The short version: Instagram shows the “AI info” label because Meta reads hidden metadata (mostly the C2PA standard) baked into a photo or video by the tool that made or edited it, not because a human at Meta looked at your post. It got rolled out fast because of election-year deepfake pressure and EU transparency rules, and it is far clumsier than most people realise, real photos get tagged all the time. I have watched it happen to my own content and to clients’ content, so this is not theory.
What the label is, in plain English
You post a photo or Reel. Under your caption, or sometimes on the media itself, a small line appears: “AI info.” Tap it and Instagram tells you the content was made or edited with AI. There is no shaming banner, no red flag, just a quiet disclosure. Meta first tried a much louder version of this in April 2024 called “Made with AI.” It lasted about six weeks before the backlash forced a rename.
That backlash matters, because it explains why the current version is softer and why it still gets it wrong. Photographers on X and Reddit started noticing their straight-out-of-camera shots getting labelled “Made with AI” simply because they had used Photoshop’s generative fill to remove a stray bin or smooth out a background, a completely normal editing step that millions of photographers use every week. Meta quietly swapped the wording to “AI info” in May 2024 to sound less alarming, but the underlying detection problem never went away. It is still there in 2026, just wearing a calmer name.
Why Meta started doing this at all
Three separate pressures landed on Meta at roughly the same time, and none of them were really about protecting your feed.
- The C2PA standard went mainstream. The Coalition for Content Provenance and Authenticity, backed by Adobe, Microsoft, the BBC, Sony and dozens of others, built a way to attach tamper-evident “content credentials” to a file the moment it is created or edited. Once camera makers, Photoshop, Midjourney and DALL-E started writing these credentials by default, Instagram simply had data to read that it never had before.
- Election-year deepfake panic. 2024 had more national elections happening in more countries than any year in living memory, and fake images of candidates spread fast on Meta’s platforms. Regulators wanted visible disclosure, not just quiet policy pages.
- EU AI Act transparency rules. Article 50 of the Act, which platforms have been working toward through 2025 and into 2026, requires that AI-generated or manipulated content be marked in a machine-readable way and, where relevant, disclosed to the viewer. Meta operates in the EU. It built one global system rather than a separate one just for European users.
Put those three together and you get a label that exists mainly to keep Meta on the right side of regulators and out of headlines, with your feed experience as a secondary concern.
What triggers the label (it is not a human reading your post)
Instagram is not analysing your face or your grammar and deciding “this looks fake.” It is checking for specific, technical signals:
- C2PA metadata written by the creation tool (Midjourney, DALL-E, Adobe Firefly, Google’s Imagen, and increasingly Canva and Photoshop’s AI features all write this automatically).
- Meta’s own internal classifier, trained to spot the tell-tale texture of certain generative models, which is why some AI images with no metadata at all still get flagged.
- Industry-standard invisible watermarks that some generators embed directly in the pixels, separate from the metadata file.
Strip the metadata (a simple screenshot does this instantly, so does re-exporting through certain apps) and the label often will not fire, even on a fully AI-generated image. That is the uncomfortable bit nobody selling you “AI transparency” content wants to say out loud: the system catches honest people who used a legitimate AI tool for a small edit, while someone deliberately faking something can strip the metadata in five seconds and post it clean. The label is much better at labelling openness than it is at catching deception.
A real example: my client’s “before and after” post
In February 2026 I was working with a coaching client, a skincare brand founder in Manchester, on a before-and-after carousel for her Reels strategy. Nothing dramatic, just a client photo where she had used the AI-powered blemish smoothing tool built into a popular editing app to even out lighting across three frames so the set matched. Standard stuff, the kind of touch-up beauty brands have done for a decade, just done with an AI tool instead of a manual dodge-and-burn.
The post went up with the “AI info” label attached. Within an hour she had three comments asking if the “before” photo was fake and one DM accusing her of misleading customers. She had not generated anything, she had lightly edited real photos, but the label reads the same whether you built the entire image from a text prompt or nudged the exposure by 5%. That flattening of very different actions into one badge is the core design flaw, and it is exactly what caused the original photographer revolt back in 2024. Nobody fixed the underlying issue, they just changed the label’s tone.
We ended up re-shooting without the AI smoothing tool for that particular post, purely because the label was doing more reputational damage than the edit was saving her in time. That is a real cost, not a hypothetical one, and it is worth factoring into how you plan content if you rely on any AI-assisted editing, right down to the caption-writing tools built into Instagram itself.
Step by step: what to do if your post gets an AI info label unfairly
- Tap the label first. It tells you specifically what it detected, metadata, watermark, or classifier match. Knowing which one tells you what to fix.
- Check your editing tool’s export settings. Apps like Adobe Lightroom and Photoshop let you strip C2PA credentials on export if you only used a minor AI-assisted feature and want that reflected honestly, though I would not recommend doing this to hide something meaningfully AI-generated, that is a different conversation with your audience’s trust.
- Use Instagram’s “request a review” option under the label if you believe it is a straight misfire. It sits in the same menu as the label explanation. Response times in early 2026 have been running around 3 to 5 days for most accounts I have seen this tried on, not instant, so plan your post timing around that if the label is time-sensitive.
- Address it in your caption if a review is not worth the wait. One honest line, “lightly AI-edited for lighting, nothing about this photo is generated,” does more for trust than silence does.
- Decide, in advance, which of your workflow steps you are comfortable being visible. This is the practical shift most creators still have not made. If you use AI for scripting, editing, captioning, or thumbnail touch-ups, assume it will surface eventually, either through a label or through someone noticing, and build your content plan around that rather than being caught out.
Where this is heading through 2026
Expect the label to spread rather than shrink. Meta has already extended similar disclosure to Facebook and Threads, and given how many Reels get uploaded to Instagram every single day, the volume of AI-touched content running through classifiers only grows. If you want the full picture of how much of Instagram’s traffic is Reels, AI-assisted content, and everything else driving the platform right now, the Instagram statistics breakdown for 2026 is worth a read alongside this.
My honest read, after watching this play out with several clients: the label will keep getting more sensitive, not less, because regulators are pushing for stricter machine-readable disclosure, and Meta would rather over-label than get fined. That means more false positives on ordinary edits, not fewer. If your content strategy leans on AI tools anywhere in the pipeline, from scripting to editing to captioning, build in a habit of checking your posts after publishing rather than assuming they went up clean. And if a label does land somewhere unwelcome and you are unsure how to fix a setting or dispute it, the general Instagram help guide covering every fix and setting in one place is a faster route than digging through Meta’s own support pages.
None of this means avoid AI tools. It means stop treating disclosure as optional or accidental. Put the same energy into deciding what you are comfortable being tagged for as you put into, say, working out where to add a clickable link in your Instagram Stories, it is a small technical decision with a real effect on how people experience your content, and it deserves the same five minutes of planning.
If your question is a different Instagram one, the Instagram guide lists every answer I have written.
Frequently asked questions
Does the AI info label mean my post will get less reach?
Meta has not published any evidence that the label itself affects distribution, and nothing in the platform’s ranking documentation ties reach to the label. The reach damage I have seen is entirely about audience reaction, comments and DMs questioning authenticity, which then affects engagement and, indirectly, how the algorithm treats that post.
Can I remove the AI info label from an already-published post?
Not directly. You can request a review through the label’s own menu if you believe it was applied wrongly, but there is no toggle to switch it off yourself. The more reliable route is preventing it at the export stage by understanding what your editing app writes into the file before you post.
Why did a photo I did not edit with AI get labelled anyway?
Some phone cameras and editing apps now write C2PA credentials automatically the moment you take or touch a photo, even for basic features like portrait mode or auto-enhance, without asking you. Check your camera app’s settings for anything mentioning “content credentials” or “AI features” and you will often find it was switched on by default.
Is Instagram the only platform doing this?
No. YouTube, TikTok and LinkedIn have all introduced comparable AI disclosure labels since 2024, largely driven by the same EU AI Act transparency requirements and similar rules taking shape in US states. Expect the labelling language to differ slightly by platform while the underlying detection method, reading embedded metadata, stays broadly the same.
