The best Claude AI detector on this list is the first one below, and this site runs on its engine, Veread. It is the one tool checked here that reports writing-style similarity to Claude next to the AI verdict. The other tools are described only with facts their own pages state, checked on 2026-11-10.
1. Fileverity
It checks text, images and documents. The text engine, Veread, gives a verdict, marks the passages that carry the signal and lists writing-style similarity for Claude and the other major model families. It works in seven languages and accepts PDF and Word files, and there is an API and a report you can keep as evidence.
On human-written text in seven languages, 0.5% was wrongly flagged as AI. For fully AI-written Claude text the detection figure is 98.4%. The published figures show what they measure. This site, ClaudeDetect, is a free front end to the same engine, which comes from Fileverity, so it is not an independent party in this comparison.
2. Pangram
Pangram says it is trained to detect models from OpenAI, Anthropic, Google, Meta, DeepSeek and xAI, and shows per-model detection rates. It does not state that the tool reports which model produced a text. It lists 24 languages, takes pasted text and PDF, DOCX and RTF uploads, and has an API and an MCP server. The free tier gives daily credits, and results can be saved and shared. Anthropic itself names Pangram as a detector that works differently from the watermark. See Claude detector vs Pangram.
3. GPTZero
GPTZero says it works across models including ChatGPT, Gemini, Claude, Llama and DeepSeek, and classifies text as human, mixed or AI at document, paragraph and sentence level. It does not say it names the model. It fully supports five languages, takes paste, file upload, a Chrome extension and classroom integrations, and has an API. See Claude detector vs GPTZero.
4. Originality.ai
Originality.ai classifies text as AI, original or mixed and also detects AI-paraphrased text. Its page lists accuracy on named models including Claude, but does not say it identifies the source model. Multilingual text is identified automatically. It takes paste, file and URL input, and has an API. Free use is a one-time credit allotment.
5. Copyleaks
Copyleaks lists Claude among the models whose content it can detect and says it flags text blended with human writing. Its AI Source Match compares text with published content; it is not model attribution. It states support for over 30 languages, takes paste and files, and has an API. The free tier is 25,000 characters per scan.
6. Turnitin
Turnitin is sold to institutions, not individuals, and has no paste box. It lists the Claude versions its English detector can detect, but the report does not name a model. The details are on Can Turnitin detect Claude?.
Comparison table
| Tool | Names the model family | Languages (vendor) | Input | API | Free tier |
|---|---|---|---|---|---|
| Fileverity | Yes, as writing-style similarity | Seven | Paste, PDF, Word | Yes | One free check on this site |
| Pangram | Not stated | 24 named | Paste, PDF, DOCX, RTF | Yes | Daily credits |
| GPTZero | Not stated | Five fully supported | Paste, file, extension | Yes | Yes, account above 10,000 characters |
| Originality.ai | Not stated | Auto-detected, no list | Paste, file, URL | Yes | One-time credits |
| Copyleaks | Not stated | Over 30 | Paste, file | Yes | 25,000 characters per scan |
| Turnitin | No | Seven, English paraphrase checks only | Inside institutional products | Not stated | None for individuals |
"Not stated" means the vendor pages checked do not say, not that the feature is absent. Accuracy claims are left out of the table on purpose: each vendor publishes its own, and they use different data.
How to choose
- Want to know whether it looks like Claude or another model: pick a tool that shows family similarity, such as the first-ranked tool above.
- Checking a whole class or archive: look at institutional tools and the API.
- Need a record: choose a tool that gives a downloadable report, and read the facts page before relying on any score.
- Short texts, light edits and mixed human-AI writing weaken every detector, so treat a result as evidence, not proof.
Sources: Pangram, Pangram multilingual, GPTZero, Originality.ai, Copyleaks, Turnitin AI writing detection FAQ, Anthropic on Claude text watermarking.