Paste your draft, see an Authenticity Score on a 0 to 100 scale, and read which specific sentences carry the AI signal. The score is the headline; the sentence-level colour map is what you actually act on. This page is the pre-Turnitin draft check students run before they submit: scan the essay, review the highlights, revise the flagged paragraphs in your own voice, then re-scan to verify the score moved above the threshold you need. Calibrated for the five common essay formats. ESL-aware. Free tier covers most student use; .edu Pro is $13.99.
This is the routine students who run a pre-Turnitin draft check actually follow. The score is the entry point. The sentence-level highlights are where the work happens. Re-scanning is what closes the loop.
Open the TextSight detector and paste your full essay. The free tier covers 1,500 words per month, which is enough for one full draft check on a standard 800-word essay plus a revision pass. Citations and reference lists are detected automatically, so you do not need to strip them before scanning. The scan returns within a few seconds for essays up to about 2,000 words.
The Authenticity Score is a 0 to 100 number where 100 reads fully human to the classifier and 0 reads fully AI. Useful as a summary; not enough on its own. Underneath, the colour map highlights every sentence the detector considers AI-feel. Green sentences passed every signal. Yellow sentences tripped one or two. Red sentences tripped three or more. Most essays have three to seven sentences carrying most of the AI signal, and those are what you act on.
Open your essay alongside the highlights and rewrite the flagged sentences before you reach for any tool. Read each one aloud. Replace one abstract claim per paragraph with a specific example, number, or named detail. Vary sentence length so two adjacent sentences are not both in the 18 to 24 word range that detectors weight. If a flagged sentence is one you genuinely cannot rewrite (a technical definition, a citation lead-in), run the AI rewriter in Light mode on just that sentence rather than the whole paragraph.
Paste the revised draft back into the detector and re-scan. Aim for above 70 on graded essays, above 80 if you want margin against the false-positive rate of whichever detector your institution uses downstream. If a single sentence still flags red, go back to step 3 for that one sentence; do not run another Maximum-mode rewrite pass on the whole essay. Then submit through your normal channel. TextSight does not interact with Turnitin and we make no promises about specific detector outcomes; we report our own score honestly so you can decide whether the draft is ready.
A number on its own does not tell you whether to submit. These five bands describe what the classifier is seeing, what downstream detectors like Turnitin tend to do with the same essay, and what the right next move is at each band.
The classifier sees almost no AI patterns. Risk of a false positive on Turnitin or any other downstream detector is low. Most professors will read this as a normal student essay. The few yellow sentences you may still see in the colour map are usually fine on their own and reflect normal academic register rather than AI tells. Submit through your normal channel and move on.
Acceptable for most submissions but worth a quick review of the flagged sentences. If they are sentences you wrote and they happen to land in AI-aligned patterns (common for non-native English writers and for highly-structured academic prose), they are usually fine. If they are sentences you generated with ChatGPT or copied from an AI outline, revise them in your own voice before you submit. Two or three targeted rewrites usually move the score into the 80 to 90 range.
The classifier sees a meaningful share of AI patterns. Turnitin will probably flag this essay in the 20 to 50 percent range, which triggers a closer instructor read at most universities. Either revise the red sentences manually or run the AI rewriter in Light mode on the stubborn ones. Target above 75 before you resubmit. Look at where the red sentences cluster: if they group in two paragraphs, you have a section-level problem; if they scatter evenly, you have a vocabulary or rhythm problem.
The essay reads AI to the classifier. Turnitin will likely flag this in the 50 to 70 percent range, which usually triggers a formal academic-integrity review at most institutions. The fix here is structural, not cosmetic. Restructure the paragraphs (merge two body paragraphs, open with a specific claim instead of a transition word, vary sentence length deliberately) and rewrite the affected sections from your own thinking. A single rewrite pass will not move a score in this band into safe territory.
Almost certainly raw or lightly-edited model output. The fix is a full restructure plus a rewrite from your own notes, not a quick edit. If the essay was AI-generated end to end, the honest move is to write it yourself or accept that no AI rewriter will get it past a careful read by your instructor. TextSight will help you check your own writing and revise authentic drafts; it will not laundromat generated text into a passing essay.
Different essay structures default to different AI signals. The scorer detects the format from the draft and weights the relevant signal more heavily, so the score reflects the actual risk for the kind of essay you submitted, not a generic average.
Introduction, three body paragraphs, conclusion. The skeleton itself is the strongest tell because ChatGPT writes this format perfectly and detectors learn that pattern. The scorer weights paragraph-structure templating more heavily here. Break the symmetry by merging two body paragraphs, opening with a specific claim instead of a thesis preview, or dropping transition words at the start of body paragraphs entirely.
Claim, counterclaim, rebuttal. The default AI tell is uniform hedging: the assistant softens every claim and every rebuttal with the same registers ("It could be argued that", "However, it is also worth considering"). The scorer weights hedge-density variance more heavily here. Sharpen the claim sentences. Pick a side and let the prose show that you have. Hedge in your own voice (a personal observation, a specific limit) rather than the assistant polite default.
The tell is perfect symmetry: the assistant mirrors every point on subject A with an exactly equivalent point on subject B. Real student writing is asymmetric because you know one side better. The scorer weights paragraph structure plus sentence-length variance more heavily here. Break the mirror by adding one extra detail on the side you actually know, or by skipping one of the symmetric points entirely if the comparison does not need it.
The tell is generic activist register ("We must take action", "It is imperative that society"). The scorer weights vocabulary fingerprint more heavily here. Swap these for specific calls anchored to a real audience: who exactly should do what, by when, and what stops them today. Specificity reads human because it reflects thinking rather than the assistant default closing template.
The tell is missing sensory detail. The assistant defaults to abstractions ("The experience was meaningful", "It taught me an important lesson"). The scorer weights sentence-length variance plus vocabulary fingerprint more heavily here. Insert a concrete sensory detail per paragraph: a smell, a name, a number, the colour of a room, the time on a clock. Narrative reads human when it shows a specific moment, not a reflection on one.
All three AI rewriter 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 →
If your first language is not English, the detector risk is structurally higher because non-native prose patterns overlap with the patterns detectors learn from AI output. The scorer on this page is calibrated for that, not against it.
In our internal evals against a sample of human-written ESL essays, the average competitor detector returned a false-positive rate around 18 to 22 percent. The TextSight 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. The score you see is the same score paid users see; there is no preview-tier gating that would make ESL false positives worse than they need to be.
The same four steps, with the emphasis on step 3 (revising the flagged paragraphs) rather than step 4 (re-scanning). Read each flagged sentence aloud; ESL writers gain more from this exercise than native writers because it surfaces sentences where formal academic register collided with non-native phrasing in a way that reads AI to the classifier. When you reach a rewrite pass, the tool defaults to Light mode and adjusts vocabulary away from idiomatic native-speaker phrasing so your second-language voice stays intact rather than getting flattened toward a native register you do not use.
We do not try to make ESL essays sound like native-speaker essays. That would erase the writer. The goal is the same authentic-voice goal as for any other student: catch sentences where assistant register leaked in, revise them in your own voice (including your second-language voice), and submit a draft that reads like you wrote it. The score is a draft check, not a fluency exam.
An honest draft check is closer to a careful proofread than to anything else. We want to be explicit about which side of the academic-integrity line this scorer sits on, so you can decide whether it fits the context you are submitting into.
Essays you wrote yourself. The thinking is yours, the argument is yours, the structure is yours. The scorer catches sentences where assistant register or uniform academic register leaked into the prose so the submitted draft reads in your own voice. The four steps on this page are designed to build a skill: the ability to read your own writing the way a detector reads it, then revise. We score honestly so you can decide what the draft needs.
It is not a Turnitin workaround. We make no promise that any specific essay will get past Turnitin or any other downstream detector, and we will not help disguise generated content as your own. If your essay is mostly AI-generated and only lightly edited, the scan will tell you that and no AI rewriter pass will magically fix it; it cannot put authentic thinking into a draft that was not yours. The score and the highlights are diagnostic, not laundering.
The habits this workflow teaches (read aloud, vary sentence length, prefer specific examples, recognise the AI tells in your own format) transfer to every essay you write for the rest of your education. A score-reduction tool that works today probably stops working in a year as detectors update. A revision habit that reads your own writing critically is yours forever. We would rather build the second.
The full student workflow, .edu discount, and how to use TextSight inside your academic policy.
Open student page →Five-step voice workflow with the AI tells specific to each essay format and the exercises that move scores.
Open the guide →The companion guide for essays where you used ChatGPT for outline, idea unblock, or editing assistance.
Open the guide →How the 0 to 100 score is computed and what threshold to aim for before you submit a graded essay.
Read the guide →Authenticity Score, sentence-by-sentence colour map, ESL-aware calibration, citations preserved. .edu Pro at $13.99.