A genuinely free ChatGPT detector with sentence-level highlights, five ChatGPT-specific signals, and roughly 40 percent fewer false positives on ESL writing than the 2024 baseline. The first scan needs nothing: no email, no account, no card. Most "free" detectors in 2026 are either email-gated previews, ad-supported result panels, or hard-locked at 250 words. TextSight's free tier was scoped to actually be usable, indefinitely, for the one essay or one article you need to check.
Honest accounting of what is in the free tier and what is not. No "actually that feature is paid" footnotes mid-workflow.
The counter resets at midnight UTC. No carry-over between days. For a student checking one essay before submission, this covers a draft, a revision, and a final pre-submission pass. For a freelancer checking client deliverables, it covers a typical day's output. The cap is generous enough that most casual users never hit it, and small enough to keep the free tier sustainable against abuse.
About 800 words, the standard college essay length. Longer pieces split into sections, scanned one at a time within the daily three-scan budget. Pro raises the per-scan cap to 10,000 characters and unlocks file upload for PDF, DOCX, and TXT so dissertation chapters and long-form articles do not need to be chunked by hand.
Every scan returns green, yellow, and red on each sentence so you see exactly which lines drove the score. Most competitor free tiers hide this behind a paid plan and show only a single overall percentage. TextSight's free tier includes sentence highlights by default because a percentage without evidence is hard to act on.
The classifier was tuned against millions of ChatGPT samples and weights five fingerprints: tripled adjective stacks, transition phrase clusters, the "delve into" vocabulary cluster, polite-assistant openers, and summary closers. The full set is in the next section; the takeaway is that the free tier scores against the same fingerprint paid users get, not a stripped-down classifier.
One paste returns the AI versus human score, a 0 to 100 Authenticity Score, and a Plagiarism Risk read flagging stock phrasings, citation-risky claims, and generic definitions. The bundled scan is the same scoring surface a paid user gets. Most free tiers in the category give one of the three; TextSight gives all three.
The first scan needs nothing. Signup is optional and only matters if you want 7-day scan history; it is not required to run the detector. The result panel has no banner ads, no interstitial upsell prompts pasted over the highlights, and no competing-tool advertising. The free tier is funded by paid plans, not by monetising attention on the result page.
Free and paid users score against the same five ChatGPT-specific patterns. The classifier weights them together at the sentence level so you see which lines triggered the flag.
"Clear, concise, and effective." "Robust, scalable, and reliable." "Comprehensive, detailed, and thoughtful." ChatGPT defaults to three-adjective groups in topic sentences at roughly four to six times the rate of human writers on equivalent topics. Humans use the rule of three sometimes; ChatGPT reaches for it almost every paragraph.
Furthermore, Moreover, In addition, Additionally, On the other hand, In conclusion. Human writers use these too, just not stacked across consecutive paragraphs. ChatGPT frequently opens four out of five paragraphs with a transition word, which produces a density signal independent of any individual phrase.
Delve, tapestry, navigate (as metaphor), multifaceted, robust, leverage, underscore, foster. These words appear in topic sentences and conclusions at roughly five to seven times the rate of human writing on the same topics. The cluster is loud enough on its own to flag a passage even when the structure looks varied.
"Certainly!", "Of course!", "I would be happy to help.", "Great question!", "Absolutely!". Even when these openers are stripped, the underlying register persists into the next sentence: a confident restatement of the prompt, then an outline of what the answer will cover. Humans usually start with the answer itself, not the meta-commentary about the answer.
"In conclusion, the interplay between..." "As we move forward, it is clear that..." "Ultimately, the path forward demands..." ChatGPT's closing paragraph almost always steps back to synthesise themes rather than ending on a specific claim. Closing sentences with this synthesis pattern, especially when paired with metaphor vocabulary (path forward, journey, landscape, tapestry), score as ChatGPT at high probability in TextSight's internal classification.
3 scans a day at 5,000 chars covers casual personal use indefinitely. Paid tiers scale into daily writing, API workflows, and team scanning. 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 →
Four steps from paste to verified rewrite. No account creation step, no email verification, no card. The free tier was scoped specifically for this 30-second loop.
Open app.textsight.ai. The text input field appears immediately, no signup wall. Paste up to 5,000 characters of the writing you want to check. A character counter ticks as you paste so you know if the passage needs to be split. Most college essays fit in one paste; dissertation chapters split into roughly six sections at the free cap.
Click Scan. The classifier runs in about six to ten seconds for an 800-word piece. The result panel shows an overall AI versus human score, the 0 to 100 Authenticity Score, and the Plagiarism Risk read in one view. Short passages under 300 words show a "directional, not precise" confidence note because structural signals are weaker on short text.
Scroll the result panel and read the colour-coded sentences. Green reads human, yellow reads borderline, red reads ChatGPT. The top three signals that fired on the passage are listed alongside (for example "tripled adjective stacks: 4 hits", "delve cluster: 2 hits"). These tell you which sentences to edit and why, not just that something is off.
No single detector is the final word. If the score reads borderline, paste the passage into a second detector (GPTZero, Originality.ai's free preview) and compare. Treat the agreement of two independent classifiers as a stronger signal than either one alone. The Rewrite button on TextSight rewrites flagged sentences in one click if you want to fix instead of cross-check; the first two uses are free.
The honest version of the AI-detector ESL problem and what the 2025 retrain actually fixed. No marketing hand-waving; specific numbers and a directional confidence note on short ESL passages.
AI detectors trained primarily on native English writing learned to read formal register, longer sentences, and lower contraction usage as AI signals. Non-native English writers often produce exactly those features for legitimate reasons (vocabulary safety, formal academic register, careful clause construction). The result through 2024 was a category-wide ESL false positive rate that flagged genuine human writing as ChatGPT at uncomfortable rates, particularly on student essays written by international students.
TextSight expanded the training corpus with roughly 1.2 million ESL writing samples from academic and professional sources across 18 source languages. The classifier learned to separate "formal register from a non-native writer" from "formal register from ChatGPT" by leaning more heavily on the five ChatGPT-specific signals listed above and less on raw register or sentence length. False positives on the ESL benchmark dropped from roughly 11 percent to roughly 6 to 7 percent, a 40 percent relative reduction.
An ESL writer running their own essay through the free detector in 2026 is meaningfully less likely to get a wrong AI flag than they would have been on the same tool in 2024. The "directional, not precise" confidence note on short passages is still shown so users know when the structural signal is weaker than usual. ESL writers should still cross-verify borderline scores with a second tool and read the per-sentence highlights to understand which lines drove the result.
Free covers the casual personal-use case indefinitely. Three patterns push users toward Starter, Pro, or Business within the first month or two.
Active freelancers, SEO writers with weekly deadlines, dissertation students editing chapters, and teachers running spot-checks across submissions all hit the 3 scans per day cap fast. Starter at $7.49 effective on annual lifts the cap to 20 scans per day and adds the Chrome extension. Pro at $14.99 effective on annual unlocks unlimited detector scans and the 10,000-char per-scan cap so long-form text does not need to be chunked.
Any workflow that wires the detector into another tool (CMS publishing pipeline, LMS integration, automated content QA, custom writing assistant) needs an API key. Starter and above include API access with per-tier monthly quotas. The free tier has no API path; paste-only on the web app is the entire surface.
Agencies running pre-publish QA on dozens of articles per week and university departments scanning student submissions in batches need the bulk endpoint that accepts an array of texts in one API call. The bulk endpoint is Business-only at $29.99 effective on annual, with 5 team seats and the shared scan history scoped for exactly this case. Free tier is one scan at a time through the web app.
The main ChatGPT detector landing page with the paid feature set and the full classifier write-up.
Open ChatGPT detector →Model-tuned classifier for the GPT-4 family specifically, with the five GPT-4 fingerprint signals.
Read the deep dive →Sister free tier for the AI rewriter side of the workflow: 1,500 words a month, all three modes.
Try the free AI rewriter →Full tier breakdown for Free, Starter, Pro, and Business. Annual billing saves 25%.
See pricing →3 scans a day, 5,000 chars per scan, sentence-level highlights, Authenticity Score and Plagiarism Risk in the same scan. Your first scan in about ten seconds.