v4 MODEL Calibrated · low false positives Detects ChatGPT · Claude · Gemini · Llama

Detect AI Text. Humanize What Sounds AI.

TextSight detects AI-written sentences, shows your Humanization Score, and helps you rewrite the parts that sound unnatural with the AI Humanizer. Detect and humanize in one workflow.

Free to try · No card · No signup for detection

1M+ texts analyzed Sentence-level AI detection AI Detector + Humanizer

NEW Now on the Chrome Web Store — detect AI on any page you’re reading

0 / 1,000 words · 0 chars ·

AI DETECTION RESULT AI HUMANIZER IDLE

—%

AI PROBABILITY

Paste text to begin

Per-sentence breakdown share of sentences in each band

AI-generated—%

Mixed / edited—%

Human-written—%

SIGNALS

Perplexity—

Burstiness—

Sentence variance—

Vocabulary range—

Readability grade—

Paste the text that sounds AI-written, then press Humanize. Up to 300 words, and it spends one of your 3 free checks a day.

How it works

From AI detection to natural writing in one workflow.

No setup. No card. 3 free checks a day, spend them detecting or humanizing. No signup.

1 PASTE

Paste or upload

Paste your text, or upload a Word, PDF, .txt or .md document.

Pasted 412 words

essay_draft_v2.docx · 412 words · 2.1 KB

2 DETECT

See what reads as AI

You get an AI probability, a Humanization Score, and a flag on every sentence that reads as AI-written.

AI probability — 82%

Humanization Score 18 / 100 · 3 sentences flagged

3 HUMANIZE

Rewrite what sounds off

Send the flagged passages to the AI Humanizer, or rewrite them one sentence at a time. Leave the rest of your draft alone.

Humanized 3 flagged sentences

Your meaning and facts carried over

4 RECHECK

Run it again and compare

Scan the improved version in the same place and see what moved. No second tool, no copy-paste between tabs.

Rechecked — AI probability 82% → 27%

Humanization Score 18 → 73 / 100

The same paragraph, before and after

See what changes when AI-sounding writing becomes more natural.

Drag the handle. Left is raw ChatGPT output. Right is the same paragraph after humanizing, with the same meaning and the same facts. Both sides were scored by TextSight's own detector.

Before · ChatGPT output 82% AI

It is important to note that climate change is a multifaceted issue that requires a comprehensive approach. Furthermore, the impacts are far-reaching and affect various aspects of our daily lives. In conclusion, addressing this critical issue necessitates collective action.

Humanization Score 18 / 100 3 of 3 sentences flagged

After · TextSight humanized 27% AI

Climate change isn't one problem — it's a stack of them, and they don't sort themselves out. The effects show up everywhere: what we eat, how we travel, where insurance still works. Fixing it is going to take all of us, pulling in roughly the same direction, for a while.

Humanization Score 73 / 100 Same meaning, same facts

Example result, measured on TextSight's own detector on 7 October 2026. Scores are probabilistic and vary by text. TextSight is built to improve naturalness and readability, not to guarantee a particular result from any detector.

Drag the handle to compare Try this on your own text

What's inside

More than a verdict — a complete workflow.

Detection, scoring, humanizing and rechecking, in one place and without switching tools.

Sentence-level AI detection

See exactly which sentences are most likely to read as AI-written. Every sentence gets its own confidence score, colour-coded inside your text, so you know what to fix instead of guessing.

In today's rapidly evolving digital landscape, organizations must leverage cutting-edge technologies to remain competitive. Our team shipped the new dashboard on a Tuesday — nobody slept much that week.

Humanization Score

A simple 0–100 score showing how natural your writing reads. Higher is more natural, and it moves as you edit.

Raw ChatGPT paragraph18 / 100

After humanizing73 / 100

Scored onEvery scan

AI Humanizer

Detected something that sounds AI-written? Humanize the flagged sentences while preserving your meaning and tone.

Tones5

Free, per rewrite260 words

Works onFlagged sentences

Rewrite Suggestions

Improve only the sentences that need work instead of rewriting the whole document. Each suggestion is one click to apply.

ScopeFlagged sentences only

To applyOne click

Before it landsReview every rewrite

Detect → Humanize → Recheck

Find AI-sounding sentences, improve them, then run the text through TextSight again without switching tools.

Step 1Detect

Step 2Humanize

Step 3Recheck

Privacy-first analysis

Your scans save to your history so you can reopen them, and one switch turns that off. We never sell your text, and excluding it from model training is on by default. How we handle your text.

Scan historyYour choice

Excluded from trainingOn by default

See all features

Need more writing tools? Explore all TextSight tools →

Our own testing

We publish the evidence behind our detector.

We ran 1,180 human-written academic papers from 2018 — before GPT-3 existed — through our own detector. Every flag it produced is a false positive by construction. The dataset, the script and the limitations are published, and these are our own tests rather than independent third-party validation.

1,180 human papers

Published in 2018, so nothing in the corpus could have been AI-written. The whole set went through the detector.

See the benchmark →

5.85% false positives

That is how often we wrongly flagged human writing, with a 95% confidence interval of 4.65 to 7.34.

See the method →

Public methodology

How the detector scores text, what it is bad at, and why we treat a low false-positive rate as the first objective.

Read the methodology →

Downloadable data

Per-document results as a CSV, with the analysis script, published under CC BY 4.0 so anyone can recompute it.

Download the dataset →

Built for

Whatever you write, whoever reads it.

The same detect-and-humanize workflow, four very different jobs.

Students

Check essays for AI-like passages and improve awkward wording while keeping your own argument and ideas.

Free tier, no card needed
Academic tone when you humanize
Sentence-level flags, not one verdict

For students →

Writers & bloggers

Find robotic or repetitive sentences and rewrite them without losing your voice.

Voice-preserving rewrite
Free Chrome extension
40,000 AI Humanizer words/mo on Starter

For writers →

SEO & marketers

Review AI-assisted content and improve clarity, naturalness, consistency and brand voice before publishing.

Unlimited scans from Pro
80,000 AI Humanizer words/month on Pro
REST API, 5,000 calls/month on Pro

For SEO teams →

Educators

Review sentence-level AI indicators while treating detector scores as one signal rather than proof.

Use any score as one signal, not proof. Talk to the student before you act.

Sentence-level evidence
Published detection methodology
False-positive rate measured and published

For educators →

NEW Now on the Chrome Web Store

The detector, in the tab you’re already in.

Select text anywhere on the web — a Gmail thread, a shared doc, a ChatGPT transcript, a draft you are about to send — and the TextSight side panel hands back the same Humanization Score and the same flagged sentences you get on this page, then rewrites them in place.

3 free scans a day, no sign-up. Sign in and the extension uses your TextSight plan instead, so there is one quota across web, browser and mobile.
Works on the text you select. Pasting a long draft or uploading a PDF or DOCX stays in the web app.
Password and payment fields are never read, and nothing leaves the page until you ask for a scan.

Add to Chrome See how it works

Free to install · detector is English-only · the same package also installs on Edge, Brave, Arc, Vivaldi and Opera

Hi Priya,

In today’s rapidly evolving digital landscape, it is imperative that we leverage synergistic opportunities to drive meaningful outcomes.

Let me know what you think.

TextSight

23/ 100

“rapidly evolving digital landscape” — reads as AI

Rewrite in place

Where the workflow ends

One workflow instead of two tools.

Most AI detectors stop at the verdict. TextSight carries on into the rewrite and the recheck — see how the AI detector and humanizer work together.

	TextSight	Typical AI detector
AI detection		Yes
Sentence-level flags		Varies
Humanization Score		Usually no
AI Humanizer		Usually separate
Rewrite flagged sentences		Usually no
Recheck in the same workflow		Usually separate

A general picture of the category, not a scorecard of any one product. See full comparison →

Pricing

Free to try. Fair as you grow.

Cancel anytime. No card to start. Detection, scoring, and rewriting included on every paid plan.

Monthly Annual Save 25%

Free

$0/forever

For students, hobbyists, and curious humans.

3 actions a day — scans, uploads and rewrites share one bucket
13,500 characters a scan (about 1,500 words)
260 words per AI Humanizer rewrite
AI detection & Humanization Score
No credit card required

Start free

Starter

$7.49/month

Billed $89.88/year — Save $30

For light, regular use.

20 scans a day, 45,000 characters a scan
40,000 AI Humanizer words / month
Humanization Score + before/after
Unlimited scan history
Email support

Start Starter

Questions, answered.

Is TextSight free?

Yes. On this page you get 3 free checks a day with no signup, and it is one allowance: spend them detecting, humanizing, or a mix of both. A free account also gives you 3 actions a day, shared between scans, file uploads and rewrites, with 13,500 characters a scan (about 1,500 words) and 260 words per AI Humanizer rewrite. There is no monthly cap and no credit card needed. Paid plans start at $7.49/mo billed yearly ($9.99 month to month).

How accurate is TextSight at detecting AI content?

On our own benchmark, TextSight is roughly 88–92% accurate on long-form English (300+ words) and 70–78% on short text under 100 words, where there is less to go on. We also ran 1,180 human-written papers from 2018 through the detector and wrongly flagged 5.85% of them (95% CI 4.65–7.34). No detector is 100% accurate, so treat any score as one signal, not proof. Read how we test.

What is the Humanization Score?

The Humanization Score is a 0–100 number showing how natural your writing reads. Higher is more natural. It is the counterpart to the AI probability on every scan, so you can watch it move as you edit. More on the Humanization Score.

What does the AI Humanizer do?

It rewrites the sentences that read as AI-written so they sound more natural, while preserving your meaning and tone. You pick a tone, review each rewrite before applying it, then recheck the text in the same place. It is built to improve naturalness and readability, not to guarantee a particular result from any detector. About the AI Humanizer, or try the free AI humanizer.

Can TextSight detect ChatGPT, Claude and Gemini?

Yes. Our test set includes text from ChatGPT (GPT-4 and GPT-4o), Claude, Gemini, Llama and Mistral, and it is refreshed with each release. GPT-5 is harder to catch than older models, especially in creative writing, so read its scores as a probability. Detection is optimized for English. Humanize ChatGPT text.

Can AI detectors be wrong?

Yes. Every detector, including TextSight, Turnitin, GPTZero and Originality.ai, makes mistakes. The most common is a false positive: human writing flagged as AI. The risk goes up with short text, very formal or formulaic writing, and writers whose first language isn’t English. That is why TextSight shows sentence-level evidence instead of a single verdict. Are AI detectors biased against ESL writers?

Does TextSight store my text?

Your text is not used to train our detection models — “Exclude my content from training” is on by default on every account. “Store my scan history” is also on by default and controls whether scans are saved; both switches are in Settings, and you can delete individual scans or wipe the whole history. We do not sell your text. How we handle your text.