Pangram

Properties

category: detector
maturity: established
approach: Sentence-level linguistic and semantic analysis (not perplexity/burstiness) trained to catch AI text including lightly edited or humanized content; reports both a probability the text is AI-generated and an estimate of how much of the document is AI.
access: freemium
owns: AI-text detection product (web dashboard, browser extension, Canvas/LMS integration, API) built for teachers, publishers, and trust-and-safety teams; became Substack's built-in AI-detection provider for posts over 100 words starting July 21, 2026.
distinctFrom: Unlike GPTZero/Originality.ai/Copyleaks (perplexity-and-burstiness classifiers marketed mainly to educators or publishers as standalone checkers), Pangram markets itself on sentence-level explainability and a low false-positive rate, and is now the detection layer embedded directly into a major publishing platform's own AI-disclosure feature rather than a separate tool users visit.
source: Pangram (pangram.com)
notes: Vendor-claimed 99.98% accuracy and third-party-reported ~1-in-10,000 false-positive rate are vendor/review-sourced, not independently reverified here. Free plan: 4 checks/day; paid subscription unlocks more credits and plagiarism detection. Substack integration announced 2026-07-21 via Substack's own product update, cited as evidence of adoption. Update (2026-07-29): Pangram raised $9M (led by Menlo Ventures, with Haystack, ScOp, Script Capital, and Cadenza participating) and launched Pangram 4, a next-generation text-detection model vendor-claimed at over 99% accuracy including improved detection of AI-humanizer output, plus Pangram Image, an AI-generated-image detector currently in research preview ahead of a wider release; this supersedes the prior 'text-only, no image detection' note. Update (2026-09-17): Pangram launched a Gmail integration that scans incoming mail and labels it directly in the inbox as AI, Mixed, or Human; Pangram's own product page states it 'Reads message text only to classify it' with 'zero data retention' and does not access sender, recipient, subject, or other mailbox metadata. Coverage (Gizmodo) confirms the labels default to flagging only AI/Mixed mail, with human-written labeling optional, sender exclusions, and a rolling dashboard of label history.

Provenance

Source: https://www.pangram.com/

Properties

category

detector

maturity

established

approach

Sentence-level linguistic and semantic analysis (not perplexity/burstiness) trained to catch AI text including lightly edited or humanized content; reports both a probability the text is AI-generated and an estimate of how much of the document is AI.

access

freemium

owns

AI-text detection product (web dashboard, browser extension, Canvas/LMS integration, API) built for teachers, publishers, and trust-and-safety teams; became Substack's built-in AI-detection provider for posts over 100 words starting July 21, 2026.

distinctFrom

Unlike GPTZero/Originality.ai/Copyleaks (perplexity-and-burstiness classifiers marketed mainly to educators or publishers as standalone checkers), Pangram markets itself on sentence-level explainability and a low false-positive rate, and is now the detection layer embedded directly into a major publishing platform's own AI-disclosure feature rather than a separate tool users visit.

source

Pangram (pangram.com)

notes

Vendor-claimed 99.98% accuracy and third-party-reported ~1-in-10,000 false-positive rate are vendor/review-sourced, not independently reverified here. Free plan: 4 checks/day; paid subscription unlocks more credits and plagiarism detection. Substack integration announced 2026-07-21 via Substack's own product update, cited as evidence of adoption. Update (2026-07-29): Pangram raised $9M (led by Menlo Ventures, with Haystack, ScOp, Script Capital, and Cadenza participating) and launched Pangram 4, a next-generation text-detection model vendor-claimed at over 99% accuracy including improved detection of AI-humanizer output, plus Pangram Image, an AI-generated-image detector currently in research preview ahead of a wider release; this supersedes the prior 'text-only, no image detection' note. Update (2026-09-17): Pangram launched a Gmail integration that scans incoming mail and labels it directly in the inbox as AI, Mixed, or Human; Pangram's own product page states it 'Reads message text only to classify it' with 'zero data retention' and does not access sender, recipient, subject, or other mailbox metadata. Coverage (Gizmodo) confirms the labels default to flagging only AI/Mixed mail, with human-written labeling optional, sender exclusions, and a rolling dashboard of label history.