Proof Loop
The Proof Loop is how RankEcho closes the loop: after you ship a fix, it re-tests the exact same prompts on a schedule and reports the observed movement — citations gained, rank, and share of voice. Because AI answers shift on their own, a re-test against a fixed baseline is the only reliable way to tell whether the work moved the brand into the answer or whether it was noise.
More: Intro to RankEcho · How it works · Full walkthrough · Walkthrough PDF
Track the prompt you fixed
After shipping, track the exact prompt the fix addresses. RankEcho records the baseline - how engines answered before - so later movement has a fixed point to be compared against.
Scheduled re-tests run per engine
The Proof Loop re-asks the same prompt on schedule across engines. The proof grid shows one row per engine and one column per run: green means cited, the amber line marks your ship date.
Get the citation alert
When an engine starts citing you on a tracked prompt, RankEcho emails you the observation with the run details - the moment your fix lands is a dated, documented event.
Read the before and after
The proof page reports baseline versus post-ship citation rate with observation counts and a confidence label. Low sample sizes say so plainly - correlation is not oversold as causation.
Why a loop, not a snapshot
A single audit tells you where you stand once. AI answers are non-deterministic and change as the web and the models change, so a one-off result can mislead in either direction. The Proof Loop re-runs the same prompts over time, turning a snapshot into a trend you can trust.
How it works
The discipline that makes proof meaningful:
- Lock the prompt set and engines so runs are comparable.
- Record a baseline before any change.
- Ship one fix at a time so movement can be attributed.
- Re-test on a fixed cadence and record the before-and-after.
Measurement
RankEcho reports correlation with controls — a fixed prompt and a strict pre/post split — not causation, and it scores confidence by sample size. When the signal is weak or the answers are volatile, it says so rather than claiming a win. The goal is evidence you can defend, not a number that flatters.
Proof you can show
The output is a before-and-after you can put in front of a stakeholder or a client: here were the prompts you were missing, here is what shipped, and here is the observed movement. That is the deliverable that justifies the work — and the part monitoring-only tools leave you to assemble by hand.
Frequently asked questions
On a fixed cadence you set — weekly for most brands, tighter while actively shipping fixes — so movement is separated from day-to-day variance.
It reports correlation under controls — a held-fixed prompt and a strict pre/post split — with confidence by sample size, rather than claiming causation.
Then RankEcho says so. Low confidence is reported as low confidence rather than dressed up as a result.