Reviewing work with missing context
An essay arrives with a polished argument or an unexpected change in style. You need a way to ask useful questions without treating an impression as evidence of misconduct. Our detector reads writing patterns in context. Its result can guide your review, but it cannot tell you what happened while the student wrote.
The type of assignment matters. In our published English tests, argumentative essays are flagged more often than encyclopedia prose. Language background matters too: calibration includes non-native writing, but human writing can still be flagged. Give students room to explain their choices and show drafts before interpreting a reading as a concern.
Put the reading beside the work
Consider the assignment
Identify the writing type and the context of the task. Review the student's available drafts and notes. An argumentative essay needs different context from a passage of reference prose.
Read confidence before the figure
Start with Our confidence, then the range and sentence marks. The figure describes the passage as a whole. It does not count sentences written by AI.
Ask about marked passages
Use the marks to prepare specific questions about argument, sources or revisions. Invite the student to explain their process. Do not turn the reading into a grade or accusation.
Keep review context together
Paid plans support batch checks. Pro adds a printable report with the reading and marked text. Keep any report alongside the assignment context and discussion, not as standalone proof.
Example: Discussing a change in register
In this fictional example, you review an essay whose final section sounds more formal than its opening. The headline is "Likely AI-written" and "Our confidence" is Medium. You inspect the sentence marks and "Most machine-like sentences" rather than treating the headline as a finding. Several marked passages appear in the final section, so you prepare a question about how that section developed during revision.
You open "Why this reading" to review the length band and the independent model's reading. During discussion, the student brings an earlier draft and explains why the conclusion became more formal. You compare that explanation with the wording and sources. You record what remains unclear without calling the reading proof. Any assessment of the work includes its argument, evidence and assignment requirements rather than relying on the detector.
A review you can discuss
You have specific passages to discuss rather than a score to enforce. The student's explanations, sources and writing record remain part of the review. A reading may help frame questions, but it is never proof about a student or the sole basis for a grade.
Keep separate notes for what the detector shows and what you learn from reviewing the work. A sentence mark describes a writing pattern; a source discrepancy or an explanation from the student is different evidence. Keeping those categories distinct makes it easier to explain your reasoning. It also prevents a strong headline from silently becoming a conclusion about conduct that the tool cannot establish.
Questions
Can I use a reading to assign a grade?
A reading must never be the sole basis for a grade. Human writing can be flagged and AI writing can be missed. Use the result to prepare a conversation about the work, then consider the assignment, sources and student's explanation. The detector does not establish authorship or whether a rule was broken.
Why can similar figures have different ratings?
The confidence rating reflects the range, not just its midpoint. We score overlapping parts of the passage and interpret the spread through its language calibration. Long, even writing can give a narrow range. Short writing or a change in register can give a wider range and a different confidence rating.
How should I treat second-language writing?
Treat the reading as uncertain evidence and give the student a chance to explain. Each language has its own calibration, including non-native writing. We publish false flags alongside AI recognition on the accuracy page. That testing does not eliminate errors, and language background should not become a reason to assume misconduct.
Which plans support batches and printable reports?
Every paid plan supports batch checks, while printable reports are available on Pro. A report includes the figure, range, rating and marked text, plus language and calibration details, model version and date. It stays in your browser. For reading preparation, the summarizer can provide key sentences without replacing your review of the full work.
Can teachers detect if you use AI?
Teachers cannot establish AI use from a detector reading alone. A tool can suggest that writing patterns lean toward AI, but human writing can be flagged and AI writing can be missed. The result does not show what happened during drafting. Review the work with the student, ask about specific choices and consider available drafts and sources before drawing conclusions about the process.
Is a low AI percentage bad?
No, a low AI percentage is not a judgment about the quality of the work or the student. The figure is the midpoint of a range describing the passage's reading, not the share of sentences written by AI. Read "Our confidence" and the range alongside it. A low figure does not certify authorship, just as a high figure does not prove misconduct.
Where can I review the detector's testing?
You can review our testing on the accuracy page, where we publish results by language with dates. We show false flags on human writing alongside recognition of AI writing and separate calibration results from held-out tests. Those distinctions matter when evaluating the evidence behind a tool. We also publish results that fail our publication rule and identify where performance is weaker.
Tools for this work
- AI detectorOne figure with our confidence rating and a mark on every sentence.
- SummarizerKey sentences, a paragraph or bullets at the length you set.
Free checks for a careful review
Free use needs no card. Each tool offers a daily free check with a word cap; the grammar checker and both counters have no cap. Paid plans add batch checks for submitted work. Pro adds printable reports to keep beside your review notes.