AI-powered creative testing predicts attention, memory and emotion directly from a creative asset in minutes, without recruiting a single participant. Traditional eye tracking measures real people's gaze using hardware or a webcam panel, typically over days or weeks of fieldwork.
Dragonfly AI's own models are validated at 91% relative accuracy against a professional eye-tracking panel – closing most of the gap at a fraction of the cost, time and scale limitations.
A traditional eye-tracking study means recruiting a panel, running sessions – in a lab or via a webcam-based panel from a provider such as Lumen Research or Tobii Pro – and analysing the resulting gaze data. That process typically takes days to weeks per study, even before a result reaches the creative team. AI-powered creative testing predicts attention directly from the pixels of the asset itself, with no consumers tested and no hardware involved, so results return in minutes.
Traditional pre-testing is priced per study. Survey-based methods such as Kantar LINK start around $4,500 per asset, and full-service studies – including eye-tracking fieldwork – run $15,000–$50,000 and take days to weeks. That's workable for a hero campaign, but not for the long tail of packaging variants, social ads and e-commerce assets that make up most of what a brand actually produces. AI-powered creative testing is typically delivered as a platform rather than a per-study fee, making it economically realistic to test every asset rather than only the ones with a research budget attached.
Even a professional, "gold standard" eye-tracking panel typically tops out around 40 participants per study, and every additional study means recruiting again. An AI-powered model, once validated, can be run against an unlimited number of assets with no panel to recruit, refresh or replace – the constraint shifts from participant availability to nothing more than how many assets a team wants to test.
This is the question that actually matters, and it's worth being precise about. Dragonfly AI's saliency algorithm doesn't learn from eye-tracking data at all – it simulates the biology of visual attention directly, an approach developed at Queen Mary University of London. Validated against a 40-person professional eye-tracking panel (the industry's own gold standard), it achieves 91% relative accuracy. Measured against the theoretical ceiling of an infinite-participant panel – used because even professional eye-tracking studies never agree with each other perfectly – that figure is 89%. The algorithm has also been independently benchmarked against industry-standard saliency datasets including MIT300 and CAT2000. None of this makes eye tracking obsolete: it remains valuable for qualitative, demographic-specific insight that a universal model isn't designed to capture. The two are complementary, not competing for the same job.
|
|
Dragonfly AI (AI-powered) |
Traditional eye tracking |
|
Speed |
Minutes – no recruitment needed |
Days to weeks – panel recruitment plus fieldwork |
|
Cost model |
Scalable platform, priced for testing every asset |
Per-study fees, from ~$4,500/asset up to $15,000–$50,000+ for a full study |
|
Scale |
Unlimited assets, no panel to recruit or refresh |
Typically capped around 40 participants per study (professional gold standard) |
|
What's tested |
The creative itself – no participants involved |
Real people's gaze, via hardware or webcam panel |
|
Accuracy vs professional eye tracking |
91% relative accuracy (89% vs theoretical ceiling) |
Baseline – the reference standard itself |
|
Best suited to |
Every asset, pre-market, at portfolio scale |
Deep qualitative or demographic-specific insight on a smaller set |