How to use AI Detect
A useful check starts with understanding its limits.
AI Detect analyzes submitted text and displays a detection score alongside readability estimates. These results support a review of writing. They do not establish authorship, honesty, or academic integrity.
What happens when you check text
You paste text, or use a paid plan to upload a document and review the extracted passage. AI Detect sends that text to its detection provider, FastDetect, and displays the returned numeric score. Readability and grade-level estimates are calculated separately. Submitted scans may be saved for the service’s recent-history feature. Review the privacy policy before sharing a document.
What the detection score means
The displayed AI signal is the provider’s AI-generation estimate on a 0–100 scale. A higher value indicates a stronger AI writing signal. The provider’s own demonstration labels the same output as AI-generated likelihood. It is not the percentage of words or sentences written by AI, proof of authorship, or a guarantee that an individual used AI. We do not assign an automatic pass/fail threshold.
Check the text itself, verify its sources, and consider how it was written. A result alone cannot identify the author, the model, or the prompt that produced a passage. See the provider’s public API documentation for the underlying service’s description.
Where detection can go wrong
False positives
Human-written text can look like generated prose. A flag is not proof that a writer used AI.
False negatives
Generated or AI-assisted text can go unrecognized, especially after edits or a mix of contributions.
Limited context
Short passages, lists, quotations, formulaic writing, and translated text can be difficult to interpret.
Different systems
Tools use different methods and score definitions. AI Detect cannot predict a Turnitin or GPTZero result.
Accuracy claims and evidence
We do not publish a universal accuracy percentage or claim that AI Detect outperforms another service. This site does not provide a representative, independently validated benchmark that would support such a claim. Language, text length, editing, and the origin of the writing can change the usefulness of a result. The English guidance on this site is not a claim of validated detection performance in every language.
If you evaluate the tool, use samples with known provenance and record the date, document type, length, and outcome. Include original human writing as well as generated text. Report false positives and false negatives separately; do not generalize from a handful of examples.
Readability is a separate measure
Reading ease and grade level estimate linguistic complexity using features such as sentence length and syllables. They are not a test of expertise, factual accuracy, or authorship. A technical passage may need specialist vocabulary. Interpret the result for the intended reader rather than aiming for the same level in every document.
Use results with a fair review process
- Read the document and verify its central claims and citations.
- Consider notes, drafts, version history, and disclosed assistance.
- Give the author a chance to explain their approach.
- Follow the relevant school, workplace, or publisher’s policy.
- Do not use a detector score as the sole basis for a penalty or rejection.
When a scan fails
If the detection provider cannot return a valid result, AI Detect shows an error. It does not substitute a randomly generated score or save a successful scan for that failed request. You can retry when the service is available.