Ghost Act

For students · Private AI checker

Check your essay for AI before anyone else does.

Paste your essay and see how it reads: which sentences a detector would flag, why, and whether the honest answer is “Uncertain”. It runs in your browser. Nothing is uploaded.

500 words free · no account · your essay never leaves this device

  • Never uploaded or stored
  • Never used for training
  • Never rewrites your work
  • Error rates published

Paste your essay, or drop a .docx, .pdf or .txt file. It is read in this browser and never uploaded.

Open the checker: 500 words free, nothing uploaded
  • 0 bytes uploaded: the model runs in this browser
  • 660 human texts and 24 AI models in the calibration
  • 5 independent 2026 datasets we never trained on
  • Every figure here is from the published validation

Example report

What you see after you paste.

Illustrative example · invented text

A sixth-form politics essay, 204 words (excerpt)

An illustrative excerpt we wrote in the formulaic style many schools teach. The report is Ghost Act’s real output for exactly this text.

57AI likelihood

Calibrated ensemble

Uncertain

57% AI likelihood for the whole text. 7 passages marked, each with the reason.

1In today's digital age, social media 2plays a crucial role in how young people form their political views. This essay will argue that platforms such as TikTok and Instagram have made teenagers more politically aware, but not necessarily better informed. It will look at where they get their news, how algorithms shape what they see, and whether any of it changes how they vote. Firstly, social media has changed where young people get their news. Ofcom's research suggests TikTok is now one of the main places 12 to 15 year olds see news, ahead of most newspapers. My younger brother is a good example of this, because he knew about the general election before anyone at home had mentioned it. 3However, what teenagers see is shaped by what they already watch. 4Furthermore, short videos rarely give the context behind a story. 5Moreover, a clip about a policy can be forty seconds long and still leave out who it actually affects. 6As a result, a student can feel informed after scrolling for ten minutes when in reality they have only seen one side. 7Overall, the evidence suggests that social media is a starting point for political interest, not a replacement for reading the news properly.

What to take from itThe mixed-authorship check found the intro reads very differently from the body: the first three sentences read as strongly AI-like, the rest at about 8%. The stock opener and stacked connectives drive the score; the details only this writer knows (the younger brother) read as human. So the honest verdict is “Uncertain”, with reasons you can see.

Check my essay free500 words free · no account · your essay never leaves this device

It is 11.40pm. The deadline is midnight.

You wrote every word. Three weeks of reading, a draft your flatmate marked up, and an introduction you rewrote four times because it still sounded stiff.

Then the thought arrives: what if the AI check says it isn’t yours? Your university runs one. You have read about the student whose results were withheld for six months before she was cleared. You have no idea what the detector looks for.

Ghost Act will not tell you what any other tool will say. It shows you how your writing reads, which sentences look formulaic and why, and whether the honest answer is “Uncertain”. You submit knowing, not guessing.

The problem

Detectors get it wrong. Students pay for it.

Even detector makers say a score should never be the only basis for action against a student. Students are still referred on a score alone, and it falls to them to explain.

  • ~4%

    Turnitin’s own sentence-level false-positive rate for its AI indicator. It also says results on documents under 300 words are less reliable and should be read with care.

    Inside Higher Ed (2023)
  • 61%

    of essays by non-native English writers were flagged as AI by seven detectors in a Stanford study. Essays by native writers were almost never flagged.

    Liang et al. (2023)
  • ~6,000

    misconduct referrals at one Australian university in 2024, about 90% of them about AI. One student waited six months to be cleared.

    ABC News (2025)
  • 38–80%

    how often light, guideline-compliant AI editing was flagged as AI in a 2026 study, more often than unedited recent writing (9–15%).

    Karr et al., arXiv (2026)

How Ghost Act helps

See what they might see, and exactly why.

“Where does my essay actually go?”

It never leaves your laptop

The language model runs inside your browser. Your essay is not uploaded, not stored on a server, not added to any database and never used to train anything. Saved reports stay on your device.

“Why would my writing get flagged?”

A reason for every flag

Every highlighted passage says what triggered it: a stock phrase, a stack of “Moreover”s, a run of same-length sentences, a rule-of-three list. Formulaic intros and conclusions are a common cause. Now you can see it.

“It says 40%. What does that even mean?”

An honest “Uncertain”

Between 29% and 78%, Ghost Act says “Uncertain” instead of guessing, because real human writing lands there too. Above that it flags; on held-out tests that catches 81% of AI text while wrongly flagging 3% of human writing.

“Is it fair to the way I write?”

Error rates, published by genre

We publish how often Ghost Act wrongly flags real writing: 7% of university essays and 0% of school essays; 1% of essays by English-language learners against 2% by native writers.

“Grammarly tidied one section”

Which parts read differently

The mixed-authorship check splits your document where the evidence changes, so you can see whether one edited paragraph reads unlike the rest instead of one number for the whole essay.

“I want something to show my tutor”

A report you can print

Every check is a full printable report: the likelihood, the reasons, the parts that differ and the limits. Context for a conversation, next to your drafts and version history.

Compared

Your essay deserves better than a black box.

Most AI checkers work by sending your text to someone else’s server. Here is what changes when nothing leaves your device.

FeatureGhost ActTypical cloud checker
Where your essay goesStays in your browser; nothing is kept or trained onUploaded to the vendor’s servers; some keep submissions
Why a passage was flaggedA named reason on every flagOften a highlight and a percentage
When the evidence is thinSays “Uncertain”Usually a score either way
False flags published by genreYes, 7% university essays, 0% school essaysRarely published
Tested on AI models it never trained onYes, 82% caught on unseen models (24 tested)Rarely published; usually one headline accuracy figure
Rewrites or “humanises” textNeverSome sell rewriting tools alongside

How it works

  1. 1

    Paste or drop your essay

    Word, PDF or plain text. It is analysed in your browser and never uploaded. The first 500 words are free, with no account.

  2. 2

    Reasons, not just a number

    See which passages look formulaic and why, whether one part reads differently, and whether the honest answer is “Uncertain”.

  3. 3

    Submit knowing, and keep your trail

    Keep your drafts, notes and version history, and save the report. If anyone ever asks how you wrote it, that is your context.

What Ghost Act is not

  • Not a Turnitin checker. It cannot predict or match what Turnitin or any other tool will say; different methods give different numbers.
  • Not proof you wrote it. No detector can prove authorship. Your drafts, notes and version history are the strongest evidence you have.
  • Not a humaniser. Ghost Act never rewrites your text and will not help anyone disguise AI writing.
  • Not perfect. It wrongly flags about 3% of human writing (7% of university essays) and catches less on independent 2026 datasets than in our tests.

Every figure on this page comes from Ghost Act’s published validation. How we measure

For students

500 words free

No account and no subscription. Need more? Pay once, from £4.99 for 5,000 words, and use the words whenever you need them.

  • 5,000 words for £4.99: a handful of essays
  • 25,000 words for £14.99: a full term of coursework
  • 100,000 words for £39.99: dissertation-scale analysis
  • Credits live in your browser; no account
  • The same full report as Pro
Check my essay free

Questions

Asked by students

How can I check my essay for AI before I submit it?

Paste it into Ghost Act or drop in the file. You get an AI-likelihood, the passages that read as AI-like and the reason for each, and a check for parts that read differently from the rest. It runs in your browser, so your essay is never uploaded. The first 500 words are free with no account. It will not tell you what your university’s detector will say; it shows how your writing reads and why.

Can Turnitin’s AI detector be wrong?

Yes, and Turnitin says so. It has reported that about 4% of sentences its AI indicator marks are human-written, that results on documents under 300 words are less reliable, and that it shows scores from 1% to 19% only as an asterisk. It also says a score should not be the sole basis for adverse action against a student. Some universities, including Vanderbilt and Waterloo, have switched its AI detection off.

Why is my writing flagged as AI when I wrote it myself?

The habits schools teach overlap with how language models write: stock openers, “Furthermore” and “Moreover” at the start of sentences, tidy three-part lists and evenly sized sentences. Writing in a second language and using grammar or editing tools can push scores up too. A 2026 study found light, permitted AI editing was flagged 38–80% of the time. Ghost Act shows which of these your essay has.

Will Ghost Act give the same score as Turnitin?

No, and it does not try to. Detectors use different models and thresholds, so their numbers differ. Ghost Act is for understanding how your writing reads and why, not for predicting another tool’s score.

How do I prove I didn’t use AI?

No detector can prove it, including this one. The strongest evidence is how you wrote the work: drafts, notes, sources and the version history in Google Docs or Word, plus being able to talk through your argument. A Ghost Act report adds context, showing which passages read as AI-like and why, but treat it as one piece of evidence among several.

I was accused of using AI when I didn’t. What should I do?

Stay calm and ask for the evidence, including the detector report. Gather your drafts and version history and explain how you wrote the work. Your students’ union can advise, and there is usually an appeal route. In England and Wales the Office of the Independent Adjudicator reviews complaints and has upheld cases where AI detection was relied on. This is general information, not legal advice.

Is my essay stored or used to train AI?

No. The analysis runs in your browser. Your essay is not uploaded, not stored on a server and never used for training. Saved reports live only in your browser, on your device.

I’m not a native English speaker. Am I more likely to be flagged?

A 2023 Stanford study found seven detectors flagged 61% of essays by non-native writers as AI; newer tools vary. On independent essays by English-language learners, Ghost Act flagged 1%; on essays by native writers, 2%. In England and Wales the student ombudsman has said universities should consider whether their assumptions could be biased against a student’s writing style.

Know how it reads before anyone else decides.

500 words free, no account, nothing uploaded. Paste your essay and see the reasons for yourself.

Check my essay free
500 words free · no account · your essay never leaves this deviceCheck my essay free