AI person in a Meta ad: is the label next to Sponsored?

Only for ads made with Meta's own generative tools. Meta's newsroom post says that when those tools include an AI-generated photorealistic human, the AI info label appears next to Sponsored instead of behind the three-dot menu (read 2026-10-03). For ads made with outside tools, Meta describes detection and a label in About this ad, not the Sponsored line.

So an avatar made in Sume and uploaded as an ad should not be expected to show a label beside Sponsored. If you want viewers to see one, put it in the creative yourself.

What exactly does Meta document?

Meta's post separates placements. For ads created or significantly edited with Meta's generative AI creative tools, the label appears in the three-dot menu or next to Sponsored. If those tools include an AI-generated photorealistic human, it appears next to Sponsored. No label is applied when there is no significant edit and no photorealistic human. Meta said it would share its approach for non-Meta tools later, and the June 2026 update says it will begin detecting third-party AI ads through industry-standard signals (read 2026-10-03).

Why add your own label if Meta may detect it?

Detection depends on signals in the file, and Meta's own help page says not all ads will have AI info while its tools roll out (read 2026-10-03). A silent avatar ad that nobody labeled is exposed in two places: Meta's system may miss it, and viewers have no cue. Laws are moving in the same direction. Our posts on California's synthetic performer rule and New York's cover when an avatar needs a disclosure in the ad itself.

A burned-in line is the one method that survives download, re-upload and cropping of the surrounding page, because it is part of the picture.

How do you burn a disclosure into the video with Sume?

Sume's video captions endpoint accepts authored cues, which skip speech-to-text. Per the docs, a cue has text, start and end, and start and end range from 0 to 60 seconds. The source must be a public HTTPS video URL. Send an idempotency key so a retry does not run twice. The request below burns a line across the first 4 seconds:

curl -X POST https://api.sume.com/v1/video-captions \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: ad-disclosure-line-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/example/ad.mp4",
    "style": "black-outline",
    "cues": [
      {"text": "AI-generated person", "start": 0, "end": 4}
    ]
  }'

What are the limits of that approach?

Cues are limited to the first 60 seconds of a video, so a longer ad only gets a line in that window. The wording is yours; Sume does not decide whether it satisfies a law or a platform rule. A standalone caption job is a fixed-price job per the docs, so confirm the figure in the catalog before a batch.

Also remember it burns text onto the picture. If the ad has speech captions as well, run the disclosure pass last so a later caption pass does not cover it.

• Pick wording that says what is generated: person, voice, or whole scene.
• Keep it on screen long enough to read, and away from platform UI overlays.
• Save the artifact URL of the captioned file as the version that ran.
• Self-disclose again in the ad's primary text if the format allows.

What belongs in the ad record?

Keep the original file URL, the captioned file URL, the cue text and timing, and the job ids of both steps. If an ad is later questioned, you can show the exact line that ran and when. For a ready list of fields, see the per-clip disclosure record.

Does a burned-in line replace the platform label?

No. A line in the picture is your own disclosure; the platform label is Meta's. They serve the same viewer, but neither substitutes for the other, and Meta applies its labels by its own process. Treat your line as the floor and Meta's label, where it appears, as a bonus.

Also decide where in the clip the line goes. A line in the first seconds is seen by everyone who stops scrolling. A line only at the end is missed by most. Check the placement against the platform's own interface elements, since captions and buttons can sit over text near the bottom and sides of a vertical frame.

Finally, keep the wording honest and specific. A line that says AI-generated person tells the viewer something; a vague tag does not.