Can you tell if text was written by AI?
You can check for writing patterns associated with AI-generated text, but a detector alone cannot prove who wrote a passage. The most useful check combines a text analysis with drafts, source notes, revision history and a clear account of how the writing was produced.
If you are asking “is it AI generated?” about your own draft, save the original before checking it. That gives you a stable reference if you later revise the wording. For someone else’s work, treat a detector result as a reason to ask a fair question, not as a verdict.
Step 1: paste the original or upload a document
Open the AI text detector and paste your English passage, or choose Upload document. Supported formats are PDF, Word DOCX, TXT and Markdown. You can also drag one file into the input panel.
Files must be 10 MB or smaller. A PDF can contain up to 100 pages and needs selectable text; a photograph or scan requires OCR first. Save an old Word .doc file as .docx. After import, inspect paragraph order, spacing and special characters because document extraction can change how a passage is represented.
Step 2: check a useful sample
Choose Check text with 80–3,000 words, at least five sentences and no more than 20,000 characters. A short email, a single sentence, source code or a list may not provide enough supported evidence. The tool explains when it cannot assign a score.
Use a continuous passage with enough context to represent the writing. Repeating a sentence to meet the minimum length does not add evidence. For a longer document, inspect relevant sections separately and keep track of which section produced each result.
Step 3: read the score and inspect the patterns
The report shows a 0–100 AI signal score and a band: fewer AI-like patterns, mixed signals, or more AI-like patterns. This scale is not the percentage of text written by AI. A score of 70 does not mean that 70% of the words were generated.
In Overview, inspect sentence-length regularity, repeated openings, repeated four-word sequences and other descriptive measurements. Open Text highlights to see the exact matched wording. These observations help you review the passage; they are separate from the learned score and do not identify individual AI-authored sentences.
Step 4: compare the result with other evidence
Check whether a draft history supports the claimed writing process. Look at notes, cited sources and revisions. Ask whether the writer can explain the choices and sources in the passage. A polished tone, standard academic phrasing or a repeated subject can be ordinary human writing.
Different checks answer different questions:
| Check | What it can show | What it cannot establish alone |
|---|---|---|
| AI text detector | Patterns associated with labeled human and AI examples | Who wrote the text or how much AI was used |
| Statistical watermark verifier | A compatible generation-time signal | Every model or every edited passage |
| Unicode inspection | Hidden characters and unusual formatting | AI authorship or a proprietary watermark |
| Source and plagiarism review | Overlap with identifiable source material | Whether an original passage was AI generated |
| Draft and revision history | Evidence about the writing process | A complete history when records are missing |
For more on false positives, read how accurate AI detectors are. For hidden characters and generation-time signals, read text watermarks and SynthID.
What if human-written text is flagged?
Keep the original text and export the report so the finding can be reviewed. Do not keep substituting words until the score falls and treat that as proof; it only shows the detector responds to the changes. A flag can reflect a false positive, the writing genre or patterns shared by human and AI examples.
If the result affects a grade, job or publication, seek a human review with the writing context and relevant records. Our methodology states the measured false-positive rate for one benchmark and the populations that have not been separately validated.
What should you do after the check?
Use the result to choose your next action. If the text needs clearer wording, review it with the paraphrasing tool. If it contains invisible or unusual characters, inspect them with the watermark checker. If you need an authorship decision, gather process evidence and discuss it with the writer.
You can export a report containing your text, engine version and measurements. Review it before sharing. Keep your original and revision notes together so the analysis remains understandable later.