AI insights

How AI insights work

In short

Copilot reads your creative, its heatmap and every score with its comparison value, then writes a summary, a three-part analysis and prioritised recommendations. It explains the numbers but never changes them; treat its suggestions as expert advice to test, not as results.

What it is

The Insights tab is where Foresight Copilot explains your results in plain language. For every analysis it writes:

• Insight: a short overall summary of how the creative performs and why. You can copy it with one click.
• Detailed Analysis: three blocks covering Emotional Impact, Attention & Clarity and Mental Demand, each shown next to the relevant scores.
• Recommendations: a prioritised list of concrete changes. Each one shows the metrics it targets and a priority (High, Medium or Low). Expand it to read the suggested action and the effect it is expected to have.

For videos, Copilot also analyses each scene. A recommendation can be tagged Overall or point to a specific scene and time range, and you can play that scene directly from the list.

What feeds it

Copilot works from the same material you see on the results page:

• every score, plus the comparison value from the benchmark you selected;
• the short description of the creative's content and, for videos, the scenes and their scores;
• for images, the predicted attention heatmap;
• the context you gave when uploading: the objective, and the creative type and sector (images) or platform (videos).

For videos tested on a platform such as TikTok, the platform is passed as context, so recommendations can take its format conventions into account. For images, insights are not platform-specific, even when the image is shown inside a platform screen.

How it's built

Copilot explains what your scores mean for this creative and suggests what to change. A few rules keep it grounded:

• Your scores are the source of truth. Copilot quotes the values from your analysis and is instructed never to invent numbers or fill in values that aren't there.
• Bigger gaps come first. The further a score sits from its comparison value (for Mental Demand, from the typical level), the higher the priority of the related recommendations.
• Brand-safe suggestions. Copilot avoids advice that would damage your brand or core idea.
• Objective-aware. If you picked an objective, the analysis is framed around it and you get fewer, more strategic recommendations.

Insights are written in English and translated automatically into the language you use Foresight in.

Scale and direction

Insights are text, not scores. The only scale is the priority on each recommendation: High, Medium or Low.

Image vs video

	Images	Videos
Summary, detailed analysis, recommendations	Yes	Yes
Uses the attention heatmap	Yes	Uses the video and scene scores
Scene-by-scene analysis and scene-specific recommendations	n/a	Yes
Context passed to Copilot	Creative type and sector	Platform

For videos, Copilot may comment on elements beyond what the scores measure, such as the soundtrack. Sound is not analysed directly; music and sound effects do not affect the scores.

How to read it

• Start with the summary, then use the detailed analysis to see which part of the creative (emotion, attention, complexity) explains it.
• Work through recommendations from High to Medium to Low priority. The metric chips tell you which score each change is meant to move.
• Treat the expected effect of a recommendation as an AI estimate, not a forecast. The only way to know the real effect is to analyse the revised creative.
• If you didn't select a benchmark, no comparison values are shown. Without a comparison group, insights are less reliable about how scores compare with typical creatives.

Common pitfalls

• Reading insights as measurements. Insights are AI-written interpretations of predicted scores. Use them like advice from an experienced colleague: a good starting point, to be checked against your brief and brand.
• Expecting insights to change the scores. They never do. Neither does the objective you pick.
• Waiting for insights straight away. Insights arrive a little after the scores. If they haven't appeared after a few minutes, reload the page later.
• Looking for insights on old analyses. Analyses made before AI insights were introduced open normally and keep all their scores, heatmaps and benchmark comparisons, but show "This study has no AI insights". To get insights for an old creative, upload it again as a new analysis.
• Insights that failed. The Insights tab says "We couldn't generate the AI insights" and offers Regenerate insights (run them again) and Report a problem (tell the Foresight team; you can send another report after a short wait). Scores, heatmaps and benchmark comparisons are complete either way. For videos, the scene breakdown comes from the same AI step, so the Scenes tab stays empty until the insights are generated successfully.