Pangram's biggest flaw is users turning its scores into public shaming
Key Points
AI text detection company Pangram hired a journalist to publicly shame social media users for using AI.
The company later cut ties with him, but CEO Max Spero has continued using Pangram scores to call out alleged AI use on social media.
Pangram only measures whether AI was involved, but the shaming implies the person didn't think or work on their own. Those are two different things.
Pangram hired an "attack dog" to shame people on social media for using AI. Two questions get tangled up in the process: whether AI was used and how it was used. The "AI hunters" accuse their targets of not thinking or working on their own, a claim about the how. But their tool only measures (roughly) the whether.
According to WIRED , journalist Rod Breslau offered his services to Pangram founder Spero earlier this year because he was fed up seeing AI-generated content on social media. Breslau advised Pangram on social media strategy while actively hunting down suspected AI users. "I wanted to take a harsher approach because I thought that people were getting off too easy," Breslau told WIRED.
Breslau's rationale for his campaign is his opinion. "Everyone proclaims to be a leader in their niche, and they build their entire personal brand around that on Twitter and LinkedIn — you have executives, tech CEOs," Breslau said in an earlier interview with GamesBeat . "If you're one of these people, you should have to write your own posts. And if you don't write your own social media posts, then I, or everyone else, should make fun of you." Ad
Pangram later cut ties with him as the company changed direction. It now assumes using AI will become increasingly accepted and wants to detect even light AI editing, positioning itself as the go-to authority for originality checks in publishing, education, and beyond, Spero tells WIRED. But even after dropping Breslau, CEO Max Spero has kept publicly scoring and calling out people based on Pangram results , or as he'd frame it, holding people accountable when they allegedly tried to hide their AI involvement. Ad
Pangram's business model depends on people equating AI use with laziness or dishonesty, because if that stigma fades, there's less reason to pay for detection.
Pangram detects AI involvement, not AI authorship
But any detection score, even a perfectly accurate one (and from my testing, Pangram's often isn't, as discussed below), only shows whether AI was likely involved. Did someone generate an entire text from a prompt? Did AI serve as a thinking partner? Did it help with wording? Did an author just run their own writing through AI to smooth the language or translate it? The score can't tell. Ad
But "Pangram shaming" implies that any AI involvement means the writer didn't think or didn't care. A high score can hit a text someone spent hours filling with original research and arguments just as easily as one prompted into existence in ten seconds. Pangram can't tell the difference. To do that, it would have had to watch the writing happen.
Case in point: The English version of this article, the one you are reading right now, is a translation from my German original, drafted with AI and then edited by hand and by AI. Pangram scored it at "28 percent AI," flagging the final paragraphs as AI-written. But every section was created and edited the same way, with roughly the same amount of AI involvement and certainly the same amount of human intention. Ad
Self-appointed AI police like Breslau or Spero would still use that score to claim the last paragraph came from an AI and doesn't reflect my thinking. Worse: people at academic institutions reject papers based on percentage scores like these. Many of those papers come from researchers who write in a second language or who are simply better at science than at prose, and who use AI to help express their results more clearly. Used this way, Pangram penalizes them for improving their work. Ad
There's a historic analogy worth considering. Alexandre Dumas had assistants like Auguste Maquet write rough drafts of his novels and then refined them himself , including classics like The Three Musketeers. The quality came from the concept at the beginning and the polish at the end. An AI detector applied to Dumas would have flagged "Maquet usage" and missed everything that made his books what they are.
AI News Without the Hype – Curated by Humans
Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.