Measuring impact
Confirmed wins
How the Action Center counts the specific answers you won after acting, and why a mention must hold for two runs before it counts as a win.
The overall visibility number is too noisy to prove that one recommendation worked. What can be proved is which specific answers changed. A confirmed win is one tracked answer that did not name you when the recommendation was written, and named you in both of the two latest runs completed after you moved the card to Done, as of the moment you took the reading.
Why the overall number is not enough
A single run-to-run comparison of one engine's mention rate cannot reliably detect a change smaller than about 28 points, while winning one extra answer is worth about 10 points for a typical brand. Comparing four runs on each side only brings that down to about 19 points, because a brand's numbers drift together from run to run rather than varying independently. No window length closes that gap. (Measured by Ranqo across 81 brands and 851 runs.)
So a real improvement of one or two answers can sit well inside the noise band and read No detectable change in the drawer, even though those answers really did change. Counting the answers themselves is how Ranqo can still show it.
What is counted
A tracked answer is one prompt asked on one engine. A recommendation's scope is the set of tracked answers its metric covers:
- Overall visibility: every tracked prompt on every engine.
- One engine's mention rate: every tracked prompt on that engine.
- One prompt category's visibility: every prompt in that category, on every engine.
When the recommendation is written, Ranqo notes, for each answer in scope, whether the latest completed run named you. When you take a reading, it compares those answers with the two most recent runs that completed after you moved the card to Done. A run that was already in progress when you pressed Done counts as one of them once it completes.
| At the start | In the two latest runs | Counted as |
|---|---|---|
| Not named | Named in both | Won |
| Not named | Named in the latest run only | Provisional |
| Named | Still named in the latest run | Held |
| Named | Not named in either | Lost |
An answer that is missing from a run, because you paused the prompt or stopped tracking the engine, is skipped rather than counted as lost. Nothing was measured for it.
Why two runs
Ranqo measured how often a mention survives to the next run. A mention seen for the first time is still there next run 54.2% of the time, close to a coin flip. A mention seen in two runs in a row is still there 96.1% of the time. So one new mention is reported as provisional, and a win needs both runs. The same rule applies to losses: an answer that drops out once is not counted as lost.
With only one run completed since you moved the card to Done, nothing can be confirmed yet, and every new mention is provisional. A card gets one reading, and it is not updated as later runs arrive, so a reading taken after one run stays unconfirmed. Wait until at least 2 runs have completed before you press Re-measure now.
Where you see it
Open a Done card after taking a reading with Re-measure now. A line under the loop in the drawer reads, for example:
- 2 answers won of 25 tracked.
- 3 answers won, 1 provisional, 1 lost of 36 tracked.
- No change across 25 tracked answers.
- 1 new mention, unconfirmed — one more run decides it. when only one run had completed at the reading.
The line describes the reading as it was taken. It does not change when the next run completes.
When the reading compared two runs, the Lift step of the loop shows the number of answers won in place of the change in points. It counts wins only; the line under the loop gives the provisional and lost answers.
A worked example
A recommendation is measured on ChatGPT mention rate. You track 25 prompts, and when it is written ChatGPT names you in 5 of the 25 answers: 20%.
You act and move the card to Done. After two runs you press Re-measure now. 2 of the 20 answers that did not name you have named you in both runs, and the 5 that already named you still do. The latest run names you in 7 of 25 answers: 28%, a rise of 8 points.
The same reading appears differently in each place:
- The drawer: the Lift step shows +2 answers won, and the line under the loop reads 2 answers won of 25 tracked.
- The Done card: the chip shows +8 pts (+8 in List and Matrix). The chip prints the change in points at any size.
- The notice after the reading: ChatGPT mention rate up +8 pts since you acted.
- Verified Lift: 8 points is inside the overall band of 15 points it applies, so this card adds nothing to the total.
The rise of 8 points is also inside the engine noise band of 20 points, so without the answer counts the drawer would show No detectable change. Both readings are true. The change in points cannot be told apart from an ordinary week, while the two answers you won are specific, and each held for two runs.
What a confirmed win does and does not prove
A confirmed win proves that an answer changed and the change held. It does not prove your action caused it: a competitor's change, a new article about you, or the engine itself can move answers too. Read wins together with what you did and when.
A recommendation with no record of which answers were missing shows the change in points only.