Detecting possible AI involvement is different from identifying the exact AI model. An AI score does not prove that ChatGPT, Claude, Gemini or another particular tool wrote a passage.
AIScan offers optional source estimates. Treat a suggested source as context for review, not a verified record of which service someone used.
Detection is not confirmed attribution
Consider a short article that receives a high AI assessment. The result cannot establish whether the author used a writing assistant, which version they used, what they asked it to do or how much they edited afterward.
A company or model label is not a transcript of that process. It cannot identify the author or prove that a particular account generated the text.
How should I read an AIScan source estimate?
Keep the estimate beside the original document and the overall result. A suggested source is a possibility, not a confirmed finding. Several suggestions do not establish that several tools were used.
If you are reviewing someone else's work, ask about their process instead of telling them that a source label proves what happened. Our accuracy page explains the limits of both detection and source estimates.
What about edited or mixed writing?
A document can combine an author's draft, a quotation, a suggested edit and a paragraph developed with assistance. One label cannot give a complete account of those contributions.
For example, an editor might receive a human-written article with an introduction rephrased using a writing tool. Ask what changed, whether that assistance was allowed and whether the author checked the final wording. A source estimate cannot answer those questions by itself.
Short excerpts, translations and repeated revisions also need care. Do not infer reliable model identification from a convincing-looking label on one example.
Does a low AI score rule out ChatGPT or Claude?
No. A low assessment cannot establish that a passage was written without AI assistance. Likewise, a high assessment cannot confirm a particular model or an undisclosed breach of policy.
Avoid using a series of detector results as a vote on authorship. Different tools can disagree, and agreement between scores does not replace the writer's explanation or available drafting evidence.
What evidence would support a model-identification claim?
A useful evaluation would name the candidate models and their versions, include examples from models outside that list, and keep detection results separate from source-identification results. It would report mistakes, uncertain cases and performance on edited text, with enough detail to understand the test's scope.
Success on a small, fixed list would not establish reliable identification of every model in everyday use. AIScan's optional source estimates are not a validated claim that the exact model can be identified.
What should a teacher or editor ask instead?
Ask what assistance was used, which parts of the document changed, and how the writer checked the result. Review relevant drafts, sources and policies. Give the writer an opportunity to explain the work.
Teachers can use our review workflow. Writers responding to a flag can use the next-steps checklist. If your concern is copied wording, AI detection and plagiarism checking address different questions.
You can check text with AIScan, then review the result alongside the writing. A source suggestion is a prompt for a question, not the answer to who wrote it.