Pangram
Pangram is an AI-text detection tool that estimates whether writing was AI-generated or human-written.
HN commenters recommend Pangram as an AI-text detector with a very low false-positive rate,1234 and some say evasion tricks work less well against it than expected.56 The most common complaint is that it misses AI text, especially from prompted or edited model output,78910 and several commenters call its evidence display misleading.1112
Why people recommend it
Low false positives · 3
Commenters say a 100% AI verdict from Pangram can be trusted because human text is rarely flagged.
“if Pangram says something is 100% AI-written, you can trust that.”
“Pangram does work, in the specific sense that when it says something was AI authored it is vanishingly unlikely that it was written by a human”
Show 1 more
“My experience is using Pangram quite often with lots of writing of all flavors”
Overall quality · 3
Users describe the core classifier as good and say detection has improved a lot.
“Give it a shot, it's quite good.”
“The main "is this AI or human?" classification is done with a machine learning model that works very well”
Show 1 more
“pangram is pretty good; i use it all of the time and pay for it.”
Mixed authorship · 3
It reports what share of a text is AI-touched rather than only giving a yes-or-no answer.
“you might want to check out the leading model”
“using Pangram free "credits"”
Show 1 more
“Pangram says 50% human. That seems to check out”
Hard to evade · 2
Commenters report that attempts to beat it through iteration or sampling tricks largely failed.
“that technique doesn't work nearly as well (at least with Pangram 4) as you think it might”
“Not convinced just using top n sigma is going to beat a SOTA detector. The fingerprints Pangram uses”
Caveats and dissent
Misses AI text · 3
Commenters say system prompts or edits can get AI text classified as human.
“if Pangram says something is 100% AI-written, you can trust that.”
“Pangram fails spectacularly in detecting AI.”
Show 1 more
“It flagged the first and last paras as "high" AI and the middle para as "high" human.”
False positives · 2
One user says it flagged their handwritten text as AI, and another says it leans toward calling text AI.
“Pangram and GPTZero also think it is written by a bot”
“Give it a shot, it's quite good.”
Misleading UI · 2
Commenters call the UI and evidence feature awful and misleading, even when the underlying model is accurate.
“I agree Pangram's UI is awful and often misleading. Their underlying classifier model is pretty accurate in my experience”
“The main "is this AI or human?" classification is done with a machine learning model that works very well”
Pricing · 2
One user calls the models pricey, and another runs through a monthly token allowance.
“Give it a shot, it's quite good.”
“if Pangram says something is 100% AI-written, you can trust that.”
Signup restrictions · 1
One user could not sign up with custom or Gmail email domains because of a suspicious-activity error.
“The error was always the same, there had been suspicious activity from that domain.”
Overreliance risk · 2
Critics warn against treating a single detector's result as truth and question how it was trained.
“This whole company is honestly such a piece of shit”
“It's trivial to see how many people think Pangram is absolute trash[1] (because it is).”
Who recommends it in their own words
From the makers not counted in the ranking
“The synthetic mirrors method for data generation used here is the same as what we use at Pangram.”
Threads where it came up
- The revolt of the reader 3 mentions
- AI Content Is Everywhere on Social Media, Especially LinkedIn 3 mentions
- Detecting LLM-Generated Texts with "Classical" Machine Learning 2 mentions
- Code-execution flaw in UEFI firmware threatens most servers on planet Earth 1 mention
- Symbolic Learning in Python and Julia 1 mention
- Loop Engineering: Optimizing Code Generation and Verification 1 mention
Data: pangram.json. Quotes are verbatim; the verdict and theme labels are written by a model from these comments, and every clause links to its sources.
All comments 28
“pangram does say 97% human and only 3% AI”
“Pangram reports 100% AI on most comment blurbs in the examples directory.”
“Do I know how to use Pangram, too? Absolutely. I got my Pangram merit badge months ago.”
“I decided to give it to pangram to find out it was 77% written by AI.”
“this is 100% ai generated as per pangram”
“using Pangram free "credits"”
“Give it a shot, it's quite good.”
“Pangram says 50% human. That seems to check out”
“Pangram and GPTZero also think it is written by a bot”
“The error was always the same, there had been suspicious activity from that domain.”
“My experience is using Pangram quite often with lots of writing of all flavors”
“that technique doesn't work nearly as well (at least with Pangram 4) as you think it might”
“This whole company is honestly such a piece of shit”
“pangram is pretty good; i use it all of the time and pay for it.”
“I agree Pangram's UI is awful and often misleading. Their underlying classifier model is pretty accurate in my experience”
“Their post gets hard-flagged by Pangram haha.”
“It flagged the first and last paras as "high" AI and the middle para as "high" human.”
“Not convinced just using top n sigma is going to beat a SOTA detector. The fingerprints Pangram uses”
“Pangram fails spectacularly in detecting AI.”
“It's trivial to see how many people think Pangram is absolute trash[1] (because it is).”
“Pangram does work, in the specific sense that when it says something was AI authored it is vanishingly unlikely that it was written by a human”
“The value is that pangram detects AI generated content which it did not in the above example.”
“if Pangram says something is 100% AI-written, you can trust that.”
“But definitely no coincidences here.”
“The main "is this AI or human?" classification is done with a machine learning model that works very well”
“I recommend checking on Pangram.”
“Pangram indeed claims the OP is 76% AI-generated.”
“you might want to check out the leading model”