Checker › Brand Study

AI disclosure in the wild: 7 brands, 23 ad assets, 0 content credentials

On 26 October 2026 we took public ad creatives from seven major consumer brands — wellness, beauty and beverage names whose ads run heavily on social platforms — and ran them through the same disclosure engine as this site's free checker. The result is not a story about seven careless brands. It is a story about a production step that does not exist yet in most pipelines.

Results by brand (anonymized)

Brands are anonymized by sector. Scores are the engine's own output on the same 100-point scale as the checker: 100 minus 30 per critical finding, 20 per high, 10 per medium, 5 per low.

The most telling detail sits underneath the zeros: every brand in the sample had at least one asset carrying a machine-readable generator indicator in the file's own bytes — the mark of an AI generator, left there by the tool itself — yet in no asset was the generator formally named in the metadata, and nowhere did the formal layer the rules ask for appear: no signed content credential, and no visible label a person could actually see. The signals exist. The disclosure step was simply never added to the pipeline.

Methodology

Sample. Seven consumer brands in wellness, beauty and beverages, chosen because their advertising runs heavily on social platforms. For each brand, the featured creatives on its public storefront (homepage heroes, featured packshots and campaign images) were collected on 26 October 2026 — 23 assets in total.

Engine. The same engine as the checker on this site: disclosure rules v2026.10.2, with the EU AI Act Article 50, Google Ads and China labeling packs applied to every asset. The engine reads declared provenance only — C2PA content-credentials manifests (JUMBF boxes), XMP packets, EXIF fields, and generator names present in file bytes — plus the visible-label state recorded at collection. It never guesses whether an asset was AI-generated; a file with no declared provenance scores as a disclosure gap, not as proof of AI use.

Limits. This is a point-in-time sample of public storefront creatives, not an audit of any brand's full campaign library or its paid placements. A brand may disclose correctly on other surfaces. SynthID-style pixel watermarks cannot be verified by file parsing and were not assessed. Scores use the checker's published arithmetic; the worked examples on the checker page reproduce the arithmetic exactly.

Source: SiteCredly research sample, 7 consumer brands' public ad creatives, scanned 26 October 2026, disclosure rules v2026.10.2. Study published 27 October 2026.

Cite this study

Journalists, researchers and creators are welcome to cite these numbers — that is what the study is published for. Please link back to this page so readers can check the method.

SiteCredly (2026). AI Disclosure in the Wild: a scan of 7 major consumer brands' public ad assets. https://sitecredly.pages.dev/brand-study — 0 of 7 brands carried a C2PA content credential and 0 of 7 attached a visible AI label; the best-scoring asset reached 25/100 (disclosure rules v2026.10.2, scanned 26 October 2026).

Copy the citation

Check your own creative

The same engine is free, runs in your browser, and your file never leaves your device: run the disclosure checker. For the wording that runs with the creative, use the disclosure copy checker. For the rules behind the scores, read the jurisdiction-by-jurisdiction guide.

Frequently asked questions