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

# Drift labels

> What it means when a day is marked because an AI engine changed how it answers.

AI engines change: a new model, a different way of searching, a habit of naming more brands or fewer. A change like that moves every brand at once. On a chart it can look as if you gained or lost ground when nothing about you changed. Citino calls this **drift**, and labels the day it happens.

## What we look for

Every night, for each engine, we compare that day's answers to your questions with the week before. We label the day when at least half of the brands that engine names change order at once, unless the websites it cites changed by half or more that day too: that is the market moving, not the engine, and the day is not labelled. We also label the day when the engine starts naming noticeably more or fewer brands per answer, whatever the websites do.

* **We look only at your own answers.** No other workspace's data is used.
* **Your first few days are never labelled.** There is no week before them to compare with.

## Where you see it

| Screen | What it shows |
| - | - |
| Overview, **Visibility over time** | A small grey diamond above the day, explained in the chart's legend, and a count of such days beside the chart. |
| Prompt detail, **Who gets named** | Each drift day in the window, with its date and engine. |
| Gaps and Source gap | How many days the engines drifted in the window, "Labelled, not removed", and a reminder that each loss is read against the leader on the same answers. |
| Objections | Which engine changed how it answers, on which day, and that every answer still counts. |
| Experiments | A **Drift in these windows** panel on each experiment. |
| Claims | Drift days on a claim's timeline, with the engines they were on. |

## Labelled, never dropped

We never remove a drift day's answers from any number. Leaving out the days that do not suit the story is cherry-picking, and you could not trust a figure built that way. So:

* **The day stays in every figure**, and the chart's line runs through it.
* **Your move is told apart from the engine's by comparing with your competitors.** An experiment's result is your change minus the competitor baseline: "The median change across the competitors you track, over the same questions and the same days. It is roughly what would have happened to you anyway."
* **Alerts do the same.** When a drift day falls in the period an alert looks at, it judges your change against the median change of the competitors you track, and its headline says the engine changed how it answers.

## The other marks on the chart

Visibility over time can carry one other mark, read the other way:

* **A teal dot: you published a change.** It marks the day you did something, so you can read a rise against it.

## What to do when a drift day lines up with a big move

1. **Check whether your competitors moved the same way** on the Brands screen. If everyone moved, it was the engine.
2. **Read the answers** from that engine before and after the day, on the Answers screen.
3. **Do not report the move as yours** until it shows against your competitors too.

## Related

> [card] [How drift is detected](https://docs.citino.ai/guides/methodology/drift.md)
>
> The method, in more detail.

> [card] [Within noise](https://docs.citino.ai/guides/within-noise.md)
>
> Why a small change is not a move.

> [card] [Experiments](https://docs.citino.ai/guides/screens/experiments.md)
>
> Measuring your change against your competitors.

> [card] [Share of voice](https://docs.citino.ai/guides/share-of-voice.md)
>
> The number a change in brands per answer moves first.


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