RankDude

Test log

StealthGPT test data

Raw results for the StealthGPT review. The verdict is in the review.

Tested
6 November 2026
Tester
Miles Foster
Account
Free demo account, signed up by RankDude, nothing paid to the brand
Tests
10
Screenshots
7
Environment
Google Chrome on desktop, the StealthGPT web app

Results

IDTestKey measurementScore
T01Detection-Bypass RateRewrite shown: 32 of 285 words3/5
T02Coverage Across Detectors–3/5
T03Meaning & Fact Preservation–3/5
T04Readability & Grammar–3/5
T05Natural, Human Voice–3/5
T06Plagiarism Introduced–3/5
T07Non-English Support–3/5
T08Input, Length & Output SupportPoints met: 3 of 101/5
T09Humanizing SpeedTime to finish: 21.5 seconds3/5
T10Pricing & Free-Tier LimitsFree documents a month: 12/5
Overall (in the review)2.7/5
T01

Detection-Bypass Rate

3/5Not Tested
Method
The plan was to rewrite 50 passages that four detectors call AI and count the rewrites that all four read as human.
Rewrite shown
32 of 285 words
Original in Copyleaks
100% AI
Rewrites scanned by a detector
0
Remark
The free plan shows only the first sentence of the rewrite, so this test could not be run. It counts as 3 by default. Scale 5 needs 44 of 50 rewrites read as human by all four detectors; none could be counted.

Copyleaks: the Valdoria passage highlighted as AI text, with 100% AI content found, 112 AI words and 0 human words.

Copyleaks: the Valdoria passage highlighted as AI text, with 100% AI content found, 112 AI words and 0 human words.
T01-1 Copyleaks on the original passage: 100% AI text, 112 words.

StealthGPT: the first sentence of the rewrite, cut off with an ellipsis, above the text 'This is your text, undetectable. Pick a plan to reveal the rest.'

StealthGPT: the first sentence of the rewrite, cut off with an ellipsis, above the text 'This is your text, undetectable. Pick a plan to reveal the rest.'
T01-2 The humanizer page after one run: the first sentence, then a plan picker.
Raw notes

The plan was to rewrite 50 passages that four detectors call AI and count the rewrites that all four read as human. The free plan allows one document a month, so one document of three passages, 285 words, went to the strongest model, Super, once. Before humanizing, the first passage (112 words) was scanned in Copyleaks.

Copyleaks read the original passage as 100% AI: 112 words of AI text and none of human text, at sensitivity 2 of 3. The humanizer page then showed 32 of the 285 words of the rewrite, ending "criss-crossing si…", under "This is your text, undetectable. Pick a plan to reveal the rest and keep humanizing".

The result page gave the rewrite its own Stealth score of 99. That is StealthGPT's number, so it is not counted. The 32 visible words are 206 characters and Copyleaks needs 350, so the rewrite could not be scanned there. GPTZero showed "Hang on while we verify your browser" and I did not retry. Originality.ai needs an account and Turnitin has no individual access, so neither was used.

  • One document, three passages, one run, no retries. The passages were written by an AI model and carry invented names and figures.
T02

Coverage Across Detectors

3/5Not Tested
Method
The results of the bypass test are split by detector and the weakest one sets the grade.
Remark
The free plan shows only the first sentence of the rewrite, so this test could not be run. It counts as 3 by default.
Raw notes

The results of the bypass test are split by detector and the weakest one sets the grade.

No rewrite could be scanned by any detector, so there is nothing to split.

T03

Meaning & Fact Preservation

3/5Not Tested
Method
Each passage carries facts written in before any tool was opened: names, dates, figures, a claim and its reason.
Remark
The free plan shows only the first sentence of the rewrite, so this test could not be run. It counts as 3 by default.
Raw notes

Each passage carries facts written in before any tool was opened: names, dates, figures, a claim and its reason. A rewrite is clean only when every fact survives.

In the 32 visible words the 8.5 million euros, the regional transport grant, Valdoria and the 42 kilometres were kept, and "largely provided" says what "drawn mostly" said. March 2024 is not in the visible words and "six districts" is cut off. The count out of 30 could not be made.

  • Thirty-two words are too few to grade, so I have not graded them.
T04

Readability & Grammar

3/5Not Tested
Method
Every rewrite is read for grammar, spelling and hidden characters, and the broken ones are counted.
Remark
The free plan shows only the first sentence of the rewrite, so this test could not be run. It counts as 3 by default.
Raw notes

Every rewrite is read for grammar, spelling and hidden characters, and the broken ones are counted.

The first words read "Backed by an 8.5 million euros largely provided by a regional transport grant": "an" before a plural amount is a slip. Nothing else could be read.

T05

Natural, Human Voice

3/5Not Tested
Method
Twenty rewrites are read blind beside human-written text for voice and kept intent.
Remark
The free plan shows only the first sentence of the rewrite, so this test could not be run. It counts as 3 by default.
Raw notes

Twenty rewrites are read blind beside human-written text for voice and kept intent.

No full rewrite was available to read.

T06

Plagiarism Introduced

3/5Not Tested
Method
Every rewrite is run through one plagiarism checker and the newly flagged ones are counted.
Remark
The free plan shows only the first sentence of the rewrite, so this test could not be run. It counts as 3 by default.
Raw notes

Every rewrite is run through one plagiarism checker and the newly flagged ones are counted.

Only 32 words of the rewrite were visible, too few to check.

T07

Non-English Support

3/5Not Tested
Method
Three passages in each of Spanish, French, German and Chinese are humanized and checked.
Remark
The free plan shows only the first sentence of the rewrite, so this test could not be run. It counts as 3 by default.
Raw notes

Three passages in each of Spanish, French, German and Chinese are humanized and checked.

The one free document was already used. The pricing page says all languages are supported on paid plans, and I did not test that.

T08

Input, Length & Output Support

1/5Tested
Method
Ten points, each counted only when it works end to end on the account used: paste, upload, batch, a cap of 2,000 words or more, a choice of mode, formatting kept, a built-in detector, an extension, an API and an export.
Points met
3 of 10
Words per request on my plan
1,000
Built-in detector on the AI original
99% human
Remark
Scale 1: three points or fewer. Three points met.

The ten points on my free plan, 6 November 2026.

  • Paste a passage

    Result
    Met
    What I saw
    285 words pasted into the box
  • Upload a PDF or DOCX

    Result
    Not met
    What I saw
    The pricing page lists .doc and .txt on Pro and .pdf and images on Max; I did not upload on my plan
  • Batch

    Result
    Not met
    What I saw
    No batch screen found
  • 2,000 words or more per run

    Result
    Not met
    What I saw
    The counter read 1,000 words per request; 2,000 is on Max
  • Choice of mode

    Result
    Met
    What I saw
    Super, Standard and Lite in a menu
  • Formatting kept

    Result
    Not met
    What I saw
    The rewrite was hidden, so I could not see it
  • Built-in detector

    Result
    Met
    What I saw
    One scan worked, up to 200 words; the second scan asked for an upgrade
  • Extension or add-in

    Result
    Not met
    What I saw
    Listed on paid plans; not on mine
  • Documented API

    Result
    Not met
    What I saw
    Listed in the sidebar with its own plans; not on mine
  • Export that keeps formatting

    Result
    Not met
    What I saw
    Copy and Download sit on the result page; I could not use them on the hidden rewrite

StealthGPT AI Detector: the Valdoria passage with a Human Score of 99%, Likely human-written.

StealthGPT AI Detector: the Valdoria passage with a Human Score of 99%, Likely human-written.
T08-1 StealthGPT's own detector: 99% human for the passage Copyleaks read as 100% AI.

StealthGPT AI Detector: a box asking Need More AI Detection Scans with a See Plans button.

StealthGPT AI Detector: a box asking Need More AI Detection Scans with a See Plans button.
T08-2 The second scan: an upgrade box instead of a score.

StealthGPT humanizer with the model menu open: Super, Standard and Lite.

StealthGPT humanizer with the model menu open: Super, Standard and Lite.
T08-3 The model menu: Super, Standard and Lite.
Raw notes

Ten points, each counted only when it works end to end on the account used: paste, upload, batch, a cap of 2,000 words or more, a choice of mode, formatting kept, a built-in detector, an extension, an API and an export.

Three of the ten points worked on my free plan: pasting 285 words, a choice of three models (Super, Standard and Lite), and a built-in detector that takes up to 200 words. The detector worked once: my second scan the same day showed "Need More AI Detection Scans? Upgrade your plan to get unlimited AI detection scans".

The built-in detector read the same passage that Copyleaks called 100% AI as 99% human, "Likely human-written", on its Standard Scan.

  • The upload, formatting, batch and export points are counted as not met because I could not show them working, not because StealthGPT says it lacks them.
T09

Humanizing Speed

3/5Tested
Method
A script pressed Generate once and stopped its clock when the page had not changed for four seconds, with the screen recorded and a screenshot every five seconds.
Time to finish
21.5 seconds
Words sent
285
Remark
Scale 3: 15 to 29.9 seconds. One run, so the median is that one time.

StealthGPT: the progress screen at 32 percent with the step 'Humanizing text' running.

StealthGPT: the progress screen at 32 percent with the step 'Humanizing text' running.
T09-1 Five seconds in: 32%, on the third of eight steps.
Raw notes

A script pressed Generate once and stopped its clock when the page had not changed for four seconds, with the screen recorded and a screenshot every five seconds.

One document of 285 words took about 21.5 seconds from Generate to the finished page. The page showed a progress screen with eight steps, from "Request prepared" to "Run AI detection", and the words "This may take a minute or two". At five seconds it read 32%.

  • One run of one document. The median needs more runs than a free plan allows.
T10

Pricing & Free-Tier Limits

2/5Tested
Method
I used the free plan until it stopped me, then read every plan and billing view from the pricing page and the terms.
Free documents a month
1
Cheapest paid plan
$29.99 a month
Remark
Scale 2: no usable free tier. The free plan shows one sentence of one document a month, and the cheapest paid plan is in the $20 to $29.99 band.

StealthGPT plans, US prices, 12 November 2026

  • Pro

    Monthly
    $29.99
    Paid yearly, per month
    $25
    Words a month
    50,000
  • Max

    Monthly
    $49.99
    Paid yearly, per month
    $41.67
    Words a month
    200,000
  • Enterprise

    Monthly
    $179.99
    Paid yearly, per month
    $149.99
    Words a month
    1,000,000+

StealthGPT pricing page: Pro, Max and Enterprise with the Weekly, Monthly and Yearly switch and the first-week offer.

StealthGPT pricing page: Pro, Max and Enterprise with the Weekly, Monthly and Yearly switch and the first-week offer.
T10-1 The plans on the pricing page, monthly view, 12 November 2026.
Raw notes

I used the free plan until it stopped me, then read every plan and billing view from the pricing page and the terms.

The free plan allows one document a month, and after it the dashboard read 1 of 1. Pro is $29.99 a month, or $25 a month paid yearly, with 50,000 words a month and 1,000 words a request. Max is $49.99 a month, or $41.67 paid yearly, with 200,000 words and 2,000 a request. Enterprise is $179.99 a month, or $149.99 paid yearly. Weekly plans also exist, and new users are offered $1 for the first three days.

The terms say "All purchases are non-refundable". A separate "Detected? Get your money back" policy refunds the current billing period when you send a screenshot from a detector it accepts (Turnit In, GPTZero, Zero GPT, CopyLeaks), but not for a response its own stealth score flagged, and not for anything flagged by Originality.ai.

  • I did not buy a plan and did not take the three-day offer, so the paid plans were not used.

Other features checked

Coverage

Tested

  • One 285-word document, three passages, in the Super model
  • The progress screen, timed with a script and recorded
  • The built-in detector on an AI passage, and the same passage in Copyleaks
  • The free-plan meter and the plan picker
  • The pricing page in its three billing views, the terms and the Trustpilot page

Not tested or limited

  • The free plan allows one document a month and shows only the first sentence of the rewrite, so seven tests could not be run
  • Copyleaks needs 350 characters and the 32 visible words are 206; GPTZero showed a browser check that I did not retry; Originality.ai needs an account and Turnitin has no individual access
  • Upload, batch, export, the extension, the API and non-English text were not tried
  • I took no card step and did not take the three-day offer

Still to test

  • The full fifty-passage test on a paid plan, with all four detectors
  • Non-English text, upload and export on a paid plan
  • A retest of the built-in detector against Copyleaks on more passages
What these results meanRead the full StealthGPT review