Test log
RewriteAI test data
Raw results for the RewriteAI 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
- 8
- Screenshots
- 9
- Method
- Published methodology
- Environment
- Google Chrome on desktop
Results
| ID | Test | Key measurement | Score |
|---|---|---|---|
| T01 | Detection-Bypass Rate | Passages rewritten: 3 | 3/5 |
| T02 | Coverage Across Detectors | Methodology detectors run: 0 of 4 | 3/5 |
| T03 | Meaning & Fact Preservation | Planted facts kept: 20 of 24 | 1/5 |
| T04 | Readability & Grammar | Rewrites with an error or artefact: 3 of 3 | 1/5 |
| T05 | Non-English Support | Rewrites that kept the meaning: 1 of 3 | 2/5 |
| T06 | Input, Length & Output Support | Points that worked: 4 of 10 | 2/5 |
| T07 | Humanizing Speed | Press to finished rewrite: about 2 s | 5/5 |
| T08 | Pricing & Free-Tier Limits | Free words a month: 500 | 3/5 |
| Overall (in the review) | 2.7/5 | ||
Detection-Bypass Rate
3/5Not Tested- Method
- Each AI passage was pasted once and the first rewrite kept.
- Passages rewritten
- 3
- Methodology detectors run
- 0 of 4
- Rewrites ZeroGPT read as human
- 2 of 3
- Remark
- Scale 5 is 44 or more of 50 rewrites read as human by all four detectors. No grade can be read from zero of four detectors, so the default 3 applies.
Raw notes
Each AI passage was pasted once and the first rewrite kept. The rewrite was then meant to go to GPTZero, Originality.ai, Copyleaks and Turnitin, the four detectors of the methodology, and count as bypassed only if all four read it as human.
None of the four could be used. GPTZero's guest box took the text but pressing Scan did nothing. Originality.ai has no free scan page and Turnitin sells only to schools. I did not open Copyleaks' guest page.
So this test is Not Tested and counts as 3 by default. For information only, ZeroGPT read the three originals as AI (100% each) and the rewrites as 0%, 0% and 84.1% AI. Sapling read the originals 96.6%, 99.3% and 95.6% AI and the rewrites 80.6%, 99.8% and 49.6% AI. RewriteAI's own detector read the originals 100%, 71% and 100% AI and the rewrites 28%, 27% and 19%, each marked Human.
- Three passages: three rewrites cannot confirm or refute the site's bypass claim, and the three outside detectors I could reach disagreed with each other.
Coverage Across Detectors
3/5Not Tested- Method
- The rewrites are split by detector so the toughest one sets the grade.
- Methodology detectors run
- 0 of 4
- Own detector: rewrites read as human
- 3 of 3
- Sapling: rewrites read as human
- 1 of 3
- Remark
- Not Tested, counted as 3 by default, because none of the four named detectors could be run. The three I could run are shown for the reader and not graded.
Raw notes
The rewrites are split by detector so the toughest one sets the grade.
Per detector, rewrites read as human out of 3: ZeroGPT 2; RewriteAI's own detector 3; Sapling 1 (the email, at 49.6% and marked Mixed); GPTZero, Originality.ai, Copyleaks and Turnitin not run.
RewriteAI's own detector is the only one that read all three as human, and it is the vendor's tool. Sapling, which is independent, read the flood plan and remote work rewrites at 80.6% and 99.8% AI.
- Free guest pages can use a different model from a paid product of the same name.
Meaning & Fact Preservation
1/5Tested- Method
- Three passages carried eight facts each (24 in all), written before the tool was opened: a news item about a flood plan, a short essay quoting a survey and a researcher, and a delay email with an order number.
- Planted facts kept
- 20 of 24
- Passages with a damaged fact
- 3 of 3
- Remark
- Scale 1 is under 14 of 30 clean, or facts routinely changed or invented. No passage came back clean, and a wrong reason in a customer email is the kind of error the scale exists to catch.
Raw notes
Three passages carried eight facts each (24 in all), written before the tool was opened: a news item about a flood plan, a short essay quoting a survey and a researcher, and a delay email with an order number. Each first rewrite was read against its list.
Flood plan: 6 of 8 kept. "Councillor Marcus Lindqvist" disappeared, and the reason, that raised barriers will hold back water during storms, became "the construction of barriers along the river".
Remote work: 7 of 8 kept. Every name, date and figure survived, but the researcher's claim that collaboration can suffer without face-to-face contact became "such collaboration is often needed to the success of many workplaces".
Delay email: 7 of 8 kept. The shipment "was held at customs" became "was held at habits", so the reason for the delay is now wrong, and the rewrite ends with a "[Your Name]" placeholder.
Option 2 repeated the "habits" error in the email, and kept the councillor in the flood plan text.
- Three passages, not 30. The scale counts clean passages out of 30, so I applied it to the share of passages with a damaged fact: 3 of 3.
Readability & Grammar
1/5Tested- Method
- Each first rewrite was read in full for grammar, spelling and sense, and checked for hidden or look-alike characters.
- Rewrites with an error or artefact
- 3 of 3
- Characters outside plain text
- 0
- Remark
- Scale 1 is more than 22 of 50 broken. Three of three is 100 percent, so the grade is 1; the sample is three.
Raw notes
Each first rewrite was read in full for grammar, spelling and sense, and checked for hidden or look-alike characters.
All three had a clear error the original did not: "a important" in the flood plan and remote work rewrites, "such collaboration is often needed to the success" in the second, and "held at habits" plus the "[Your Name]" placeholder in the email. Option 2 repeated "a important" and "habits".
No character outside plain keyboard text was found in any of the three rewrites, so the errors are plain mistakes, not hidden characters.
- Three rewrites of 85 to 103 words.
Non-English Support
2/5Tested- Method
- One AI passage of 30 to 47 words, with a name, a date, a figure and a reason, was written in each of Spanish, French and German, rewritten once and read against its facts.
- Rewrites that kept the meaning
- 1 of 3
- Languages tried
- 3
- Remark
- Scale 2 is 3 to 5 of 12; one of three is the same share (about 4 of 12). Bypass was not measured, only meaning and grammar.
Raw notes
One AI passage of 30 to 47 words, with a name, a date, a figure and a reason, was written in each of Spanish, French and German, rewritten once and read against its facts. The detectors were not run on them.
Spanish: date, name, 40 million euros and 20 percent kept, but the reason (dedicated bus lanes) became "the time that buses spend travelling through the city streets", which says something different.
French: date, name, 25 million euros and 30 percent kept, but the reason (longer opening hours) was dropped.
German: every fact and the reason kept, with a slightly awkward last clause ("da alle Dächer von Schulgebäuden Solaranlagen installiert werden").
The site's own FAQ says the tool is "optimized to work with English language" and that results in other languages "may not be as accurate".
- Three passages, no Chinese, no detectors, and no native reader for the German phrasing.
Input, Length & Output Support
2/5Tested- Method
- Each of the ten points was tried on the account used: paste, upload, batch, a 2,000-word cap, a choice of mode, formatting kept, a built-in detector, an extension, an API and a copy of the rewrite.
- Points that worked
- 4 of 10
- Words allowed per input
- 300
- Remark
- Scale 2 is 4 or 5 points.
Raw notes
Each of the ten points was tried on the account used: paste, upload, batch, a 2,000-word cap, a choice of mode, formatting kept, a built-in detector, an extension, an API and a copy of the rewrite.
Worked: paste; formatting kept (the email came back with its greeting, paragraph break and sign-off); a built-in AI detector (a free page on the site that gave a reading in about two seconds); the Copy Text button, which put the rewrite on the clipboard. That is 4 of 10.
Not counted: no upload button or batch box on the humanizer; the free cap is 300 words an input (1,500 on Pro); no mode choice on the humanizer on the free plan (the Advanced engine and Academic mode start on Pro); no browser extension seen; the API is documented on the account page with a key form and a POST /api/v1/humanize example, but I did not create a key or call it.
- Some points may work on a paid plan; the methodology counts only what works on the account used.
Humanizing Speed
5/5Tested- Method
- The page was polled every quarter of a second from pressing Humanize and the time of the first poll that showed "Done.
- Press to finished rewrite
- about 2 s
- Slowest rewrite
- under 3.6 s
- Rewrites timed
- 6
- Remark
- Scale 5 is a median under 8 seconds. The median of six is about 2 seconds.
Raw notes
The page was polled every quarter of a second from pressing Humanize and the time of the first poll that showed "Done. Your result is ready." was kept. The first rewrite was also recorded on screen.
Six rewrites. Four were ready at the first reading after about 1.9 seconds, and the other two (the remote work essay and the Spanish passage) by about 3.5 seconds at most. The screen shows "Humanizing your text" with an Analyzing spinner and then the result.
- Six passages of 30 to 103 words on the free plan; the poll step is a quarter of a second and the log was read every 1.5 seconds, so the times are read to about a second and a half.
Pricing & Free-Tier Limits
3/5Tested- Method
- The free plan was used until I stopped, short of 85 percent of the allowance; the plans were read from the account's billing page and the site's pricing section, in both views, on the test date.
- Free words a month
- 500
- Free rewrites I ran
- 6
- Remark
- The free tier is real but small, and the scale's top grades need 500 words a day or a paid plan under $12 that lifts limits and adds the strongest mode. Basic is under $12 but keeps the basic engine, so I set 3.
Raw notes
The free plan was used until I stopped, short of 85 percent of the allowance; the plans were read from the account's billing page and the site's pricing section, in both views, on the test date.
The free plan is 500 words a month and 300 words an input, renewing on 1 December 2026. It gives the Basic Humanization Engine and two options per rewrite. Six rewrites used 391 words; the balance counts the words you put in, not the words that come out.
Basic is $9.99 a month, or $5.99 a month billed yearly, for 5,000 words a month and 750 an input on the same Basic engine. Pro is $24.99, or $14.99 yearly, for 30,000 words and 1,500 an input, and adds the Advanced engine, Academic mode and "Bypass all AI detectors". Max is $49.99, or $29.99 yearly, for 200,000 words and 3,000 an input.
On Basic, 1,000 words cost about $2 on the monthly price and $1.20 on the yearly price. On Pro, about 83 cents and 50 cents. A free monthly allowance of 500 words is about 17 words a day.
The account page also offers up to 30,000 extra words for a short video about the tool posted to TikTok, Instagram or YouTube (10,000 for X, LinkedIn or Facebook); I did not try it.
- No plan was bought; checkout was not opened, so I could not check how the words renew or whether unused words roll over.
Coverage
Tested
- Three English AI passages rewritten once each on the free plan, with 24 planted facts
- One passage each in Spanish, French and German
- The rewrites read again in ZeroGPT and Sapling, in RewriteAI's own detector page and in GPTZero
- The word balance before and after, the history page, the copy button, the API page and the plans
Not tested or limited
- Six rewrites came from 500 free words, so every result is from three English passages, not 50
- GPTZero's guest scan took the text but pressing Scan did nothing; Originality.ai has no free page, Turnitin sells only to schools, and I did not open Copyleaks' guest page. The four detectors of the methodology were not run
- ZeroGPT, Sapling and RewriteAI's own detector are not among the four, so their readings are reported but do not set a grade
- Naturalness and plagiarism were left out: they need many rewrites and a plagiarism checker
- The languages test uses three passages (one each in Spanish, French and German), not twelve, and no Chinese; the detectors were not run on them
- Upload, batch, a 2,000-word run, a mode choice and the Advanced engine are not on the free plan, and the API was read but not called
- The English passages were written by one model (Claude), not six
- Trustpilot has no page for rewriteai.com (it returned a 404 on 6 November 2026), so there is no user rating to report
Still to test
- Fifty passages from six models on a paid plan
- The four named detectors, including Originality.ai and Turnitin
- The Advanced engine and Academic mode on Pro
- Plagiarism introduced, on a plagiarism checker
- Twelve passages in four languages
- Naturalness read blind beside human controls
- Upload, batch, a 2,000-word run and the API