the ends read “Over 99%” and “Under 1%”.
The points under each result
These sentences say what the result rests on. The ones you may see:
“Compared with …”: what your answers were compared with. It can be your own answers on assistants or questions the change does not reach; “similar businesses that changed nothing”, meaning your other locations or businesses on the same platform as you; or other tracked businesses on the same assistants, which Proof uses when you have none of those.
“Nothing to compare against yet, so a move that hit every business cannot be ruled out.”: there was no comparison, so the verdict can be at most Probably helped. This is common for a change aimed at every assistant and every question.
“No drift in the weeks before the change.”: your answers and the comparison moved alike before the change, which a fair comparison needs.
“Your answers were already changing before this went live, so the change cannot be credited with confidence.”: they were drifting apart before the change, so it cannot be Verified.
“Another change to the same answers went live in the same weeks. Their effects cannot be told apart.”: two of your changes overlap, so neither can be Verified.
“Answers moved … for everyone in these weeks, likely a model update. That part is subtracted.”: the comparison moved too, and that part is taken out of your number. See When an AI update moves everyone’s results.
“There is not enough history before the change to check that you and the market were moving alike, so this cannot be verified yet.”: the comparison is other businesses, and there were too few answers before the change to check it.
“OAI-SearchBot (OpenAI) re-read the changed page on …”: when each AI crawler fetched the page you changed. It needs the page address on the change and AI Crawlers set up.
“Based on 280 answers before and 280 after, across 10 question and engine pairs.”: how much evidence there is.
“Marked done. The next readiness check will confirm it is live.”: shown on a Diagnosis fix and on an Action Center action. Here the readiness check means running Diagnosis again. For a Diagnosis fix, select Run again on the Diagnosis page once the fix is live; if that run no longer finds the problem, the line becomes “The next readiness check confirmed the fix is live.” after the next nightly update. An Action Center action keeps the first line, because no check follows it.
Not enough answers and Already named every time
Not enough answers, with a line such as “0 of the 172 answers per window needed to see a 15-point change”, means there is not enough to measure yet. Right after a change goes live this is normal, because the four weeks of answers after it have not started. If it is still there at the end, the verdict is Not proven.
Adding questions does not help a change that is already being measured: a question added now has no answers from the four weeks before the change. For your next change, track more questions about what you plan to change at least four weeks before you make it. See Add, edit or pause the questions you track.
Already named every time means assistants named you in almost every answer before the change, so there is no room for a rise to show.
Frequently asked questions
My visibility score went up. Why is my change Not proven?
Your score moves for many reasons, including model updates that move every business. Not proven means the rise could not be told apart from those with enough confidence. See Why the numbers differ between pages.
Can a verdict change later?
A finished verdict stays as it is. It changes only if you select Measure now on it, which works it out again (for example after you leave out an overlapping change), or if you mark its Diagnosis fix done again, which starts a new measurement.
What is a percentage point?
The difference between two percentages. Going from 20% of answers to 30% is a rise of 10 points, even though 30 is half as much again as 20.