How Holdout works

Did your brand improve, or did the AI change?

Holdout checks how AI assistants describe and recommend your brand. It helps you see whether a change is about your brand, a wider shift, or simply not clear yet.

Three useful comparisons

We do not trust one score by itself.

Your brand over time

We ask the same questions before and after a brand makes a change.

Your brand against similar brands

We check whether similar brands moved too, even though they made no change.

Each AI assistant on its own

ChatGPT, Claude, Gemini, and Perplexity can disagree. We show their results separately.

A simple example

A bigger number does not always mean the work succeeded.

Your brand rises by 14 points. Similar brands rise by 10 points at the same time. That leaves 4 points that may be linked to your change—but the honest answer can still be “not clear yet.”

Your brand+14

Similar brands+10

Difference+4

The answer

Holdout says what the evidence supports.

Likely improvementYour brand moved more than the similar brands.

Likely declineYour brand moved down after the wider shift was removed.

Not clear yetThere is not enough evidence to call the change real.

Nothing hidden

Every result can be checked.

Holdout keeps the AI assistant and model, the questions asked, the date, the comparison brands, the sources, and the steps used to reach the result. If something important changes, Holdout starts a new comparison instead of quietly joining unlike results.