How it works
Fatal Recall is a simulated day-after recall test, not a real one. Here is exactly what happens under the hood, what the score means, and where the limits are.
Each market panel contains eight AI consumer personas. Every persona has a defined demographic, a media diet, and a way of speaking that matches the market they represent.
They think and respond in the language of their market. A Slovak panel writes in Slovak, with Slovak references and media habits. A US panel writes in English, with US references and media habits. The panel is not a single model answering eight times; it is a structured cast of simulated consumers.
Personas are designed to disagree. Some notice visuals first. Others remember claims. Some are skeptical of advertising. That disagreement is what makes the transcripts readable and stops a single viewpoint from dominating the result.
Each persona sees the creative once, for five seconds, in the language and media context of their market. No brand cues, no questions in advance.
A simulated overnight decay period lets salience compete with forgetting. Strong cues consolidate; weak cues fade the way real memory does.
Personas are asked what they remember, without seeing the creative again. Their answers are captured as free-recall transcripts, not forced choices.
Each transcript is checked against the creative's actual components. What survived is retention; what disappeared is leakage.
Did the persona remember who the ad was for, unprompted? This is the hardest and most important survival signal.
Did the memory attach to the right product or service category, or did it drift somewhere irrelevant?
Did the core claim, promise, or offer survive the night, even if the brand name did not?
Did the creative leave a unique trace, or is it interchangeable with any other ad in the category?
Did the ad leave a feeling, tone, or reaction that the persona could still report after the decay period?
The panel configuration is frozen and versioned. Every report carries the panel version it was run against, so scores are comparable over time.
Changing a model, a persona definition, or a scoring rubric creates a new version. We do not silently re-score old reports against new panels. What you ran is what you keep.
Current panel: v1.0
Before a panel version is released, it is run against paired creatives whose relative recall outcome is already known, for example the same execution with and without the brand present. The engine has to put them in the right order. This tests direction, not absolute accuracy. Published results for the current panel version are in preparation.
This is not a replacement for a recall study. It is a fast prediction built on proven memory principles: encoding bottlenecks, salience decay, free recall vs. recognition. Results reflect probability, not certainty.
Run a simulated day-after recall test on your own ad. Get a Memory Score, retention and leakage diagnostics, and readable persona transcripts.
Run your first autopsy free