Most universities now run every submission through an AI check before a human ever reads it. A flag at 20 percent can mean a referral, a meeting, or worse. The fix is not to hope your essay reads human, it is to check your own draft first, see which sentences flag, and rewrite the specific lines that read AI. TextSight does that in around 30 seconds: paste up to 5,000 characters, get an Authenticity Score, sentence-level highlights for every flagged line, Plagiarism Risk, and a one-click AI rewriter if you want to fix the red lines without rewriting from scratch. Free tier, no signup needed for your first scan, 3 scans every day.
The free tier is built for the actual student workflow, not a teaser. You get every detection mode and the same sentence-level evidence paid users get. The paywall is volume, not features.
That covers a standard 800-word college essay plus a revision pass before you submit. The free quota resets at midnight UTC, so a busy assignment week with three classes is still inside the free band as long as each draft fits a single scan. No card, no email collection, no progress lock. If you write longer than 5,000 characters in one go, split the essay into two scans or upgrade to Pro for 10,000 characters per scan.
Every scan returns the Authenticity Score on a 0 to 100 scale, the AI likelihood band, and the Plagiarism Risk score in the same pass. You do not have to pick a mode or upgrade for the second score: a single click runs all three so you see whether the essay is failing on AI-style cadence, soft plagiarism patterns, or both. Most student drafts only have one of those two problems, but knowing which one drives the score is what makes the revision targeted instead of guessed.
This is the difference between a number and a fix. Every flagged sentence is colour-coded green, amber, or red, and clicking a flag tells you the reason: AI-style cadence, stock phrasing, uncited specific, or stock textbook definition. You know which line to rewrite and why. The free tier ships the full highlight set, not a paywalled preview.
An AI detector that scores every essay the same way misses format-specific tells. TextSight detects the format first, then weights the signal that actually betrays AI in that format.
The skeleton is the tell. Intro with three signposted claims, three body paragraphs with topic sentence plus three supporting points, conclusion that restates the thesis. A real student writing this format breaks the symmetry: one body paragraph runs long, another lands a memorable line, the conclusion drifts. AI almost never breaks the skeleton. The checker weights template uniformity heavily on this format so a too-perfect structure flags even with otherwise human prose.
The hedging gives it away. Uniform hedge density, "many scholars argue", "it can be argued that", every claim wrapped in the same modal verb. Human writers vary their commitment level paragraph to paragraph; AI flattens the register. The checker tracks hedge frequency and modal-verb variance, then weights both higher when it detects argumentative scaffolding.
The symmetry is the tell. AI writes comparison essays with mathematically clean parallel structure: each side gets exactly the same number of supporting points, presented in identical order. Human writers play favourites or run long on whichever side they care about more. Perfect parallelism on a comparison essay is one of the strongest single-signal flags.
Generic activist register is the tell. "We must act now", "the time has come", "society must come together". Pulled from RLHF training that rewards safe persuasive tone. Real student persuasive writing names specific actors, specific policies, specific costs. The checker flags the generic register pattern and treats specificity as the human signal.
Missing sensory detail is the tell. AI narrative defaults to event sequencing and emotion labels: "I felt happy. Then I felt sad." Human narrative anchors in physical detail: the smell of the room, the colour of the light, what someone was wearing. The checker weights sensory-density on detected narrative paragraphs, and a flat emotion-label narrative flags hard even if the grammar is clean.
Free includes 3 scans a day with all three detection modes and sentence highlights bundled in. Paid tiers raise the per-scan character limit and add file upload, the Chrome extension, and the REST API. Yearly billing saves 25%. Students with a .edu email get Pro at $13.99/mo.
Billed $89.88/year — Save $30
Billed $179.88/year — Save $60
Billed $359.88/year — Save $120
Students with a .edu email get Pro at $13.99/mo instead of $19.99. View full pricing
The whole loop runs in around 15 minutes for a standard essay. The point is to make it fast enough that you actually do it the night before submission instead of skipping it.
Open the scanner and paste up to 5,000 characters on the free tier, around 800 words. That fits most college essay assignments in a single scan. No signup is required for the first scan; the daily quota resets at midnight UTC. If your essay runs longer, split into two sections or upgrade to Pro for 10,000 characters per scan plus PDF and DOCX upload.
One click runs AI detection, Authenticity Score, Plagiarism Risk, and citation parsing in the same pass. Results land in around 30 seconds for a standard essay. The result panel shows three score gauges side by side plus the full sentence-level colour map underneath. No mode picker, no upsell screen between paste and result, no "verify your email to see your score" wall.
The colour map is the actionable part of the scan. Green sentences read human. Amber sentences are borderline. Red sentences read strongly AI to the classifier. Click into any flagged sentence to see the reason: AI-style cadence, stock phrasing, uncited specific, or stock textbook definition. You always know which line to work on and why, and you can keep sentences you wrote even if they happen to land in the AI pattern.
For each red sentence you have three choices: edit it yourself, click Rewrite and let the tool rewrite it for you, or cut it if it was filler. Re-scan to confirm the score moved into the safe band (70 plus for graded work, 80 plus if you want margin). Most essays clear in one revision pass. A heavily AI-assisted draft might need two. The Starter and Pro tiers exist for exactly this rhythm of scan-revise-rescan across multiple drafts.
The hidden tax most AI detectors charge: they over-flag essays from non-native English writers because formal ESL prose patterns overlap with what detectors learn from AI output. TextSight is calibrated against an ESL writing sample so the score is gentler for you, not stricter.
Formal English-as-a-second-language prose is structurally clean: precise grammar, balanced sentence length, careful vocabulary, controlled hedge density. Those are exactly the patterns AI defaults to as well, because both were trained on the same kind of formal source material. A detector tuned only on native-speaker writing reads ESL prose as AI and flags it hard. Indian, Filipino, Nigerian, and Chinese students see this constantly, and the penalty for being a good writer in your second language ends up being an academic misconduct meeting.
The classifier is trained against a dedicated ESL writing corpus alongside native-speaker text, and the sentence-level evidence layer exposes which signals fired. An ESL student can keep clean structured paragraphs that an averaging detector would mark red, because the highlight panel shows that the flag was triggered by sentence-length uniformity, not by genuine AI cadence. In our internal evals, the calibration returns roughly 40 percent fewer false positives on non-native English than the average competitor. The Authenticity Score reflects this calibration on every tier including free.
Click into the flag and read the reason. If the trigger was sentence-length uniformity, vary the lengths in that paragraph: one short sentence, one long, one medium. If the trigger was hedge density, drop one of the modal verbs and commit to the claim directly. If the trigger was generic register, swap one stock phrasing for a specific detail from the source you are discussing. None of these changes require sounding less academic, only sounding more like you.
Turnitin runs inside your university submission system after you hand the essay in. TextSight runs on your own draft before. Both have a role, and the honest workflow uses them in that order.
Turnitin runs its own AI classifier inside the submission portal your university or college uses. You do not choose when it scans, you do not see the per-sentence breakdown, and you do not get a chance to revise once the report exists. The score lands on your professor first, and by then the conversation has already shifted from your essay to your integrity. Turnitin is doing the job your institution asked it to do; the problem is that you only see the result after the only useful editing window has closed.
TextSight runs on your own draft, on your own laptop, before any submission portal sees the file. You see the same kind of score, plus the sentence-level evidence Turnitin does not show you, plus a built-in rewriting option for the flagged lines. The Authenticity Score correlates with Turnitin within roughly 5 to 10 percentage points in our internal testing, so a green TextSight result is a strong signal you will pass the Turnitin pass cleanly. Use TextSight to revise, then submit, then let Turnitin do its job knowing your essay reads human to a machine.
The honest setup is to run TextSight on every meaningful draft during writing and let Turnitin do the final certification at submission time. That is the same pattern most professional writers settle into: a lightweight continuous tool for iteration plus a heavier final check at the gate. Three jobs, two tools, no pretence that either one replaces the other.
The full student workflow plus the verified .edu discount on Pro at $13.99/mo.
See the student planThe standalone scoring page with ESL calibration and the 0 to 100 Authenticity metric explained.
Open the scorerRewrite flagged sentences in one click. Citations preserved, three rewrite modes, no meaning drift on Light or Balanced.
Open the AI rewriterStylistic plagiarism patterns and citation-risk flags bundled into the same scan as the AI detector.
Read the explainerFree, no signup, 3 scans every day, 5,000 characters per scan. All three detection modes plus sentence-level highlights in every pass. Students with a .edu email get Pro at $13.99/mo.