AI Detector Checker is a conservative AI-writing-signal flagger. It reads English text of at least 100 words and returns a 0-100 AI-writing signal score with a stability range and one of three bands: Likely AI-written, Uncertain, or Likely human-written. The score is an estimate of signal strength; it is not a percentage of AI-written words and not the chance that AI wrote the text. A result is not the chance that AI wrote the text, not the share of AI-written words, and never proof of authorship.
- Read the labelOne document-level score with a stability range and a band for the whole passage.
- Read the text yourselfThe tool points at nothing inside the passage; you do the reading.
- Never a verdictUse provenance and human review for permitted revision.
On this page
Result anatomy
What an AI detector result contains
A AI Detector Checker result has one label for the whole passage, a short explanation of what that label means, and a disclaimer. Alongside it: the score stability range, the measured false-flag rate for the text’s length, and the patterns observed in the text. No individual sentences are marked.
Document-level result
Likely AI-written
A score and a band, not a chance: the displayed score is a frozen mapping of the detector’s internal signal; it is not the chance of AI authorship, which is why it comes with a stability range and a band.
01
The label
Means: whether the whole passage crossed the detector’s conservative flag threshold, sat close to it, or did not.
Does not mean: the chance of AI authorship, a percentage of AI-written words, authorship, intent, or misconduct.
02
The explanation
Means: a plain-language note on how often human writing receives that label on the validation sets, and what the label cannot tell you.
Does not mean: a certainty level, an accuracy figure for your text, or validation of the individual result.
03
The disclaimer
Means: the result does not prove AI or human authorship, and is never sufficient on its own for a decision about a person.
Does not mean: that the result is optional context; the disclaimer is part of every result.
Worked example
How to interpret one illustrative response
The values below show the fields a result contains. This is not an accuracy test, a benchmark row, or an authorship finding.
- Start with the band. Uncertain means the passage sat close to the flag threshold: the signals were mixed. It is a valid product outcome, not an error and not a “50/50” finding.
- Check the word count. Results are only issued for 100 words or more. Shorter input returns “More text is needed for a meaningful analysis” instead of a label.
- Do not look for a number. There is none to look for. The internal signal is mapped to the displayed score by a frozen monotonic mapping; the band is decided by the raw thresholds only.
- Do not infer ground truth from the response. To investigate authorship or process, bring in drafts, revision history, notes, sources, citations, task instructions, and the writer’s explanation.
Why the wording is cautious: the detector is tuned so that strong flags on human writing stay rare. The price of that setting is that much AI-written text receives “Likely human-written”, and how much depends heavily on which model produced it.
The three bands
How to read Likely AI-written, Uncertain and Likely human-written results
Use the band exactly as displayed. None of the three proves authorship.
Below the threshold
Likely human-written
The passage did not cross the detector’s conservative flag threshold.
Do not conclude: “This text is human-written.” At this setting the detector misses a large share of AI-written text; edited, paraphrased, or atypical generated writing routinely receives this label.
Near the threshold
Uncertain
The signals were mixed; the passage sat close to the threshold without crossing it.
Do not conclude: “The text is a known human–AI mix.” Uncertain describes the detector’s result, not the writing workflow.
Above the threshold
Likely AI-written
The passage matched the AI-writing patterns the detector measures strongly enough to cross its threshold.
Do not conclude: “AI use is proved.” Human formal, technical, translated, template-driven, or assistant-style prose can produce a false flag.
Threshold note: the flag threshold is fixed for the current release and is documented on the model card and in the methodology. It is not adjusted per user, per document type, or from live traffic.
Why a stability range
Why the score comes with a stability range
The detector’s internal signal ranks texts, but the same value can mean different things for different kinds of human writing, which is why the score is shown with a stability range and the false-flag rate is published per text length. Showing it as “N% AI” would overstate what is known.
It is a ranking
Higher internal values correspond to stronger matches to the patterns measured; the product only reports whether a fixed threshold was crossed.
It is not a chance of AI authorship
No number on the result page would tell you the chance that AI wrote the text, because that chance depends on context the detector cannot see.
It is not a certainty figure
The detector does not report certainty about an individual result, because it has no validated way to know it.
For a process overview, read how AI Detector Checker analyzes text; for versioned technical context, see the evaluation methodology and model card.
Passages to review
Reading the text yourself
The current detector returns one label for the whole passage and marks nothing inside it. When a result deserves a closer look, the reading is yours to do.
A passage that reads as generic and unsupported belongs on a review list; it does not become proof of machine authorship.
Read the whole passage
Check whether the writing is unusually generic, formulaic, smooth, repetitive, or stylistically different from the writer’s other work.
Check evidence and specificity
Look for sources, examples, reasoning, quotations, first-hand detail, and accurate attribution—not merely a preferred detector result.
Compare the writing process
Use drafts, revision history, notes, and known samples when authorship matters. The detector cannot reconstruct those facts.
Expect misses
A “No strong signal” label is common for AI-written text, especially after editing. Absence of a flag is not clearance.
Mixed or inconclusive evidence
How to read an Uncertain AI detector result
An Uncertain result is a valid product outcome, not a technical error. It means the passage sat close to the flag threshold without crossing it.
Why it can happen
- The passage is only just over the 100-word minimum or contains little connected prose.
- The style is highly formal, technical, translated, template-driven, or written as a reply to an instruction.
- The draft has been heavily edited, paraphrased, or assembled from different sources.
- The measured patterns sit close to the fixed decision threshold.
What to do next
- Confirm that you submitted a complete, coherent passage of at least 100 words.
- Read the passage yourself with the checks above in mind.
- Check provenance, drafts, citations, and revision history.
- If you legitimately revise the text, analyze the revised version as a new result.
Do not convert Uncertain into Human, AI, or “50/50.” A technical failure, a refused short input, or a result that was not run is a separate state and should never be silently relabeled.
Scenario matrix
How to read common result scenarios
The band describes the detector’s output for the passage you submitted. It never tells you whether that output is correct for your text.
No strong signal · your own draft
Expected for most human writing, and also common for AI-assisted drafts. Treat it as “not flagged”, never as “verified human”.
No strong signal · a text you suspect
The detector misses a large share of AI-written text at its conservative setting. Suspicion should rest on provenance and process evidence, not on this label.
Strong signal · formal or templated writing
Human formal, technical, translated, and assistant-style prose can produce a false flag. Read the passage and gather external evidence before drawing any conclusion.
Strong signal · repeated on a revised draft
Each run is a separate result. A repeated flag is still a flag on measured patterns, not proof; improve the text for readers rather than to change the label.
Evidence and limits
Can an AI detector result be wrong?
Yes. Pattern classifiers produce false positives, false negatives, and Uncertain outcomes. The error risk changes with the sample, language, domain, model, prompt, length, translation, and editing history.
False positive
Human-written text receives “Likely AI-written”. Formal, technical, translated, polished, non-native, template-driven, or assistant-style prose can share measured patterns with generated text.
False negative
AI-generated or AI-assisted text receives “Likely human-written”. Editing, paraphrasing, creative variation, mixed workflows, and model differences all weaken measured cues, and the detector is deliberately tuned to accept many misses.
What the validation evidence actually establishes
The current engine (V3, released 5 September 2026) was scored once on sealed test data. Two of its six release targets were not met and it was released under a published owner override; the full result is on the evidence page. In a sealed test on human-written texts of 150 words or more, from sources the system had never seen, it gave a wrong AI signal for about 2 in every 1,000 texts. For texts of 100 to 149 words it was about 2.4 in every 100. In our sealed test of texts from the tested current AI models, about 55 in every 100 received a strong AI-writing signal. A text with no signal is not evidence that a person wrote it. Short texts, technical writing such as software bug reports, and volunteer replies written in the style of an AI assistant are harder for the system; the figures for each type are on the Evidence page. That last figure describes one model family only; it is not a general rate for unseen or current models.
Those figures describe that disclosed release, sample, and protocol. They do not guarantee performance for every model, language, document type, adversarial rewrite, or editing condition, and human writing produced in reply to an instruction was not part of the sealed human set: development evidence showed materially higher false-flag rates on that kind of writing. Read the full validation page, including denominators, uncertainty intervals, tested sources, and limitations.
- Short input: anything under 100 words is refused rather than labelled, because the detector has no validated behaviour there.
- Translations and other languages: the current validation is strongest for English text; other languages have not been validated for the current detector. See supported input and language limitations.
- Mixed authorship: one label cannot recover how much assistance occurred or which parts came from which source.
- Source-model identification: the result cannot prove that text came from ChatGPT, Claude, Gemini, or another specific system.
- Quality and truth: a label does not measure factual accuracy, originality, plagiarism, usefulness, or writing quality.
For edge cases and error handling, read AI detector limitations and false positives. AI detection and source matching are different tasks; see AI detection vs. plagiarism checking.
Responsible workflow
What should you do after reading the result?
Treat the detector as a low-stakes review aid for your own text or text submitted with informed consent. It can point to a passage worth revising or fact-checking, but it is not a triage tool for people, submissions, cases, or eligibility.
-
1
Check the input
Use a coherent English passage of at least 100 words. Exclude navigation, references, boilerplate, or copied prompts if they are not part of the writing under review.
-
2
Read the label
Record the displayed score and band without converting them into the chance of AI authorship, a percentage of AI-written content, or a statement about the author.
-
3
Read the passage yourself
Ask whether wording is generic, repetitive, unsupported, or inconsistent with the rest of the draft.
-
4
Gather external evidence
Consult drafts, revision history, citations, sources, task requirements, and the writer’s explanation. These can bear directly on process; the label cannot.
-
5
Choose a low-stakes next step
Use the result only to guide revision, fact-checking, source verification, or a consented editorial discussion. Do not use it to evaluate a person or trigger any adverse action.
If your text was unexpectedly flagged, follow the practical steps in what to do when an AI detector flags your writing. Improve the work for clarity, evidence, accuracy, and authentic voice—not merely to change the label.
Privacy: do not submit confidential, regulated, medical, financial, identity, or other sensitive material. Review the Privacy Policy and security guidance before using the tool in a routine workflow.
Quick answers
AI detector result FAQ
Is an AI detector result the percentage of text written by AI?
No. AI Detector Checker returns a 0-100 AI-writing signal score with a stability range and one of three bands for the whole passage; the score is not a percentage of AI-written words. A score never means that a certain share of the words were written by AI or that there is a certain chance of AI authorship.
Does “Likely AI-written” prove that AI wrote the text?
No. It means the passage matched the patterns the detector measures strongly enough to cross its conservative threshold. Human writing can receive this label, so authorship requires broader evidence.
Does “Likely human-written” prove the text is human-written?
No. The detector is tuned to keep false flags on human writing rare, and the price is that it misses a large share of AI-written text. Generated text routinely receives this label after editing, paraphrasing, or simply because it came from a model family the detector rarely flags.
Why does the score come with a stability range?
Because the score mapping varies across source families: the range shows that variation. It is not the chance that AI wrote the text. The same value can mean different things for different kinds of human writing, so presenting it as a percentage of AI writing or as a certainty level would overstate what is known.
Does the detector show which sentences were written by AI?
No. The current detector returns one label for the whole passage and marks nothing inside it. It cannot identify which parts of a text came from which source.
What does an Uncertain result mean?
It means the passage sat close to the flag threshold without crossing it: the signals were mixed. It does not mean the text is known to be half human and half AI, and it is not the same as a technical failure or a refused short input.
Can the result identify ChatGPT, Claude, Gemini, or another model?
No. AI Detector Checker reports a general AI-writing signal; it cannot prove which person or model produced the text.
Should I run the detector again after editing?
A second check can be useful after legitimate revision. Treat it as a new review signal and keep the original result, edits, and context together if the review needs an audit trail.
Need a broader answer? Visit the complete AI Detector Checker FAQ.
Use the result in context
Review text for AI-writing signals
Run a first-pass check, then read the passage yourself and use external evidence as the separate, decisive part of a responsible review.