Test log
AI Text Tools test data
Raw results for the AI Text Tools 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
- 7
- Screenshots
- 9
- Method
- Published methodology
- Environment
- Google Chrome on desktop
Results
| ID | Test | Key measurement | Score |
|---|---|---|---|
| T01 | Detection-Bypass Rate | Passages rewritten: 4 | 3/5 |
| T02 | Coverage Across Detectors | Detectors that read it as human: 1 of 2 rewrites for ZeroGPT, 2 of 2 for Sapling | 3/5 |
| T03 | Meaning & Fact Preservation | Passages rewritten: 4 | 2/5 |
| T04 | Readability & Grammar | Rewrites with an error or artefact: 4 of 4 | 1/5 |
| T05 | Input, Length & Output Support | Points that worked: 2 of 10 | 1/5 |
| T06 | Humanizing Speed | Seconds to a rewrite: 14.7 median | 4/5 |
| T07 | Pricing & Free-Tier Limits | Free humanizer words a month: 1,000 | 3/5 |
| Overall (in the review) | 2.4/5 | ||
Detection-Bypass Rate
3/5Not Tested- Method
- Each AI passage was pasted once on the Basic setting and the first rewrite kept.
- Passages rewritten
- 4
- Methodology detectors run
- 0 of 4
- 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 on the Basic setting 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 my text but Scan returned no result. Copyleaks' guest page showed a Cloudflare "Verify you are human" box, and I did not pass or retry it. Originality.ai has no free scan page and Turnitin sells only to schools.
So this test is Not Tested and counts as 3 by default. For information only: ZeroGPT read the first rewrite 0% AI (the original read 100%) but read the second rewrite 100% AI (the original also 100%). Sapling read the two rewrites 0.8% and 10.8% AI (the originals 64.7% and 99.5%). The tool's own detector read the first rewrite 4.3% AI (the original 72.8%).
- Four passages and no named detector. The two outside detectors I could reach disagreed with each other on the second rewrite.
Coverage Across Detectors
3/5Not Tested- Method
- The rewrite is split by detector so the toughest one sets the grade.
- Detectors that read it as human
- 1 of 2 rewrites for ZeroGPT, 2 of 2 for Sapling
- Methodology detectors run
- 0 of 4
- Remark
- Not Tested, counted as 3 by default, because none of the four named detectors could be run. The ones I could run are shown for the reader and not graded.
Raw notes
The rewrite is split by detector so the toughest one sets the grade.
Per detector, original then rewrite, on two passages (Brightwater, then Valdoria): the tool's own detector 72.8% then 4.3% (Brightwater only); ZeroGPT 100% then 0%, and 100% then 100%; Sapling 64.7% then 0.8%, and 99.5% then 10.8%; GPTZero not run (no result); Copyleaks not run (bot check); Originality.ai and Turnitin not run (no free access).
The built-in detector scored the rewrite far lower than ZeroGPT scored the second rewrite, which ZeroGPT called AI in every sentence. A checker made by the same company as the humanizer is not an independent reading.
- Free guest pages can use a different model from a paid product of the same name. The detector words came out of a separate 1,000-word detector allowance (131 used).
Meaning & Fact Preservation
2/5Tested- Method
- Each passage carried planted facts: names, dates, figures, a claim and the reason given for it.
- Passages rewritten
- 4
- Rewrites clean
- 1 of 4
- Numbers, names and dates kept
- 4 of 4
- Remark
- The scale counts clean rewrites out of 30. One clean in four is 25 percent, which on a strict count sits at grade 1. Every figure and name survived in all four, so I set grade 2 and say why.
Raw notes
Each passage carried planted facts: names, dates, figures, a claim and the reason given for it. Every rewrite was read against the original line by line, and counted clean only when every fact and the argument were unchanged.
All numbers, names and dates survived in all four rewrites: 14 March 2024 and $48 million, 42 kilometres and 8.5 million euros, 6 percent and 3,200 pupils, 240 chairs and 1,150 dollars.
One rewrite of four was clean (Valdoria). The Brightwater rewrite lost who said what: the mayor's statement became "As Mayor Elena Ruiz, the transit system will reduce average commute times by 18 percent". The Halvorsen rewrite added a sentence that is in no part of the original: "we believe that good sleep, health, and well-being should always be a priority for our children". The Northwind email turned "a delay at our Rotterdam warehouse" into "an overload at our Rotterdam warehouse".
- Four short passages. Facts kept here do not show they are kept every time.
Readability & Grammar
1/5Tested- Method
- All four rewrites were read in full for grammar, spelling and sense, and the saved text was checked for hidden or look-alike characters.
- Rewrites with an error or artefact
- 4 of 4
- Characters outside plain text
- 0
- Remark
- Scale 1 is more than 22 of 50 broken. Four of four is 100 percent, so the grade is 1; the sample is four.
Raw notes
All four rewrites were read in full for grammar, spelling and sense, and the saved text was checked for hidden or look-alike characters.
No hidden or non-standard character in any rewrite. Every rewrite still carried a broken sentence or an odd phrase: "As Mayor Elena Ruiz, the transit system will reduce...", "first, and foremost sourced from a regional transport grant", "In a study conducted by the researchers at the Halvorsen Institute in June 2023 revealed an increase", and an email whose subject line runs into its greeting ("Subject: Northwind order Dear Mr. Okafor").
- Four rewrites of 66 to 108 words on the free Basic setting. The paid Premium mode may read differently.
Input, Length & Output Support
1/5Tested- 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 an export.
- Points that worked
- 2 of 10
- Remark
- Scale 1 is 3 or fewer points, or paste-only with a tight cap.
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 an export.
Worked: paste; a built-in detector tab. That is 2 of 10.
Not counted: Upload File opened the "Unlock Premium Mode" window, although the pricing page lists Word and PDF upload under Free; no batch box was shown and bulk processing is listed under Exclusive; the free cap is 300 words, and a 320-word paste greyed out the button; only Basic is open of nine settings, so there is no choice of strength; formatting was not tested because my passages were single paragraphs; no extension was shown; the API is an enterprise offer by email; the copy and download icons were seen, not used.
- Some points may work on a paid plan; the methodology counts only what works on the account used.
- The copy and download icons were seen but not pressed, and no passage with headings or lists was pasted, so formatting and export are unconfirmed rather than failed.
Humanizing Speed
4/5Tested- Method
- The screen was recorded while I pressed Humanize and the time was read from the press to the first moment the finished rewrite showed in the output box.
- Seconds to a rewrite
- 14.7 median
- Rewrites timed
- 4
- Remark
- Scale 4 is a median of 8 to 14.9 seconds. The median of the four is 14.7 seconds, with the free-plan countdown counted because the text is not shown until it ends.
Raw notes
The screen was recorded while I pressed Humanize and the time was read from the press to the first moment the finished rewrite showed in the output box.
The four rewrites took about 13, 14.5, 14.8 and 17.4 seconds. The page showed a countdown of 10 seconds ("Free users wait 10 seconds. Upgrade to Pro for instant results."), which began 2 to 7 seconds after each press, so the work itself took 2 to 7 seconds.
- Four passages of 65 to 112 words on the free plan. The paid plans list instant results.
Pricing & Free-Tier Limits
3/5Tested- Method
- The free plan was used and its meters read on the dashboard before and after; the pricing page was read on the test date.
- Free humanizer words a month
- 1,000
- Words used in my tests
- 350
- Remark
- Scale 3 is a tight free tier or a paid plan of $20 to $29.99. The free tier is about 33 words a day with a 10-second wait; Lite is under $12 but does not include bulk, which sits on the top plan.
Raw notes
The free plan was used and its meters read on the dashboard before and after; the pricing page was read on the test date.
Free gives 1,000 words a month of humanizing, 1,000 of detection and 1,000 of AI generation, up to 300 words a request, with a 10-second wait and no saved history. After my four rewrites (350 words) the dashboard showed 650 humanizer words left, and 869 detector words left after 131 words of detection.
Lite is $9.99 a month (shown as reduced from $12.99), for 40,000 words, 1,200 words a request and no wait. Pro is $12.99 (reduced from $15.99) for 80,000 words and 2,000 a request. Exclusive is $19.99 (reduced from $24.99) for 170,000 humanizer words, 3,000 a request and bulk processing. A "2 days left" tag sat on Pro.
On Lite, 1,000 humanizer words cost about 25 cents, on Pro about 16 cents and on Exclusive about 12 cents.
- No plan was bought and checkout was not opened. The paid plans were read from the pricing page, with a note that the Lite plan is described as a weekly plan but priced per month.
Coverage
Tested
- Four rewrites of AI passages with planted facts, one on a news item, one on a city project, one on a school study and one on a customer email
- The built-in detector on the first original and its rewrite, ZeroGPT and Sapling on two originals and two rewrites, and attempts at GPTZero and Copyleaks
- The 300-word cap, Upload File, the mode and language menus and the dashboard meters
- The pricing page and the Trustpilot page, read on the test date
Not tested or limited
- The free plan allows 300 words a request and 1,000 words a month, so every result is from four passages, not 50
- GPTZero's guest box took the text but returned no result, and Copyleaks showed a Cloudflare "Verify you are human" box that I did not pass; Originality.ai and Turnitin have no free page, so the four detectors of the methodology were not run
- ZeroGPT and Sapling are not among the four, so their readings are reported but are not used for a grade
- Only the Basic setting is open on the free plan, so the strongest mode of the methodology was not tested
- Naturalness, plagiarism and non-English were left out: they need many rewrites, a plagiarism checker and the locked languages
- Upload File, history, Premium mode and the other eight settings are locked on the free plan, so batch, file input and export were not confirmed
- Passages came from my own corpus, not from six models
Still to test
- Fifty passages from six models on a paid plan
- The four named detectors, including Originality.ai and Turnitin
- Plagiarism introduced, on a plagiarism checker
- Twelve passages in four languages on a plan that unlocks them
- Naturalness read blind beside human controls
- Premium mode, file upload, batch and an export