If you used any AI tool to outline a homework essay, draft a lab report, or unblock a tough problem set, the prose now reads like an AI tool. TextSight runs a sentence-level check across output from GPT-4 and GPT-4o, Claude, Gemini, Copilot, and Llama, then rewrites the flagged sentences toward your authentic voice. Built for K-12 through college, ESL-aware, citations preserved. Self-check before submission, not a shortcut — talk to your teacher about your school's AI policy.
ChatGPT is the loudest name, but students use a whole stack. The detector and AI rewriter were tuned across the five model families below and treat their output uniformly, because the AI-tell prose patterns overlap more than the brand names suggest.
The most common source. Defaults to the five-paragraph skeleton, opens body paragraphs with transition words, and over-uses words like "delve," "tapestry," and "navigate." If your draft reads like a Wikipedia article wrote itself, you probably used GPT.
Claude tends toward longer, more elegant sentences with a noticeable signposting habit ("First, let's consider," "It's worth noting that"). The voice is slightly warmer than GPT but still gives itself away on the second-pass read. The AI rewriter treats Claude-style output exactly the same as GPT-style.
Gemini leans factual and list-heavy. Assignments drafted with Gemini often arrive over-structured with bullets and numbered headings where prose would be expected. The detector picks this up; the AI rewriter unwinds the list-iness back into paragraphs.
Copilot is GPT-4 under the hood for prose generation, so the tells are similar to OpenAI's own output, with extra summarisation-heavy tendencies inherited from the search-grounded mode. Same workflow applies.
Llama variants and the open ecosystem (Mistral, Qwen, DeepSeek) produce prose that often reads more "formula-driven" than the closed-source models. Detectors catch this easily. The AI rewriter treats it the same way: rewrite the flagged sentences toward your real voice.
Not every school assignment is a five-paragraph essay. The AI rewriter is tuned against the formats students juggle weekly across K-12 and college coursework.
Standard five-paragraph, argumentative, comparison, persuasive, and short narrative formats. The most common case and the one detectors are most aggressive on, because they were trained on student essays. Light mode is the default. Aim for above 70 on graded essays, above 80 if you want margin.
Structured scientific writing with fixed sections (intro, methods, results, discussion). The methods section is where AI shows up most often because the format is predictable. Add the specifics only your group knows — which trial failed, which timer you actually used, what you recalibrated — and the methods section reads human again.
Math, physics, CS, and economics problem sets that ask for a paragraph of reasoning alongside the answer. AI-written explanations sound textbook-perfect. Real student reasoning has hedges ("I think this is because"), specific number references, and the occasional sidetrack. The AI rewriter brings those back.
Weekly forum posts and reading response journals are low individual stakes but high frequency. Most slip past Turnitin's automatic pipeline but get spot-checked by TAs who notice a polished tone. Drop contractions and a personal reaction into AI-drafted journal entries and they read student again.
A spoken presentation script written by an AI tool sounds like an essay being read aloud. Long sentences, no contractions, no rhetorical questions. The AI rewriter reshapes the prose for spoken delivery: shorter sentences, contractions everywhere, casual link words.
School assignments are the use case where the line between legitimate self-checking and academic dishonesty matters most. We want to be explicit about which side of that line we are on, especially for students under 18.
If you are in K-12, talk to your teacher about your school's AI policy before submitting anything that involved an AI tool, even at the outline stage. Some schools allow AI assistance for brainstorming and editing; some ban it entirely. The AI rewriter cannot make a policy violation go away. Knowing the policy is part of the work.
Assignments you wrote yourself, with an AI tool used as an outline assistant, an idea unblocker, or an editing helper. The thinking is yours, the argument is yours, the structure is yours. The AI rewriter helps you catch sentences where the assistant register leaked into the prose so the submitted draft reads in your own voice rather than the AI voice. This is closer to a careful proofread than to anything else.
If you already wrote the assignment without any AI help, run TextSight as a sanity check. Detectors flag fully human work at 1 to 2 percent of submissions; on a class of 200 students, that is several false positives per week. The Authenticity Score and the sentence-level highlights tell you whether a draft reads as your authentic voice before a teacher's detector sees it.
Generating an assignment with an AI tool and submitting it under your name in a graded context is academically dishonest regardless of which AI rewriter you run it through. We will not pretend otherwise. If that is the situation you are in, we would rather you used the detector to understand which parts of the draft read AI and then rewrote them in your own words, building the skill the course is trying to teach.
The reason schools care about AI assistance is that writing is how you build the thinking. A rewritten AI essay you did not actually understand will not help you on the exam. Use the scan and the highlights as a learning tool: notice which sentences read AI, understand why, and write the next assignment more in your own voice from the start.
Same workflow whether your draft came from ChatGPT, Claude, Gemini, or Copilot. The model that wrote the AI sentences does not change the fix.
Paste the draft you intend to submit. TextSight returns an Authenticity Score from 0 to 100 and a sentence-level colour map. Anything below about 35 reads detector-level AI; above 70 reads consistently human. The score itself matters less than the highlights.
The colour map tells you which sentences pull the score down. In a typical 500-word homework essay, three to seven sentences carry most of the AI signal. The rest of the draft is usually fine. You do not need to rewrite the whole thing.
For each red sentence, either rewrite it in your own words or click Rewrite. Light is the default for academic work because it preserves formal register and citation context. Switch to Balanced only for longer pieces (1,500 plus words) or reflection essays where a less formal voice is acceptable.
Paste the revised draft back in. Aim for above 70 on graded assignments, above 80 if you want margin. Then submit through your normal channel (LMS, email, Google Classroom). TextSight does not interact with Turnitin and we make no promise about specific detector outcomes; we report our own score honestly and let you decide whether the draft is ready.
Maximum can flatten the academic register your teacher expects, so the default we suggest for school assignments is more conservative than the default for blog content.
Light makes mild edits and preserves academic register, citation context, and sentence structure. Score gains per pass are smaller, but the output still reads like a draft you would hand in. This is the starting mode for any graded school work and the default the AI rewriter picks when an ESL register is detected.
Balanced runs moderate rewrites and is right for assignments above 1,500 words, reflection essays, blog-format coursework, and pieces where the teacher expects a less formal register. It shifts vocabulary and rhythm more aggressively than Light without flattening your voice.
Maximum runs the most aggressive rewrite and produces the biggest score gain per pass. The caveat is real: aggressive rewrites can flatten authentic voice, replacing your distinctive phrasing with generic conversational patterns that read fine for a blog post but flat for an assignment. Use Maximum on isolated red sentences after a Balanced pass, not on the whole draft as a first move.
Default recommendation for school assignments: start on Light. If the score is below 70 afterward, run Balanced on the remaining red sentences. Only reach for Maximum on the few sentences that still flag after both passes.
Detectors have a documented false-positive bias against non-native English writers. The prose patterns common in ESL writing (slightly more formal register, less idiomatic phrasing, more careful sentence structure) overlap with the patterns detectors learn from the model side. That bias can flag fully human-written assignments as AI.
In our internal evals against a sample of human-written ESL assignments, the average competitor detector returned a false-positive rate around 18 to 22 percent. TextSight's detector returns roughly 11 to 13 percent on the same sample. That is around 40 percent fewer false positives, not zero, and we report this honestly because the gap matters for students whose first language is not English.
When ESL register is detected on a scan, the AI rewriter defaults to Light mode and adjusts its vocabulary substitutions away from idiomatic native-speaker phrasing. The goal is to keep your voice (which is in part your second-language voice) and only fix the AI-tell sentences. We do not try to make ESL writing sound like native-speaker writing; that would erase the writer.
If you are an ESL student who used an AI tool for editing assistance, detector risk is higher both because of the assistant's influence on the prose and because of the underlying bias. Running the pre-submission scan matters more, not less. The threshold to clear is the same; the path to clearing it is gentler.
All three modes available on every paid plan. Students with a verified .edu email get Pro at $13.99 per month instead of $19.99. Full details on the pricing page.
Billed $89.88/year — Save $30
Billed $179.88/year — Save $60
Billed $359.88/year — Save $120
Yearly billing saves 25%. View full pricing →
The essay-only deep dive: five-paragraph, argument, comparison, persuasive, narrative.
Open essay guide →Multi-week projects, research reports, group docs, presentation scripts.
Open projects guide →The full student workflow, .edu discount, and how to use TextSight inside your academic policy.
Open student page →How the score is computed and what threshold to aim for before you submit a graded assignment.
Read the guide →Free to try, no card. Works on ChatGPT, Claude, Gemini, Copilot. Sentence-level highlights, three modes, ESL-aware, citations preserved.