Limitations and responsible use
Foresight predicts how a typical viewer is likely to react to what is visible in a creative. It doesn't hear sound, know your audience or culture, understand the story, or judge whether text is understood. Use it to compare versions and catch issues early, not as the only basis for big decisions.
Foresight gives you a fast, consistent read of how a typical viewer is likely to respond to a creative. Knowing what it does not see is what lets you use it well.
What predictions can't tell you
Your specific audience
Foresight predicts the response of a typical viewer, averaged over many people. It doesn't know your audience's age, culture, familiarity with your brand, mood or intent, and none of these are inputs to any model. A specific person or segment may react differently.
Cultural context
Cultural references, humour and symbols are interpreted from a general point of view. A gesture, colour or joke that lands one way in your market may be read differently elsewhere, and Foresight won't flag it.
Sound
Sound is not analysed directly. Music and sound effects do not affect the scores. Spoken words can be reflected indirectly when the content description mentions them. A video that relies on its soundtrack will be judged essentially on its pictures.
What the text says
The attention model sees text as visual shapes: it predicts whether a headline draws the eye, not whether it is read or understood. The emotion model also uses a short description of the creative's content as context, and that is the only way the meaning of visible text reaches it. Mental Demand reflects visual complexity only. None of the scores tells you whether your message is understood or persuasive.
Story, pacing and narrative
For videos, Foresight analyses frames and scenes. It doesn't judge the story, the pacing or the narrative arc, and overall video values are averages over time, so a short emotional peak in a long video counts for less in the overall value than it does on the emotion timeline.
The real feed
When you pick a platform, the creative is analysed inside a simplified version of the app's screen. That approximates the interface, not the real feed: scrolling, sound-off viewing and competing posts are not modelled. See What changes when you pick a platform.
Individual elements in context
Areas of interest measure the share of predicted attention, not the share of viewers who looked. The reference values for faces, logos and text come from images in general, not your category, so treat comparisons against them as indicative.
Image vs video
Images and videos are analysed with the same models, but videos are analysed frame by frame and averaged over time, and each media type has its own benchmarks. Every score is positioned against creatives of the same media type (see how scores are positioned). Compare images with images and videos with videos. See Image vs video benchmarks.
Using Foresight responsibly
- Compare, don't crown. Foresight is strongest at comparing versions of a creative and spotting issues early.
- Look for big gaps and consistent patterns. Don't over-read small differences between two creatives; focus on large, consistent differences and on patterns that several KPIs agree on.
- Keep humans in the loop for high-stakes decisions. Don't use predictions as the only basis for major decisions; confirm with in-market data or research with real people where the decision warrants it.
- Don't make claims about individuals. Predictions describe a typical viewer, never a particular person.
- Treat AI insights as advice. Copilot's recommendations interpret the predicted scores; they are not measurements. See How AI insights work.