How to Tell If a Video on Bluesky Is AI Generated in 2026

How to tell if a video on Bluesky is AI generated in 2026? First, accept the hard truth: Bluesky gives you almost no help. Unlike Threads or TikTok, it has no built-in AI label, no detection badge, and no provenance display. A test of ten social platforms in October 2026 found exactly that — “X and Bluesky offer no AI label for posts” — and the tester had to write the disclosure into each post by hand. The work falls on you: your eyes, the account, and a quick check outside the app. Videos can now run up to ten minutes.

Step one: understand what Bluesky does — and does not — label

Bluesky runs on the AT Protocol, where moderation works as a layer on top of the network. Its Community Guidelines (last updated September 19, 2025) contain no general rule requiring AI content to be labelled. What they do ban is deception: the “Be Authentic” principle forbids content intended to misrepresent, defraud, trick, deceive, or mislead, and the Trust and Transparency section bans financial scams and phishing.

The guidelines are specific where AI causes direct harm: they ban AI-generated sexual content involving minors, non-consensual sexual content including synthetic versions, and sharing tools that create intimate images without consent — including “deepfakes” and “nudify” apps. Impersonation, handle-squatting, and identity churning are also banned. But an ordinary AI video — a fake speech, a fabricated event — is not automatically labelled or removed; it is only actioned if it crosses one of those harm lines.

One partial exception: community labelers. Anyone can run a labelling service on the AT Protocol, and some flag AI-generated images for subscribers. These are opt-in, third-party, and not applied by default — and we could not verify that any reliably covers video.

Visual and audio tells that still work on Bluesky

With no label to lean on, watch the video twice — once for the story, once for the details. Hands, teeth, and text are still the most reliable tells: fingers that merge or bend wrong, teeth that blur into one shape, and signs or logos with misspelled or warped letters. AI motion often glides too smoothly — people walk without weight and the camera drifts unnaturally. Faces flicker across cuts: ears, glasses, and hairstyles that change shape between seconds are a strong signal.

Check physics and sound: shadows pointing in different directions, reflections that do not match the scene, and backgrounds warping as the subject moves all point to AI. Narration that is too even, with no breaths, plus lips drifting out of sync, should raise suspicion. Finally, judge the story: a “breaking news” clip no outlet covers is suspect. One flag is a curiosity; three or more, plus no source, means do not share it.

The 30-second verification routine

1. Read the post text and the profile (10 seconds). Honest creators often disclose AI use in the post text or bio, because the platform gives them no label for it. Check the handle: Bluesky lets a domain you control become your handle, so an account posting from a matching domain carries more weight than a lookalike handle. A new account with a flood of viral clips and no history is a warning sign.

2. Search the claim elsewhere (10 seconds). Search the key claim on the web. If a dramatic event appears only on Bluesky — no news site, no official account — it is probably fabricated. Real events leave traces.

3. Run a detector on a downloaded copy (10 seconds to start). If you can save the video, upload it to Google’s SynthID Detector at synthid.com, which opened to everyone on October 7, 2026 and checks for invisible SynthID watermarks in content from Google, OpenAI, Nvidia, and Kakao. A hit confirms AI involvement; no hit proves nothing. You can also run the clip through https://aivideodetector.site/ as a second signal. For the same routine on labelled platforms, see our guides to AI video on Threads and X.

The scam angle: what AI videos on Bluesky are used for

Bluesky’s smaller, trust-heavy community is fertile ground for impersonation: fake accounts copy a real person’s name and avatar, then post AI videos of that person “announcing” a crypto giveaway or endorsing an investment. The video is the hook; the scam lives in the reply, the bio link, or the direct message. Scams are banned under Bluesky’s guidelines, but removal happens after the fact.

Never send money, click a wallet link, or share personal details because a Bluesky video told you to. Verify the claim on the person’s official website or another platform first, and report impersonators through the post menu.

Honest limits: what we could not verify

No detector is 100% reliable, and on Bluesky the gaps are wider because there is no label system to fall back on. We could not verify any Bluesky support for C2PA Content Credentials or other provenance display as of November 2026, nor whether Bluesky strips metadata on upload or download — so do not assume a downloaded file keeps its original provenance. We also could not verify that any community labeler reliably flags AI video. A clean result means “no AI signals found,” not “this video is real.” The strongest evidence is outside the video: a credible source, a consistent account, and independent confirmation.

Frequently asked questions

Does Bluesky label AI-generated videos?

No. As of November 2026, Bluesky has no built-in AI label — a test of ten platforms in October 2026 confirmed it offered none — so disclosure depends on the poster writing it in the post or bio.

Does Bluesky ban AI-generated content?

Not generally. Its guidelines ban deceptive content and specific AI harms — sexual content involving minors, non-consensual synthetic intimate images, and deepfake or “nudify” apps — but no verified rule requires AI videos to be disclosed.

Can I check a Bluesky video with SynthID Detector?

Yes, if you can download the file — upload it at synthid.com. A detected watermark confirms AI involvement from a supported tool; no watermark found does not prove the video is human-made.

How long can a Bluesky video be?

Up to ten minutes and 300 MB, one video per post, since August 26, 2026 — three minutes before that, and 60 seconds at launch in September 2024.

The bottom line

On Bluesky there is no label to save you — no badge, no provenance display, just the video and the account behind it. Read the post and profile, search the claim, and run a downloaded copy through a detector before you believe or share a dramatic clip. A realistic video with no disclosure, no source, and no corroboration is one you should not trust. For a detection tool to use as one signal, visit https://aivideodetector.site/.