Do AI images get labelled on Instagram and LinkedIn?
· How it works
Meta can add an AI label when it detects standard metadata. LinkedIn's label reads signed Content Credentials only. Social Bee writes an XMP packet.
Sometimes, and the two platforms differ. Meta says it labels images on Facebook, Instagram and Threads when it detects industry standard indicators, naming the C2PA and IPTC standards. LinkedIn shows its Content Credentials icon only for cryptographically signed media. Social Bee writes an XMP provenance packet into every image it encodes, which is the kind of metadata Meta reads.
Take the platform statements first, because they are the part you cannot change.
Meta announced that it would "label images that users post to Facebook, Instagram and Threads when we can detect industry standard indicators that they are AI-generated", and described building tools that can "identify invisible markers at scale", specifically the "AI generated" information in the C2PA and IPTC technical standards (about.fb.com, read on 13 September 2026). So an image that carries a standard AI source-type marker can arrive on Instagram already saying what it is.
LinkedIn's help page is narrower. It states that "Image and video content that is cryptographically signed using C2PA Content Credentials will be noted with the C2PA icon", and that the panel behind the icon shows the assertion about the media, the app or device used, who created it, who issued the credential and when it was issued. The page says nothing about IPTC and nothing about unsigned metadata (linkedin.com/help/linkedin/answer/a6282984, read on 13 September 2026). The practical reading is plain: no signature, no panel.
Now what Social Bee writes, exactly, so you can check it against the platform statements yourself.
Every image Social Bee encodes carries one XMP metadata packet. Three fields name the tool: xmp:CreatorTool, photoshop:Credit and photoshop:Source all read "Social Bee by HYVE Labs". One field names the owner: dc:creator carries the organization that owns the brand, or the brand name when no organization name is known. On a generation whose model is recorded, dc:description names that model, in the form "Generated with", the model name, "through Social Bee by HYVE Labs". And one field states the origin in a standard vocabulary: Iptc4xmpExt:DigitalSourceType.
That last field is the one a platform can act on. A model output carries trainedAlgorithmicMedia. A derivative of a generation, which means a platform export, an Instagram feed autocrop, a crop or an editor render whose lineage reaches a generation, carries compositeSynthetic. A customer's own uploaded photograph, and any derivative of one, carries no source type at all. The IPTC vocabulary defines trainedAlgorithmicMedia as "Digital media created algorithmically using an Artificial Intelligence model trained on captured content" and compositeSynthetic as a "Mix or composite of several elements, at least one of which is Generative AI" (cv.iptc.org/newscodes/digitalsourcetype, read on 13 September 2026).
The rule behind those choices runs in both directions. Nothing Social Bee writes may deny that an image is AI-generated, and nothing it writes may claim a synthetic element in an image that has none. The origin is never guessed from pixels: a generation is recorded as such when it is materialized, and a derivative asks whether its source, or anything that source was derived from, was a generation. A crop of your own photograph therefore carries the credit and makes no source-type claim.
Nothing else is added. EXIF, IPTC and ICC blocks stay stripped, and no metadata is carried forward from a source image. The packet is deterministic: it carries no timestamp, no job id, no counter and no randomness, so the same image encoded twice produces the same bytes.
Where it is applied: at materialization, once per image, in JPEG, PNG and WebP; on the JPEG Meta fetches for a direct Instagram publish; on the feed-ratio rescue crop; on post and Library crops; and on editor resize, retouch and background removal renders. LinkedIn, X and Snapchat receive the stored bytes as they are, so what materialization writes is exactly what those platforms receive. Instagram receives a second encode, which carries the packet as well.
What each platform does with it, stated as what the platforms themselves say.
Meta reads both standards and names the IPTC one, so an image carrying a standard AI source type may show Meta's "AI info" label on Facebook and Instagram. That is truthful and expected.
LinkedIn states that it reads C2PA, and its help page does not mention IPTC at all. Nothing about this packet changes what LinkedIn shows, because the Content Credentials panel reads a signed C2PA manifest only.
TikTok states that it reads C2PA and does not mention the IPTC digital source type, so it states nothing either way about this packet. X has published no statement about either standard.
Two honest limits follow from that, and they matter more than the feature itself.
Social Bee does not sign anything. There is no C2PA library, no certificate and no signing step in the product today. Making LinkedIn's panel read "Issued by HYVE Labs" requires a C2PA manifest signed with a certificate on the C2PA trust list, which means a conformance record for Social Bee as a generator product and a claim-signing certificate in the organization's name, with the private key held in a key management service. A self-signed certificate is refused by the C2PA software, and a test certificate renders as an unrecognised issuer, which is worse for a reader than showing no label at all. So the honest statement today is that LinkedIn's label reads signed Content Credentials only, and Social Bee's packet is XMP.
And no platform is known to display the credit or creator-tool string to a viewer. Those fields exist for archives, customer downloads and any tool that inspects the file. They are not a way to make "Social Bee" appear inside a platform's label, and we do not present them as one.
What to do with this as a brand. Decide your disclosure position yourself rather than leaving it to a platform's detector: if a campaign image is generated, say so in the caption where it matters, because a label you did not control may or may not appear. Keep the original files, since the packet travels with them and answers the question months later.
Image generation is metered: Social Bee reserves an Image Generation Credit before a job and settles it against the delivered result, releasing the reservation on a confirmed failure. Creator is $100 a month with 30 credits, Starter $400 with 150, Growth $1,000 with 400. Creator and Starter start with a 7-day trial: a card at checkout, the first charge on day 8, one trial per business, capped at 2 Image Generation Credits, 20 publishing actions and 2 connected profiles.