Pangram AI detector review: all three verdicts right, all three hedged

Free plan facts↑ Contents

Results↑ Contents

The headline result for each text, with default settings. Green means the verdict matched what the text really is.

Every result carried “Less accurate on short text” and the line “This text is too short to classify with certainty, but we believe it may be AI generated” (or “human written”). One scan per text, so we can’t say how stable these are. Pangram’s word counter differed from ours by a few words (111 or 112 for the 109-word text A, 121 for the 122-word text D, 146 for the 145-word text F); we don’t know why, and the result screen repeated each full text.

Against other detectors↑ Contents

Text D is where free detectors split in our detector comparison. Pangram sides with Copyleaks, GPTZero, Originality.ai and Sapling, and not with ZeroGPT.

One scan per text per tool. The other detectors’ results are from their own scan dates (Oct 8 to Oct 19). Pangram’s “100%” is the share of the text it calls possibly AI or possibly human, not a probability that the text is AI.

Findings & screenshots↑ Contents

• The percentage and the headline disagreed in tone. The gauge said 100% and the Details tab listed one “High” confidence segment over the whole text, yet the headline read “Possibly AI” with “too short to classify with certainty”. All three of our texts were between 109 and 145 words. We don’t know where Pangram draws the line between short and long.
• Only the rewrite got a rewrite flag. On text D, the Undetectable AI rewrite, the Overview tab added “This text appears to have been paraphrased or rewritten” with the line “This appears to be AI text that has been paraphrased or rewritten” and a “View paraphrased segments” button showing 1. Texts A and F had no such row. The flag did not change the headline or the 100% gauge.
• Text F, the 1889 passage, came back Possibly human, 100%, the whole passage highlighted green. One old, famous human text is a thin control, so read this as one data point.
• The home page says “1 in 10,000 False positive rate” and “#1 In studies from independent researchers”. We did not check either claim, and three short scans say nothing about accuracy.

The three texts↑ Contents

The same texts we gave the other detectors. Copy them to check another tool.

Artificial intelligence has fundamentally transformed the way students approach their academic work. In today's rapidly evolving digital landscape, AI-powered tools offer unprecedented opportunities to enhance learning outcomes and streamline research processes. Furthermore, these technologies enable learners to access personalized feedback, identify knowledge gaps, and develop critical thinking skills more efficiently than ever before. However, it is important to note that the integration of AI in education also presents significant challenges, including concerns about academic integrity and over-reliance on automated assistance. Ultimately, striking the right balance between leveraging AI capabilities and fostering independent thought will be crucial for educators and students alike as they navigate this new era of learning.

Artificial intelligence is changing how students do their schoolwork. With all the new digital tools available, students can learn more easily and get help with their research. These tools can even give students personalized feedback, help them find areas where they need to improve, and make it easier for them to think critically. But we also need to be careful - using AI in education can be tricky. Some people worry that students might cheat or rely too much on computers to do their work. So, teachers and students need to find a way to use AI that helps them learn without doing all the work for them. This will be important as we move forward in this new way of learning.

We were all feeling seedy, and we were getting quite nervous about it. Harris said he felt such extraordinary fits of giddiness come over him at times, that he hardly knew what he was doing; and then George said that he had fits of giddiness too, and hardly knew what he was doing. With me, it was my liver that was out of order. I knew it was my liver that was out of order, because I had just been reading a patent liver-pill circular, in which were detailed the various symptoms by which a man could tell when his liver was out of order. I had them all. It is a most extraordinary thing, but I never read a patent medicine advertisement without being impelled to the conclusion that I am suffering from the particular disease therein dealt with in its most virulent form.

Text D comes from our Undetectable AI review. Text F is public domain (Project Gutenberg #308).

Pros and cons↑ Contents

Method & updates↑ Contents

On Nov 1, 2026 an automated browser opened pangram.com as a guest, pasted each text into the detector and pressed “Check for AI” once, in the order A, F, D. It clicked the human-check box once. We recorded the results, the credits and the locked parts; Jamie C. checked the records and wrote this page. Full method.

Not tested

• What happens when the guest credits run out, or whether they reset by day
• Texts over a few hundred words, where the short-text notice may not appear
• The paraphrase flag on other rewrites, or on a human text that was lightly edited
• The AI-assisted view, the “AI + Human” sample and non-English text
• Similarity analysis, Notes, the Chrome extension, the API and anything behind an account or paid plan
• The “1 in 10,000” false-positive claim and the independent-study ranking on the page
• Pangram’s image detector

Update log

• Nov 1, 2026First published, from three guest scans.

↑ Back to top