Skip to content
ClaudeDetect
Language: English

Claude detector vs Pangram

By Yang Zheng · Updated

Pangram and ClaudeDetect both read writing patterns, not the Claude watermark. Pangram is a general detector trained on models from six vendors, including Anthropic. ClaudeDetect runs on the Veread engine and adds writing-style similarity per model family. This site runs on that engine, so the table describes each tool and does not declare a winner. Pangram facts come from its own pages, checked 2026-11-10.

Side by side

ClaudeDetect Pangram
Output Verdict, marked passages, style similarity per family AI-generated and AI-assisted text, highlighted segments
Names the model family Yes, as style similarity Not stated
Claude covered Yes Yes, listed among trained models
Humanized or translated text Not claimed on this page Vendor says it detects it
Languages Seven 24 named
Input Paste; PDF and Word in the full detector Paste; PDF, DOCX, RTF; extension; Google Docs; LMS
API Yes Yes, plus an MCP server
Reports Evidence report in the full detector Scan history, downloadable and shareable reports
Free use One free check on this site Daily free credits
Minimum text 100 words Not stated

Where this engine differs

Where Pangram differs

Accuracy claims

Pangram says it is over 99% accurate with a false positive rate of 1 in 10,000, calculated on public datasets. Those are vendor claims. Veread's, including 0.5% of human-written text wrongly flagged across seven languages and 98.4% for fully AI-written Claude text, are on the facts page. Compare them knowing the data differs, and run both tools on the same text. For the wider field see best Claude AI detectors.

Sources: Pangram, Pangram multilingual, Pangram free account, Pangram data privacy.

Questions and answers

Does Pangram detect Claude?

Pangram says it is trained to detect models from OpenAI, Anthropic, Google, Meta, DeepSeek and xAI, and shows per-model detection rates on its home page.

Does Pangram say which model wrote a text?

Its pages do not say the tool reports the source model. ClaudeDetect reports writing-style similarity per model family, as a signal and not proof.

Why does Anthropic mention Pangram?

Anthropic names Pangram as detection software that works differently from its watermark, because such companies do not hold Anthropic's key. See Claude watermark status.

Which is more accurate?

This page does not rank them. Pangram's figures are its own claims, and ClaudeDetect's are on the facts page. They use different data.