Accessibility checks for AI-generated campaign media

Start with the job the campaign image performs

Do not ask an image model to write the alternative text and treat the result as approved. First place the final exported asset in its real campaign context, then identify what a reader must understand or do if the pixels are unavailable. WCAG 2.2 Success Criterion 1.1.1 requires a text alternative that serves the equivalent purpose for non-text content, while content that is purely decorative must be implemented so assistive technology can ignore it. The same generated visual can therefore need different treatment when it is a product claim, a linked call to action, or atmosphere behind nearby copy.

Write from the approved message and the delivered image, not from the generation prompt. A prompt records intent; the export may omit, distort, or add details. Describe the information that survives review and matters in this placement. Do not identify a person, location, product feature, certification, interface state, or measured outcome unless the campaign owner has verified it. If nearby live copy already communicates the image's full purpose and the visual adds no meaning, an empty alternative can prevent repetition. If the visual carries meaning, provide a concise equivalent rather than a list of colors and objects.

Keep campaign copy live whenever the medium permits

Separate exact words from generated pixels before approval. Headlines, prices, dates, offer terms, disclaimers, and calls to action should remain editable text in the page, email, or ad fields when the destination supports that presentation. WCAG 2.2 Success Criterion 1.4.5 says to use text rather than images of text when the technology can achieve the presentation, with exceptions for customizable images and cases where the particular presentation is essential; logotypes are identified as essential. A brand logo does not justify flattening the rest of the campaign message into the image.

For a generated key visual, export a clean plate and place approved copy in the delivery layer. That keeps the words available for correction, localization, resizing, and destination-specific truncation. If a platform forces a poster-like image, repeat material copy in accessible nearby text and include text shown only in the image in its alternative. The W3C alt decision tree distinguishes image text repeated nearby, image text used only for visual effect, functional text, and text unavailable elsewhere. Record which branch applies instead of reusing one generic description across every placement.

Inspect the exported pixels at every intended crop

Run contrast checks on the delivered wide, square, and vertical files, not only on a design-system swatch. WCAG 2.2 Success Criterion 1.4.3 sets at least 4.5:1 contrast for ordinary text and 3:1 for large-scale text, subject to its exceptions, and says threshold values must not be rounded. Generated backgrounds can vary beneath a single line, so sample the weakest relevant area. Add a stable solid panel or scrim, move the live copy, or reject the crop when any required word loses contrast. Recheck after compression and platform cropping because the approved composition may not be the served composition.

Meaningful icons, chart marks, product-state indicators, and other graphical objects needed to understand the content have a separate check. WCAG 2.2 Success Criterion 1.4.11 requires 3:1 contrast against adjacent colors for graphical objects needed to understand the content, except where a particular presentation is essential. Do not infer a pass from a text-contrast result. Inspect each required visual cue, and provide the information in nearby text when a generated diagram cannot be repaired without changing its meaning.

This research pass did not generate or download an OfflineCreator campaign asset, inspect a source prompt, measure pixels in a real delivered export, or test a crop in a live ad placement. The checklist is grounded in current primary accessibility guidance, but no page-specific asset is claimed to pass. A future editorial review should attach the exact exported files, dimensions, contrast measurements, alternative text, destination previews, and reviewer decision before this route is considered verified.

Choose checks by media role, not by how it was generated

Use an empty alternative only for a genuinely decorative image or one whose meaning is fully redundant with nearby real text. Use a brief, context-specific alternative for an informative photograph or simple graphic. Describe the destination or action when an image is the only understandable content of a link or control. For a complex chart or diagram, provide the important information elsewhere rather than forcing all detail into one attribute. These decisions come from the image's purpose in context, which is why a single asset library description is not an adequate campaign-wide answer.

Extend the gate for motion. WCAG 2.2 requires captions for prerecorded audio in synchronized media unless the media is a clearly labeled alternative for text, and captions include meaningful non-speech audio as well as dialogue. Moving, blinking, or scrolling content that starts automatically, lasts more than five seconds, and appears alongside other content needs a way to pause, stop, or hide it unless the movement is essential. A silent generated loop, a narrated product concept, and a decorative animation therefore require different review paths; the word “AI-generated” neither creates nor removes those accessibility duties.

Decorative visual

Hide it from assistive technologyUse this only when removing the image would not remove information or an understandable action.

Informative visual

Describe its purpose in contextBase the alternative on the approved message and the inspected export, not on prompt history.

Motion with audio

Review captions and visual informationCheck the delivered media and player behavior rather than assuming an automated track is complete.

Run an accessibility gate from brief to placement

At briefing, label every required message, action, claim, and piece of text that must remain exact. During concept selection, reject layouts that depend on unreadable lettering or visual details that cannot be described honestly. Before export, create the live-copy version, decide whether each image is decorative, informative, functional, or complex, and draft its alternative from the approved purpose. For motion, inventory dialogue, meaningful sound, important visual-only action, autoplay, duration, flashing, and player controls.

At output review, compare the prompt, approved brief, full-resolution result, each crop, and the final destination preview. Check every visible word, test text and meaningful graphical contrast, read the alternative without looking at the image, and verify that it does not duplicate nearby copy unnecessarily. On LinkedIn, current first-party help says authors can add alt text to feed and article images, and its current single-image ad specification exposes optional image alt text for screen-reader users. Platform support is not proof of good copy: manually inspect what was saved in the actual campaign record.

At release, store the asset version, dimensions, destination, nearby live copy, alternative text, caption file or transcript where relevant, measurements, reviewer, and decision. Reopen the live placement after publication because crops, truncation, or media processing can change the reader's experience. Stop release when the reviewer cannot explain the image's purpose without the pixels, when required words exist only as unreliable generated lettering, when a contrast sample falls below its applicable threshold, or when synchronized audio lacks a reviewed equivalent.

Move to the next task only after the failed check is named

Return to the use-case directory when the asset still needs a production path. Use the client-deliverable packaging guide when the accessibility record, crop set, and approval notes need to travel with a review package. These links address the reader's next operational decision; they do not replace the image-purpose, live-copy, contrast, caption, and placement checks on this page.

If a failure comes from unreadable generated lettering, rebuild the words outside the image rather than polishing the alternative text around a visual defect. If the failure comes from an inaccurate product depiction or unsupported claim, return it to the claim owner. If the destination has no usable way to carry the needed alternative, choose another accessible presentation or provide the equivalent information in the surrounding content and record the limitation.

Keep this page inside its evidence boundary

This page owns the narrow task of checking AI-generated campaign media for context-specific alternatives, live text, text and graphical contrast, basic motion controls, captions, and final-placement failures. It does not certify WCAG conformance, provide legal advice, guarantee that a platform preserves accessibility fields, or claim that generated media is inherently more or less accessible than media made by another method.

The required recent-evidence run returned 91 items across arXiv, Digg, GitHub, Hacker News, Instagram, Reddit (partial), Techmeme, TikTok, YouTube, and web grounding; Polymarket returned no results and X was not configured. Zero items met the relevance floor for this page's accessibility-check topic; ten were marginal (EU AI labeling, AI-in-advertising commentary, accessibility metadata in other products); ten were promotional; two were duplicates; one was unsafe; and one was unverifiable. The page therefore makes no claim about common practitioner behavior, sentiment, adoption, accessibility outcomes, or comparative product quality. Current primary W3C and LinkedIn sources support the documented checks, while real output review remains an explicit evidence gap. Keep the route draft, unreviewed, researched, and unpublished until an editor reviews an actual campaign asset and the remaining coverage limits.