Pangram AI Detector: How Accurate It Is and What It Costs
Pangram AI is a detector that estimates whether a piece of text was written by a person, assisted by AI, or generated by AI. Made by Pangram Labs, it has a strong reputation in independent tests and is now used by platforms like Substack, which has made it both influential and controversial.
The quick answer: Pangram is among the most accurate AI text detectors available, and it’s a good tool for checking large amounts of writing. A result is still a probability and shouldn’t be treated as proof. It’s weaker on short text and many non-English languages, and a false flag can do real harm to a writer or student.
Pangram at a glance
Checked against Pangram’s site and press coverage on November 2, 2026.
| Pangram | |
|---|---|
| What it is | An AI detector for text and images |
| Founders | Max Spero (CEO) and Bradley Emi (CTO) |
| Funding | $9 million round led by Menlo Ventures, July 2026 |
| Languages | More than 20 |
| Paid plans | Individual $20/month, Professional $65/month, Team $20/seat |
| Other tools | Chrome extension, Google Docs, Gmail, API and MCP server |
| Best for | Editors, teachers, platforms and anyone reviewing writing in bulk |
What Pangram does
Paste in text and Pangram labels it as human, AI-assisted or AI-generated, highlighting the likely AI sections. Pangram 4, released on July 29, 2026, is its latest model, and it can also detect text run through “humanizer” tools. Pangram also checks for similarity to existing sources, and it has a newer image detector.
The company has become a kind of public referee. Fast Company reports that its results have fed scandals over an award-winning short story, a popular novel and a billionaire’s Wall Street Journal column. Spero calls himself a “slop janitor” and argues that people need a way to know what’s AI before they can decide how they feel about it.
How accurate Pangram really is
Accuracy claims need careful reading, so here’s what each number means.
Pangram’s site cites a false positive rate of about 1 in 10,000. For Pangram 4, its model card publishes 0.0041%, roughly 1 in 24,000, with a false negative rate of 0.3396%. Those figures come from Pangram’s own testing.
Independent work is favorable. Fast Company says Pangram routinely beats other detectors in independent testing, and a University of Chicago study is the one most often cited. Those studies are generally too small to confirm a rate as low as 1 in 24,000, though, and the ones we saw tested earlier Pangram versions.
Pangram documents its own limits as well:
- Short text. It doesn’t work on anything under 50 words.
- Certain formats. Code, tables of contents, technical manuals and templated writing are out of scope.
- Other languages. Published miss rates are several times higher for languages like Urdu and Persian than for English.
- Context. A passage can score differently alone than inside a longer document.
- Confidence. Its confidence labels aren’t calibrated probabilities, so “High” doesn’t mean 95%.
The Substack backlash
The hardest test of Pangram has been social. Substack added Pangram-powered detection in July 2026, letting readers scan posts, notes and comments. Some writers called it a “witch hunt”, and NPR’s coverage described a divided community.
Critics also argue detectors misfire on non-native and neurodivergent writers. Writers who earn a living on the platform worry a false flag could cost them income. Patrick Traynor, a University of Florida professor who studies detectors, put the core issue in four words: “What if you’re wrong?”
Even a low error rate matters at scale. At 1 in 10,000, a platform scanning a million human-written posts would still flag about 100 of them. Each one is a real person who has to defend their work.
Pangram pricing
| Plan | Price | What you get |
|---|---|---|
| Individual | $20/month | 300,000 words, 100 image scans, MCP, Docs and Gmail, $10 in API credits. 7-day free trial |
| Professional | $65/month | 5x the words and image scans, $200 in API credits |
| Team | $20/seat, two-seat minimum | Individual features plus an admin dashboard and unified billing |
Pangram also offers free credits to try it, and sells API credits to developers. Organizations such as universities can contact sales. Annual billing saves $60 on Individual and $240 on Professional.
How to use a Pangram result fairly
- Treat it as a lead. A flag is a reason to look closer, not a verdict.
- Ask for process. Drafts, notes and version history often settle the question faster than a score.
- Talk first. Speak to the person before taking action, and let them respond.
- Set a policy. Decide in advance what counts as acceptable AI help. Grammar tools and full drafts are different things.
- Avoid high-stakes use alone. Don’t fail a student, fire a writer or reject an applicant on one score.
Pangram compared with other detectors
GPTZero is popular in education, and Originality.ai is common with publishers. Turnitin is the default at many universities. Pangram stands out for accuracy on longer English text, and it costs more than some rivals. Our Pangram alternatives guide compares seven options by job.
If you write with AI and worry about being flagged, disclose it where your audience expects that, and edit until the work sounds like you.
FAQ
Is Pangram accurate?
It’s among the most accurate detectors in independent tests. Its published false positive rate of about 1 in 24,000 for Pangram 4 is the company’s own number, and accuracy falls on short text and many non-English languages.
Can Pangram be wrong?
Yes. No detector is perfect. A result is an estimate, so it should never be the only evidence against someone.
How much does Pangram cost?
Individual costs $20 a month, Professional costs $65 a month, and Team costs $20 per seat with a two-seat minimum. There are free credits to try it.
Does Substack use Pangram?
Yes. Substack added Pangram-powered AI detection in July 2026.
What is the minimum length for Pangram?
Pangram needs at least 50 words. Longer text gives more reliable results.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
