Platform guide

Trends and history

How to read the trend chart, what the range and granularity filters do, how series breaks are marked, and what the action pins mean.

By the AI Native team · Updated 2026-06-11

The Trends page plots recommendation rate, mention rate, and sentiment across every live scan of a product. It answers the question: did your work move the number? See Scans and results for what each metric means before reading this page.

Range and granularity filters

Two filter strips sit at the top of the page. The range strip selects how far back to look: 7 days, 30 days, 90 days, or all time. The granularity strip groups data points by day, week, or month. Both filters apply to the chart without a page reload. The right combination depends on scan frequency and how long you have been measuring.

Reading the chart

Three series appear on the chart:

Recommendation (green) - the share of answers for this product where the AI actively recommended it, expressed as a percentage.

Mention (cyan) - the share of answers where the AI mentioned the product by name, whether or not it recommended it.

Sentiment x100 (amber) - the average sentiment score across branded answers, multiplied by 100 so it plots on the same axis as the rate percentages.

Each data point represents one scan aggregated to the chosen granularity. Data labels appear on the recommendation series. The y-axis is labelled in percentage points for all three series.

Action pins

Cyan vertical markers appear on the chart wherever an action was shipped. Each pin is labelled with the lever type. Below the chart, the "Actions shipped" section lists every pin with its lever, the affected query cell, and the run number. This makes it possible to connect movements in the series to specific actions.

Series breaks

When the product's prompt set changes between scans (prompts added, removed, or re-classified), an amber dashed vertical line appears on the chart at the run where the change happened. The break is labelled with the version change, for example "v2 to v3."

A break means the two sides are not directly comparable because the prompt set changed. The platform warns you explicitly rather than silently plotting points that use different denominators. Compare runs within the same version segment.

KPI tiles

Four summary tiles appear above the chart. The first tile (highlighted in green) shows the most recent recommendation rate. The remaining tiles show the most recent values for the other key metrics. These are point-in-time snapshots of the latest scan rather than period averages.

Questions

Why does my chart show fewer points than I expected?

The granularity setting groups scans by time bucket. If you ran three scans in one day and chose weekly granularity, those three scans appear as one weekly data point. Switching to daily granularity shows each scan separately.

Does the trend include demo scans?

No. Demo scans are excluded from the trend chart. Only live scan results are plotted. This keeps the chart from mixing real measurements with synthetic fixture data. See Scans: demo mode vs live mode for how modes stay separated.

What does a series break mean practically?

It means the set of prompts changed between those two runs. Rates before and after the break are calculated over different prompt sets, so a jump in the chart at a break point may reflect the prompt change rather than a real change in AI behaviour. Check what changed in the prompt library when you see a break.

Can I see which action caused a movement?

Action pins mark the run where an action was recorded. If the recommendation series rises after a pin, that action is a candidate cause. The Experiments page provides a more formal before-and-after comparison for actions you freeze as experiments.

Why do all three series use the same axis?

Recommendation and mention are percentages (0 to 100). Sentiment is normally between -1 and 1, so it is multiplied by 100 to plot on the same scale. A sentiment x100 value of 45 corresponds to a raw sentiment score of 0.45.

What is the difference between mention rate and recommendation rate?

Mention rate counts answers where your product appears by name. Recommendation rate counts only the answers where the AI actively recommended your product. Recommendation is a subset of mention. A product can be mentioned in a list without being recommended.

How far back does "All time" go?

All time shows every live scan that has ever been run for the product since it was added to the workspace. There is no fixed cutoff.

Back to Platform guide or the documentation hub.