AI slop is growing on social media, accounting for more than half of all posts on LinkedIn and up to 45% on TikTok, with even more slipping past shoddy auto-label tools. To help users separate the genuine content from the garbage, one company has a powerful AI detector that is meant to bring order to the online chaos, but does it actually work?
Pangram’s AI detection tool is a versatile scanning service that openly outs any piece of content that was put together by bots. While it is being touted for its video detection capabilities, Pangram’s service actually goes even farther. It is designed to identify AI-generated videos, still images, and even text, standing in as a full-fledged AI-busting bouncer for the internet.
AI detector tools haven’t been very reliable in the past, sometimes erroneously marking man-made content as AI.
Pangram’s AI detection tool is only useful if it works. The company claims that it has a low false positive rate of only 0.01% and that it is ranked first by independent researchers. Naturally, we wanted to test these claims for ourselves to see if Pangram actually does what it says on the label.
I tested several different types of content to put it through its paces:
real public images pulled from X to serve as the control variable.real public images from X that were knowingly altered with AI.AI images and videos generated by AI.real text written by me and my own two hands.confirmed AI-generated text written by AI.
Text check
First, I had Grok write a poem about daffodils — something simple and innocuous in hopes that I could skirt by the detector. Pangram returned a verdict of “possibly AI” with 85% certainty. The service suggested that it could be more certain if the poem was longer.
For that, Pangram gets a bit of credit. However, 85% certainty leaves room for doubt, and if you’re looking to use this service to identify real AI slop on the web, you need to be sure if what you’re reading is real or not.
For a second pass, I gave ChatGPT, Claude, Gemini, and Siri the same assignment and ran it all through Pangram. This time, all four samples came back as 100% AI with no deviations. That’s better.
To conclude the text test, I ran my own copy — the first three paragraphs of this article — through Pangram. The service immediately returned a verdict of “human written” with 100% certainty. Now that’s impressive.
RELATED: Real or AI? This new ‘slop detector’ can spare you embarrassment — or worse
Infinite Creations/Getty Images
Image check
Next, I moved on to still images. On the first attempt — which was a confirmed real photo of a woman standing in an autumn scene — Pangram correctly noted that no AI was used on the image whatsoever.
The second photo, which was a slightly altered version of the original image, rightly returned an “AI detected” label with high certainty. It even included a heat map that scanned the image for AI identifiers, using that to further confirm the image’s AI origins.
Video check
Last, I prompted Gemini to create a video of a boat sailing across the East River beneath the Brooklyn Bridge in New York. When I ran it through Pangram, it immediately pointed the video out as AI with 100% certainty. Again, this was correct.
Verdict
AI detector tools haven’t been very reliable in the past, sometimes erroneously marking man-made content as AI. Some of these mistakes have even caused trouble for students who have had to defend their work under false accusations that they didn’t complete assignments on their own.
Pangram, however, appears to be the real deal. I was impressed at how accurate it was at identifying AI-generated content with near-certainty, as well as sorting out the real stuff from the fakery that pervades the internet today.
The only part where it fell short was the poem from Grok. Especially when encountering AI-generated text on social media, posts may not always be long enough to give Pangram enough content for a 100% verdict. That said, Pangram explicitly states on its website that the service is built to identify text created with ChatGPT, Claude, and Gemini, so the fact that it spotted Grok’s linguistic style at all deserves some credit.
How to try Pangram
If you want to test Pangram for yourself, some of its features are available without an account, while its more advanced capabilities — like uploading files and scanning URLs — require you to sign in.
During my test period, Pangram charged credits for each scan — 1 credit for every 100 words checked, five credits to check an image, and one credit for every one second of video. These differ slightly from the official rules on Pangram’s support page, so your mileage may vary. With a 20-credit daily allowance, you’ll burn through these fairly quickly if you have a lot of content to scan.
As an occasional AI checker, though, Pangram is worth keeping around, just in case you want to know if that post you’re reading online is slop or not.
Like Blaze News? Bypass the censors, sign up for our newsletters, and get stories like this direct to your inbox. Sign up here!
Artificial intelligence, Ai slop, Gemini, Grok, Claude, Chatgpt, Tech
