Image ads need visual review, not just OCR

The words can be present and the ad can still mislead

Image ads create a simple trap for compliance review: the required words are technically present, so the asset appears to pass.

But customers do not experience an image as a text transcript. They see hierarchy, size, contrast, colour, imagery, spacing and the relationship between the benefit and its qualification.

A bold promise at the centre of an ad and a faint risk line at the edge do not necessarily create a balanced impression simply because both can be extracted by OCR.

That is why image review must examine the creative, not only the words inside it.

Prominence is part of meaning

Regulated marketing frequently depends on information being sufficiently prominent.

The FCA's FG24/1 social-media guidance says promotions across advertising channels should be fair, clear and not misleading, provide a balanced view of benefits and risks and support consumer understanding.

In a visual asset, balance is expressed through design as well as language.

Reviewers should ask:

• Can the risk information be read at the size and placement in which the ad will appear?
• Is the qualification close enough to the claim it changes?
• Does the benefit dominate the asset so completely that the warning feels incidental?
• Is important text likely to be cropped by the platform?
• Does a button, arrow or image create urgency or certainty beyond the written claim?
• Does the visual imply an outcome the copy avoids stating directly?

These are not aesthetic preferences. They affect what the customer is likely to understand.

Imagery can make its own claim

An image of luxury, safety, family security, effortless growth or environmental benefit can carry a commercial message without expressing it in words.

The same is true in gambling advertising, where personalities, sporting references and visual style can affect appeal to children and young people. The CAP Code's gambling rules are designed around social responsibility and protection from harm; the people and imagery used in the ad can therefore be central to review.

A text-only process can miss this entire layer.

The intended format matters

The master creative is not always the asset the customer sees.

Platforms crop images, compress text and apply interface elements. A square post may become a vertical story. A mobile placement may make a warning unreadable. An agency may export several sizes from one source file, each with different spacing.

Review should capture the intended channel and, where the differences are material, the actual variants.

Approving a desktop layout does not automatically approve every mobile execution derived from it.

What useful image review should return

A useful finding should point to the visual element behind the issue.

For example:

• the warning is present but materially less prominent than the return claim
• the image implies capital security that the product does not provide
• essential terms are likely to be cropped in the story format
• the featured personality creates a strong-appeal concern
• the call-to-action obscures a material eligibility condition

The treatment should be practical: move the warning, change the hierarchy, replace the image, narrow the claim, adjust the crop or create a separate channel variant.

Preserve the creative in the record

If the review record stores only extracted text, it cannot show what the reviewer actually saw.

The asset, its format and the relevant visual finding should remain connected to the decision. That becomes important when the campaign is reused, resized or challenged later.

A simple test for compliance software

Give the platform an ad where every required word is present but the qualification is visually overwhelmed.

Then ask:

1. Did it identify the hierarchy problem?
2. Did it point to the relevant part of the image?
3. Did it explain why the overall impression matters?
4. Did it suggest a usable design change?
5. Did the exported record preserve the image reviewed?

If the product only returns the extracted text, it supports a document check. It has not yet demonstrated visual compliance review.