AI image and layout tools produce work fast, and the visual habits of the models that generate it show up in predictable places once you know where to look. Spotting these tells does not require training in design. It requires knowing where to point your attention.

Patterns that repeat across tools

Text inside AI-generated images is often the giveaway: letters that almost form words, kerning that drifts, or a headline that reads correctly at a glance but falls apart under a second look. Icons and illustrations from the same set frequently mix styles, a flat icon sitting next to a shaded one, because each was generated on its own. Gradients and glow effects turn up more often than the brief calls for, because they are a cheap way for a model to add visual interest without solving a layout problem.

Consistency failures across a set

A single AI-generated image can look convincing. A set of them rarely holds together, because each generation is a separate guess. Check for:

• Colour values that shift slightly between pages that should share a palette
• Spacing and margins that vary between components doing the same job
• A logo or mark that changes proportion or weight from one placement to the next
• Corner radii, shadow depth or line weight that differ across buttons or cards that should match

A visual identity is a small number of rules applied consistently across everything a business puts out, and consistency is what generative tools have no mechanism to enforce across separate outputs. A written design system is what stops this drift on human-made work; without one, each new asset is a fresh guess whichever way it was produced.

Details a trained eye checks

Hands, hardware and small mechanical objects are still where most image models struggle, so a hand holding a pen, a pair of glasses, or a piece of furniture with visible joinery is worth a second look. Symmetry is another tell: faces, logos and icons that should be perfectly mirrored often are not, by a margin small enough to miss on a quick scan but visible once you hold a straight edge against them. Drop shadows falling in different directions within the same image suggest the elements were composited.

None of this makes AI-generated work unusable. Some of it is a starting point worth refining, and recognizing that is the judgement a client needs before signing off a set of assets for print or for a live site.

Accessibility is not a visual style choice, it is a set of technical requirements, and AI tools trained mainly on how designs look tend to fail the same checks over and over. Knowing the common failures means you can check for them without needing to be an accessibility specialist yourself.

Contrast

AI-generated palettes often look fine on screen but fail the contrast ratio a screen needs to be readable for someone with low vision, particularly pale grey text on white or light-coloured text on a mid-tone background. This is checkable: a contrast checker tool gives you a pass or fail number against the recognised threshold, and it takes under a minute per colour pairing.

Type sizing and hierarchy

AI output frequently sets body text too small, or uses size as the only signal of what is a heading and what is not. A screen reader (software that reads a page aloud for someone who cannot see the screen) depends on the underlying code marking a heading as a heading.

Semantic structure

This is the one non-designers most often miss, because it does not show up on screen at all. AI-generated interfaces regularly produce a page that looks correctly structured but is built from generic containers with no real heading tags, no landmark regions and no logical reading order underneath, which means a screen reader announces it as an undifferentiated block of text. You cannot judge this by looking at the design; you have to look at the code or run it through a screen reader.

Alt text

Alt text is the written description attached to an image so a screen reader user knows what it shows. AI-generated designs either omit it, or generate something generic like "image" or a restated file name that tells the user nothing about the actual content. Every image that carries meaning needs a specific description; a purely decorative image needs to be marked as decorative so it is skipped.

A checklist before accepting AI-generated work

• Run every text and background colour pairing through a contrast checker and note the ratio.
• Check that headings are marked as headings in the code, and that the reading order matches the visual order.
• Check that every meaningful image has specific alt text, and that decorative images are marked as decorative.
• Check that body text is a legible size and that the design does not rely on colour alone to convey information, such as an error shown only in red.
• Test keyboard navigation: can every interactive element be reached and used without a mouse?

None of these checks require special software beyond a free contrast checker and, ideally, a screen reader trial run. What they require is knowing to look, since AI output that fails every one of them can still look, at a glance, like a finished design.

If a design has already gone out and you are not sure whether it meets these points, that is a design refresh: an assessment and correction pass on existing work. You can start a design job and describe what exists and what you are unsure about, or read how to brief a designer for what to include so the scope is clear from the outset.