Prove
Impact
See every change you made and its measured effect on your visibility at 4, 8 and 12 weeks, adjusted with a control group so market-wide moves are not credited to your work.
Impact answers the question every report eventually gets: did the work actually move the number? Every change Rixe knows about — a page edit, a listing that went live, published content, a finished action — is recorded with its date. Rixe then compares your visibility before and after that date on the same questions, at 4, 8 and 12 weeks, and nets out what moved for everyone.
Where to find it
Actions › Action plan, then the Impact tab (https://app.rixe.ai/actions?tab=impact). Inside it, Timeline shows the interventions and their effects, and A/B tests is covered in A/B tests.
Where interventions come from
You do not log interventions in a separate place; they are picked up from the work itself:
| Kind | Recorded when |
|---|---|
| Page edit / Technical fix | You mark work on a page: Analytics › LLM Analytics, Pages tab, open a page, and use Mark under Work completed (what was done, the date, an optional note). Schema work counts as a technical fix. |
| Source listing | An outreach task reaches Live with its proof link (Sources, outreach and PR). |
| Content published | Content you created in Rixe reaches published (Content). |
| Technical fix / Other | An item on the action plan is marked done. |
The date that counts is stamped once and never moves, so re-marking the same work does not reset the clock.
How the effect is measured
- Equal windows. For each intervention, Rixe takes the same number of weeks before and after the date — 4, 8 and 12 weeks — on the questions the work targeted.
- Mention rate before and after. In each window, the share of measured answers that name your brand, with a Wilson confidence interval on the "after" rate and a 95% interval on the change itself.
- A verdict. A change counts as real only when it is both statistically significant (p < 0.05) and at least 10 points. Either test alone produces results nobody can act on: huge samples make trivial moves "significant", and small samples make noise look dramatic.
- A control group. The same two windows are measured on answers the work could not have touched (see below), and the net effect is your change minus the control's change.
Impact is computed every night at 05:30 UTC. The first window closes four weeks after the change, so a new intervention shows First measurement is done 4 weeks later.
Verdicts
| Verdict | Meaning |
|---|---|
| Real change | Significant and at least 10 points. |
| Change too small | The move is under 10 points. |
| Cannot be told from noise | Big enough, but not statistically significant. |
| Cannot tell yet | Not enough measurements in the window (fewer than 4 runs on either side). This does not mean the work failed — we do not know yet. |
The control group
A before/after comparison on its own credits your work with whatever else happened in those weeks: an engine update, a seasonal swing, a competitor's launch. The control strips that out.
| Control | Used when | What it is |
|---|---|---|
| Control: the site's other questions | The work targeted specific questions. | Your own other questions, same brand, same engines, same weeks — only the work differs. |
| Control: competitors on the same questions | The work was site-wide, so no question is untouched. | Your competitors' mention rate on the same questions stands in for "the market moved". |
Each effect then shows control moved N pts and Net effect N pts (low to high), with its own verdict: Significant against control, Too small against control, Not distinguishable from control, Not enough data for a control or No control group.
Read the net effect first
If your visibility rose 12 points and the control rose 10, the work earned about 2 — and that is usually inside the noise. The net effect is the number to put in a report.
Reading the screen
- Measured — interventions with at least one window closed. Significant — those with a statistically significant effect. Worked — the share of significant effects that went up.
- An insight line names your most effective intervention.
- Intervention history — each intervention with its kind, date and a link (Open the page) where relevant, followed by its 4, 8 and 12-week results: the change in points, the change range, the "after" range, the number of runs and the control comparison.
- What actually works — what each kind of work returned across organisations in your market: Median effect, Positive share and Sample, at a given Window. It is anonymous and needs at least five organisations; your own result is shown beside it when you have one. If you have not tried a kind of work yet, it says so.
- Market benchmark — appears once enough organisations in the same market have data; it is never built from a single account.
Limits and plans
Impact measurement is in every plan. It uses only measurements Rixe already collects, so it costs nothing extra. See Plans and limits.
Common questions
Impact is empty. Nothing has been recorded yet. Mark a change on a page, take an outreach task live, publish content or finish an action-plan item.
It has been two weeks and there is no result. The first window closes at four weeks.
The verdict is "Cannot tell yet" after 12 weeks. The targeted questions have too few measurements. Check that they are active and measured on enough engines (Prompts).
Two changes on the same page in the same month — which one gets the credit? Both are measured from their own dates, and their windows overlap, so neither can be separated cleanly. If you need a clean answer about one change, run an A/B test.
Related
Still stuck? Write to [email protected] and a person will answer.