Test log
StealthWriter test data
Raw results for the StealthWriter review. The verdict is in the review.
- Tested
- 7 November 2026
- Tester
- Miles Foster
- Account
- Free demo account, signed up by RankDude, nothing paid to the brand
- Tests
- 10
- Screenshots
- 6
- Method
- Published methodology
- Environment
- Google Chrome on desktop
Results
| ID | Test | Key measurement | Score |
|---|---|---|---|
| T01 | Detection-Bypass Rate | Fixed detectors that answered: 0 of 4 | n/a |
| T02 | Coverage Across Detectors | Fixed detectors that answered: 0 of 4 | n/a |
| T03 | Natural, Human Voice | Rewrites read blind: 0 of 20 | n/a |
| T04 | Plagiarism Introduced | Rewrites checked: 0 of 8 | n/a |
| T05 | Meaning & Fact Preservation | Rewrites that kept every fact: 0 of 8 | 1/5 |
| T06 | Readability & Grammar | Rewrites with a fault: 8 of 8 | 1/5 |
| T07 | Non-English Support | Rewrites with meaning and grammar intact: 0 of 8 | 1/5 |
| T08 | Input, Length & Output Support | Points met: 4 of 10 | 2/5 |
| T09 | Humanizing Speed | Median per rewrite: 1.2 s | 5/5 |
| T10 | Pricing & Free-Tier Limits | Free humanizations a day: 10 | 5/5 |
| Overall (in the review) | 2.5/5 | ||
Detection-Bypass Rate
n/aNot Tested- Method
- Planned: every rewrite scanned once by GPTZero, Originality.ai, Copyleaks and Turnitin.
- Fixed detectors that answered
- 0 of 4
- Built-in detector, rewrites read human
- 8 of 8
- Remark
- Not graded. The test needs all four fixed detectors and none gave a verdict.
Raw notes
Planned: every rewrite scanned once by GPTZero, Originality.ai, Copyleaks and Turnitin. Run: GPTZero and Copyleaks, on the public pages with no account.
GPTZero: I took the scan again once on the original and once on the rewrite of passage 1, one press of Scan each and no retry. The page stayed on the pasted text, with the Scan button and no verdict or message, so GPTZero gave no result. Copyleaks showed a Cloudflare "Verify you are human" box on the original of passage 1, and I stopped there. Originality.ai and Turnitin need accounts I do not hold.
StealthWriter's own detector read the original of passage 1 as 0% human ("AI Detected") and its rewrite as 100% human on engine V2. Its automatic scan after each rewrite, engine V1, read 90, 100, 100, 100, 100, 100, 100 and 88 percent human for the eight passages.
That is a vendor reading its own work, so it is shown and not graded.
- A bot check is not retried. The test needs the four fixed detectors on a plan or account that reaches them.
Coverage Across Detectors
n/aNot Tested- Method
- Planned: the bypass results split by detector, graded on the weakest.
- Fixed detectors that answered
- 0 of 4
- Remark
- Not graded.
Raw notes
Planned: the bypass results split by detector, graded on the weakest.
There is no per-detector table: GPTZero and Copyleaks gave no verdict, and the other two need accounts.
The built-in detector has two engines. V1 read the eight rewrites 88 to 100 percent human; V2 read the one rewrite I gave it 100 percent human.
- Needs the four fixed detectors.
Natural, Human Voice
n/aNot Tested- Method
- Twenty rewrites read blind beside ten human-written controls.
- Rewrites read blind
- 0 of 20
- Remark
- Not graded.
Raw notes
Twenty rewrites read blind beside ten human-written controls.
Not run: the free plan allows 10 rewrites a day, and a blind read needs 20 plus ten human controls.
- Run on a plan with more rewrites a day.
Plagiarism Introduced
n/aNot Tested- Method
- Every rewrite through one fixed plagiarism checker.
- Rewrites checked
- 0 of 8
- Remark
- Not graded.
Raw notes
Every rewrite through one fixed plagiarism checker.
Not run: the one checker I tried, Copyleaks, showed a bot check, and I did not retry.
- Needs a plagiarism checker I can reach.
Meaning & Fact Preservation
1/5Tested- Method
- Eight passages, each with five facts written in before any tool was opened, rewritten once at level 10.
- Rewrites that kept every fact
- 0 of 8
- Rewrites with an invented detail
- 4 of 8
- Remark
- Scale 1: under 14 of 30 clean, or facts routinely changed or invented. 0 of 8 is 0 percent.
Raw notes
Eight passages, each with five facts written in before any tool was opened, rewritten once at level 10. Each rewrite was read line by line against its fact list.
0 of 8 rewrites kept every fact and added nothing.
Invented text: passage 1 gained "Today's Tale of the Week therefore asks: What will the future of work be?" and turned "employees can control their own schedules" into "employers will not interrupt the work of employees". Passage 7 became "The Vireo, a new line of series 100 laptops".
Changed or lost facts: Martin Luther became "Marty Luther" (2); "90 percent recycled steel" became "90% of all bottles are recycled steel" (3); "20 minutes" disappeared and a one-minute run became "1 minute lunges" (5); the subject line with order 48213 was replaced by "To: maik [LINK]From: bossy [LINK]" (6); passage 8 invented "a mean of 81%" and "50%" of recalled pairs (8). The supporters' point in passage 4 became "comparable plans have got rid of Antwerp and Copenhagen contributed to a greater than 20 per cent rise in cycling".
- Eight passages, not 30. The result is the same on every one, so a larger sample is unlikely to change the grade.
Readability & Grammar
1/5Tested- Method
- All eight rewrites were read in full and marked clean or broken: a grammar or spelling error the original did not have, a mangled word, a sentence that no longer parses, or a hidden character.
- Rewrites with a fault
- 8 of 8
- Remark
- Scale 1: more than 22 of 50 broken. 8 of 8 is 100 percent.
Raw notes
All eight rewrites were read in full and marked clean or broken: a grammar or spelling error the original did not have, a mangled word, a sentence that no longer parses, or a hidden character.
8 of 8 rewrites were broken.
Examples: "the insulated stainless blaster" for a stainless steel bottle; "1 minute lunges followed by 2 minutes of walking"; "comparable plans have got rid of Antwerp and Copenhagen contributed to"; "Warm up each session for 5 minutes and relaxation is imperative"; "Gutenburg" misspelled.
No hidden or look-alike characters were found: the only non-standard characters were dashes, the euro sign and one inch mark. The FAQ says the output has "no typos, no broken sentences".
- Eight passages, not 50.
Non-English Support
1/5Tested- Method
- Eight passages written before the tool was opened, each with five planted facts: two in Spanish, three in French, three in German.
- Rewrites with meaning and grammar intact
- 0 of 8
- Rewrites with English words inside
- 6 of 8
- Remark
- Graded 1 because none of eight rewrites kept its meaning and its language intact; the scale gives a 1 to two or fewer of twelve.
Raw notes
Eight passages written before the tool was opened, each with five planted facts: two in Spanish, three in French, three in German. Each was pasted once at level 10 on Ghost 5.2 Mini and the first rewrite kept, then read against its facts and for grammar in that language.
None of the eight rewrites came back clean. Six had English words dropped into the French, German or Spanish text, for example "Investment costs were 3.5 million euros, and were required because the cranes are outdated" in the middle of a German paragraph, and "Subject to some 180 pupils, each week" in the French school story.
Facts broke too: the Spanish doctor "Javier" became "Javiera", a German library event gained a church that was never in the text ("in die Kirche"), a French bakery story gained "la quesada", and the German running event lost its month ("Am 15." then a new sentence). Others had broken grammar, such as "une potager" and "daron".
StealthWriter's own detector read all eight as human on its automatic scan (100% human). No other detector was used, so that is the vendor marking its own work.
The FAQ says "StealthWriter supports many languages beyond English — including Spanish, French, German, Italian, Portuguese, Dutch, and more."
- Eight passages in three languages, no human controls, and only the vendor's own detector, so bypass is not graded separately.
Input, Length & Output Support
2/5Tested- Method
- Each of the ten points was checked on the free plan on the test date, by doing it or by reading the FAQ answer.
- Points met
- 4 of 10
- Remark
- Scale 2: 4 or 5 of 10.
Raw notes
Each of the ten points was checked on the free plan on the test date, by doing it or by reading the FAQ answer.
Met (4): paste a passage; a choice of strength (levels 1 to 10, and the Drafts setting); paragraphs kept through the rewrite; a built-in detector.
Not met (6): no file upload on the humanizer page; no batch; the free cap is 1,000 words, below the 2,000 the scale asks for (paid plans allow 5,000); no browser extension and no API: the FAQ says there is no public API and that a browser extension is only on the roadmap; the Copy button was not tested because the daily allowance was used up.
Formatting held for paragraphs but not for the email's subject line, which was replaced by invented "To:" and "From:" lines.
- Upload and batch were looked for on the pages I could reach; a paid plan may add them.
Humanizing Speed
5/5Tested- Method
- Each of the eight rewrites was timed from pressing Humanize to the finished rewrite on screen, with a script that checks the page four times a second.
- Median per rewrite
- 1.2 s
- Slowest rewrite
- 1.6 s
- Remark
- Scale 5: a median under 8 seconds.
Raw notes
Each of the eight rewrites was timed from pressing Humanize to the finished rewrite on screen, with a script that checks the page four times a second.
Median 1.2 seconds, fastest 0.8, slowest 1.6.
No rewrite failed or queued on the free plan.
- Eight rewrites of about 180 words, not 50 of 300. Longer text may take longer.
Pricing & Free-Tier Limits
5/5Tested- Method
- The free tier was used until I reached 80 percent of its daily allowance, and the plans page was read in both billing views on 6 November 2026.
- Free humanizations a day
- 10
- Cheapest paid plan, yearly
- $16.67 a month
- Remark
- Scale 5: the free tier gives 10 rewrites of up to 1,000 words a day, well above the 500 words a day the scale asks for. The strongest model, Ghost 5.2 Pro, sits on the paid plans, which start at $16.67 a month billed yearly.
Raw notes
The free tier was used until I reached 80 percent of its daily allowance, and the plans page was read in both billing views on 6 November 2026.
Free: 10 humanizations and 10 scans a day, up to 1,000 words each, so up to 10,000 words a day, with Ghost 5.2 Mini only. No card was asked for.
Starter, the cheapest plan with Ghost 5.2 Pro and Legacy, is $20 a month, or $16.67 a month paid yearly. It gives 50 humanizations a day and 5,000 words an input.
Plus, Pro, Business and Scale rise to 150, 350, 800 and more humanizations a day.
- No plan was bought; Ghost 5.2 Pro was not tested.
Coverage
Tested
- Eight of the ten fixed passages, level 10, Ghost 5.2 Mini, first rewrite kept
- Every rewrite read against the five facts written into its passage
- Time from pressing Humanize to the finished rewrite, eight times
- The built-in detector on the original and on the rewrite of passage 1, and its automatic scan of every rewrite
- GPTZero on the original and on the rewrite of passage 1, one scan each, and Copyleaks on the original, once
- The plans page read in both billing views, and the FAQ page with six answers opened
- Eight short passages in Spanish, French and German (two, three and three), level 10, Ghost 5.2 Mini, first rewrite kept
Not tested or limited
- The free plan allows 10 humanizations a day: eight English passages (6 November) and eight non-English passages (7 November) were run, so every count is out of eight, not 50, 30 or 12
- GPTZero gave no verdict on either scan of passage 1 (the page stayed on the pasted text after Scan) and Copyleaks showed a Cloudflare "Verify you are human" box; I did not retry either, and Originality.ai and Turnitin need accounts, so the two detector tests are not graded
- Naturalness (a blind read against ten human controls) and plagiarism (no checker I could reach) were not run
- Non-English: three languages (Spanish, French, German), not four, because the FAQ names no Chinese; passages of 52 to 82 words, under the 100 words the page recommends; no human controls; the only detector for the language rewrites was StealthWriter's own, so the test is graded on kept meaning and correct language
- No plan was bought: Ghost 5.2 Pro, Ghost 5.2 Legacy and the 5,000-word input were not tested
- Result screenshots at 2x for the English rewrites were not possible after the fact: the free allowance was used up, so those pictures are from the detector page
Still to test
- The four fixed detectors on all 50 passages
- Naturalness and plagiarism on a paid plan
- Chinese and the other languages the FAQ lists
- Ghost 5.2 Pro and Legacy against Ghost 5.2 Mini
- Lower levels, which the FAQ says change less
- Non-English passages of 100 words or more, the length the page recommends