> ## Documentation Index
> Fetch the complete documentation index at: https://docs.rankraft.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Measuring results

> How RankRaft AI decides whether an action moved your numbers, and what each verdict means.

When you click **I've done this**, RankRaft AI freezes where your brand stands on the card's prompts, waits for the engines to catch up, then compares the same prompt-and-engine pairs afterwards against the prompts you left alone.

## The timeline

| Stage | Length |
| - | - |
| **Before window** | The 14 days before you marked it done |
| **Wait** | Technical: 3 days · Content: 10 days · Off-site: 14 days |
| **After window** | 14 days, extended once to 28 if there aren't enough answers |

The detail page tracks progress through **Suggested → You marked it done → Measuring → Result**. The **Verification** panel shows the baseline we recorded at mark time (mention rate, citation rate and average spot, with the target metric tagged **Target**) and when the result is due.

If your scans run too rarely to compare a before and after, marking done says so straight away.

## Verdicts

| Verdict | What it means |
| - | - |
| **Measuring** | The after window hasn't closed yet. |
| **Improved** | Your target metric rose. If it moved across enough prompt-engine pairs while your untouched prompts held still, the card says so. With fewer pairs, it shows you the change but says it can't show your edit produced it. |
| **Declined** | Your target metric fell. The card says whether your other prompts held steady. |
| **No measurable change** | Before and after are within normal day-to-day variation. |
| **Can't separate it** | Something else moved at the same time. Either an engine changed its underlying model during the window, or the prompts this action didn't touch moved the same way, so the change looks like it came from the engine rather than your site. |
| **Changed, cause unknown** | The action targeted every prompt you track, so there's nothing left to compare against. |
| **Not enough data** | The prompt was removed, or the engines returned too few answers after you marked it done. |

Every verdict is written with exact counts, for example: *AI named **your brand** in 7 of 15 answers, up from 1 of 14.* For citation actions it reads *linked to you in*.

## Why you can trust this

* **The comparison unit is the prompt-and-engine pair.** Fourteen days of the same question on the same engine is one noisy observation, not fourteen independent ones, so we don't inflate the sample.
* **Untouched prompts are the control.** They can only *withhold* a claim, never create one. If the engines moved everywhere, we say so instead of giving you the credit.
* **We never say proved, caused or because.** At the sample sizes a single prompt produces, no honest tool can.
* **Results are frozen.** Once a verdict is reached it's stamped with a date and never recalculated.
* **Baselines are snapshots.** The before numbers are recorded at the moment you click, so later changes to your setup can't rewrite them.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.