AI Chat Search Interest

Methodology

Google Trends rescales every export so that the largest value in the export equals 100, which makes separate exports incomparable. We combine six overlapping worldwide exports covering eight AI chat products and estimate one scale factor per export by weighted least squares in log space, using every term that appears in more than one export as a bridge. Dividing each export by its factor puts all series in one common unit; the result is re-anchored so the strongest month overall (ChatGPT, October 2025) equals 100. A round-trip check reproduces 98% of the raw export cells within rounding error.

Ambiguous names contaminate raw search data, so each series is cleaned before publication. A series starts at the product’s public launch, and a constant background estimated from the pre-launch average is subtracted across the whole series — removing, for example, searches for “kimi” as a given name, the racing driver Kimi Räikkönen, and the 2022 film, or “claude” as a first name. Non-positive residuals and the current in-progress month are excluded; positive residuals are retained so valid low-signal months remain visible.

The y-axis is logarithmic because the series span three orders of magnitude. On a log scale, parallel lines mean equal growth rates, and a doubling covers the same vertical distance anywhere on the chart.

Frequently asked questions

Why do the series start on different dates?

Each series begins at its product’s public launch. Search volume recorded before launch measures something other than the product — given names, films, common words — so it is removed, together with a constant pre-launch background that would otherwise inflate ambiguous terms after launch. Positive residuals after that subtraction are retained to avoid artificial gaps.

Can Google Trends values from different exports be compared?

Not directly: every export is rescaled so its own maximum equals 100. We estimate one scale factor per export from the terms shared across exports and divide it out, which places all series in one common unit before the final re-anchoring to 100.