Short answer: Instagram and Facebook label AI images by reading data stored inside the file, not by looking at the picture. In our tests, every AI image that still had that data got labelled, and every copy with the data removed did not. Pinterest labels some AI images, but rarely. TikTok labels some AI photos, yet in our test it didn't label even untouched AI images posted from a computer.

We tested this ourselves in September and October 2026, using our own accounts and real files from six AI image generators. Here's what we did, what each platform did, and what it means for you.

What we tested

We made one image each in Gemini, ChatGPT, Meta AI, Midjourney, Grok and Google Flow, and kept the files exactly as downloaded. For each one we also made a cleaned copy with AI Metadata Cleaner, which removes the AI data and then re-reads the new file to confirm nothing is left.

Then we posted originals and cleaned copies side by side. Every test post was private ("Only me" or "Only you"), and we left each platform's own "AI label" switch off, so any label you see below was added by the platform itself.

Instagram: labels from the file (9 of 9)

In September 2026 we posted 9 images to Instagram:

  • Originals with AI data (Gemini, Meta AI, ChatGPT): all three labelled.
  • Cleaned copies (Gemini, Meta AI, ChatGPT, Midjourney): none labelled.
  • Two AI images that arrived with no AI data at all (from Google Flow and Seedream, never touched by us): neither labelled.

Those last two are the important ones. They are plainly AI images, yet Instagram didn't label them, because their files carried nothing to read. We checked again after 3 minutes, 25 minutes and about a day; nothing changed.

Facebook: same result, within seconds (3 of 3)

In October 2026 we posted three pairs to Facebook, each an original and its cleaned copy:

Image Original Cleaned copy
Gemini Labelled "AI content" No label
ChatGPT Labelled "AI content" No label
Meta AI Labelled "AI content" No label

The originals were labelled the moment they were posted. Facebook, Instagram and Threads all use Meta's labelling, so this is what we expected, and now it's measured.

Pinterest: labels some AI images, rarely

Pinterest shows an "AI modified" tag in the corner of a pin when it decides the image was made or edited with AI. The tag only shows on the pin's own page, not in the search grid.

  • We opened 10 obvious Gemini AI portraits from a Pinterest search. 1 had the "AI modified" tag; the other 9 had nothing.
  • We checked the 10 most recent pins on our founder's own Pinterest account, all AI images cleaned with our tool before pinning. None had the tag.

That fits the same pattern, but it's weaker evidence than Instagram and Facebook: we didn't post matched pairs on Pinterest, and we can't see what was in other people's files.

TikTok: inconclusive

TikTok does label some AI photo posts. Public Gemini photo posts in search show "Contains AI-generated media" under the caption.

But when we posted our six test images as photo posts from TikTok Studio on a computer, none was labelled after about 40 minutes, including the Gemini and ChatGPT originals that still carried their AI data. Because the originals weren't labelled either, this test can't tell us anything about cleaning. TikTok may treat posts from its phone app differently, or not label private posts. We'll update this section if we test from the app.

What each AI generator writes into your image

This is what the files we downloaded on October 9, 2026 carried. Generators change this often. Google Flow added nothing in September, and by October it was adding Google's Content Credentials.

Generator AI data in the file
Gemini Content Credentials (C2PA) from Google, including a note that a SynthID watermark was applied
ChatGPT Content Credentials from OpenAI naming the image model
Meta AI An IPTC tag saying the image was made by AI, plus a Facebook data block. No Content Credentials
Midjourney Your full prompt, Job ID and Midjourney username, plus an IPTC "made by AI" tag
Grok Content Credentials naming Grok Imagine, plus an xAI signature in the EXIF data
Google Flow Content Credentials from Google, with the same SynthID note as Gemini

Our cleaner removed all of it from all six files. The Midjourney row is worth a second look even if you don't care about labels: anyone who downloads the original file can read your prompt and username.

Does this fool AI detector websites? No

Removing metadata changes what the file says, not the picture. We ran the three originals and their three cleaned copies through an AI detector website that judges the pixels. All six scored 99% AI.

The same goes for invisible watermarks like Google's SynthID, which live in the pixels. The cleaned Gemini copy still has it. What our tests show is that, today, Instagram and Facebook didn't use it to label our posts.

So cleaning helps with the platform labels that read the file. It does nothing against tools that analyse the image itself, and no metadata tool can honestly promise otherwise.

Tips if you post AI images

  1. Post the cleaned copy, not the original. It's easy to pick the wrong file from your downloads. The original still carries the AI data.
  2. Upload the cleaned file directly. Opening it in another editor and re-exporting can write new data back in.
  3. Label it yourself when it matters. If an image is a realistic AI picture of a real person or event, platforms expect you to disclose it, and their "AI label" switch is the honest way to do that. See the legal side of removing AI metadata.

If a label is already on an Instagram post, here's how to remove AI info after posting, and why Instagram says AI info covers retouched real photos.

Frequently asked questions

Does Facebook label AI images automatically?

Yes. In our test, Facebook labelled AI images from Gemini, ChatGPT and Meta AI "AI content" the moment they were posted, even with its own AI label switch off. It reads AI data stored in the file, the same way Instagram does.

Does Instagram detect AI from the picture itself?

Not in our tests. Two AI images that came with no AI data in the file were not labelled, and cleaned copies of labelled images were not labelled either. Meta has said it is working on other detection methods, so this could change.

Does Pinterest label AI images?

Sometimes. Pinterest shows an "AI modified" tag on some pins. In our check, 1 of 10 AI portraits from search had it, and none of 10 AI images cleaned with our tool did.

Does TikTok label AI photos?

Some, yes: public AI photo posts can show "Contains AI-generated media". In our desktop test it didn't label any of our photo posts, including untouched AI originals, so we couldn't measure the effect of cleaning there.

Will removing metadata get past AI detectors?

No. Detector websites analyse the pixels. Both our originals and our cleaned copies scored 99% AI on one. Removing metadata only affects labels that are read from the file.