Banner and display ad analysis

Banner and display ad analysis predicts how a display creative will be seen in the moment it appears next to content the viewer came for. This article explains how to run the analysis and what each part of the output shows for display creative.

RUNNING THE ANALYSIS

Open AI Creative Insights and upload the banner as a .jpg or .png at the exact size it will serve at. The analysis runs and opens the results when it completes.

A display set is not one creative. A 300x250, a 728x90, and a 160x600 built from the same artwork have different layouts and different relationships between elements, so each size is uploaded and analysed separately. The results will not agree with each other, which is why each size needs its own run.

WHAT THE ANALYSIS RETURNS

Fogmap

The fogmap shows which parts of your banner are noticed and which are missed. It applies a visibility threshold to the gaze prediction: elements that draw enough attention to register stay clear, and elements that fall below the threshold are faded out.

This is the first thing to read on a display ad. What stays clear is what the banner communicates in the exposure a display placement gives. If the brand or the offer falls below the threshold, the creative is not carrying it. See Fogmap interpretation.

Heatmap

The heatmap shows where people pay attention on your banner and how strongly. It is built from predicted gaze points, where each point marks a place the eye is predicted to land, and the points collect together into a map across the creative. Areas drawing the most gaze appear in warm colours, with red at the strongest concentration, running through to cooler colours where attention is lighter.

On display creative the heatmap tells you what is drawing attention away from the message, which is usually an image or a second piece of text competing with the offer. See Heatmap interpretation.

Area of Interest scores

Marking an area of interest gives you scores for a specific element instead of an overall impression. On a banner, mark the logo and the call to action, and add the offer or the price if the creative carries one.

Attention: the percentage of total viewing time spent looking at that element once it has been noticed. It measures how well the element holds someone after they reach it. A high value means they stay on it. A low value means they look at it and move on.

Time to Discover: how long it takes before the element is noticed for the first time. It measures how quickly the element is found, and it is the score that tells you the order elements are reached in. A low value means it is found early. A high value means it is found after other parts of the creative.

Time to Discover carries more weight on display than on most formats. A banner is seen for a moment, so an element found late is an element not found. See Area of Interest (AOI) setup on creatives and Time to Discover.

Design Analysis Report

The Design Analysis Report is a written summary of what the analysis found in the creative. Use it to read the finding directly rather than working it out from the maps, and as a starting point when writing up results for your team.

Design Improvement Report

The Design Improvement Report lists the specific changes the analysis recommends, based on the objective the creative is assessed against. This is the part you act on when you revise the banner. Check the objective shown alongside it first, because a banner assessed for brand recall produces different recommendations from the same banner assessed for clicks. See Prescriptive guidance.

Persona results

If a synthetic audience is applied to the analysis, the results also show how the creative performs for each persona rather than for a general viewer, through Persona Fit, Persona Insights, and the Personas tab. See Using a Synthetic Audience in AI Creative Insights.

COMPARING SIZES AND EXECUTIONS

You can upload multiple creatives into the same study. To compare two versions of a banner, upload both into one study and assess them against the same objective, with the same elements marked as areas of interest. Reading the scores across both tells you which version holds attention on the logo and the offer, and which one reaches them faster.

To understand how a version is likely to perform for a particular group rather than for a general viewer, apply a synthetic audience to the study. This gives you the scores per persona, so you can see whether a version that performs well overall also works for the audience you are buying against. See Using a Synthetic Audience in AI Creative Insights.