Meta political ads: disclose photoreal AI video and realistic audio
Meta's rules for social issue, election and political ads require disclosing photorealistic AI images, video and realistic audio. What counts in a Sume clip.

Meta holds social issue, election and political ads to a stricter AI rule than ordinary ads. According to the Meta ad standards page, read on 2026-10-05, advertisers must disclose when such an ad contains a photorealistic image or video, or realistic-sounding audio, that was created or edited with third-party generative AI tools.
The wording points at the output, not the vendor: it is about realism, and it names third-party tools. A clip generated through the Sume video API is made with a third-party tool from Meta's point of view.
What triggers it
| Ad content | Disclosure needed? |
|---|---|
| Photorealistic AI video of people or scenes | Yes, per the page |
| Realistic-sounding AI voice or audio | Yes, per the page |
| Photorealistic AI image | Yes, per the page |
| Obviously stylised animation | Not described on the page; judge by realism |
What to do with a Sume clip
If the ad is in a covered category, assume a realistic generated presenter, crowd or street scene needs the disclosure. Use Meta's ad-creation disclosure control for it, and keep a record of the clip's origin.
For records, a Sume job carries a metadata field that Sume stores and does not send to the provider. Put a line there such as meta_siep_ai_disclosed: true, and keep the job id with the ad name. That makes it easy to answer a review request later.
Sume's docs do not say that clips carry a platform flag or credentials, so the disclosure has to be set in Meta's own ad flow.
Edge cases
Rules like this change; the date I read the page was 2026-10-05, and I only describe what it said then. Sume's docs describe a metadata field on video, image and music jobs that Sume stores on the job and does not send to the provider. I found nothing in the docs about Sume adding, keeping or removing C2PA credentials, watermarks or platform labels, so treat that as undocumented and check the delivered file.
- A real recording with a generated voice-over still contains realistic-sounding audio.
- A real spokesperson clip with a generated background may count as edited with third-party tools; err toward disclosing.
- If you are unsure whether an ad is a social issue ad, read Meta's definitions on the same page before building the creative.
Sources
Related posts
More in Use cases
- Meta text overlay rules: one CTA, contrast, safe zone, Sume captions
Meta's overlay advice is four lines: clean font, contrasting hue, one call to action, stay in the safe zone. How to meet each with a Sume caption job.
- Midjourney 100K-image history crash fixed; archive your files anyway
Midjourney's Oct 2 changelog fixes a crash opening images in 100K+ histories. Keep your own copy: download outputs and log prompts, settings and cost.
- Midjourney collaboration is 'upcoming'; hand off with Sume metadata
Midjourney's Oct 2 changelog lists collaborative generation and edit history as upcoming. On Sume, put a handoff note in the job metadata field.
- Mixed-language recording: detach ranges, STT with a hint per range
Split a bilingual video into audio ranges with Sume audio detach, then transcribe each range with its own language_code. Detach output caps at 900 s.
Written by Sume