Test log
Originality.ai test data
Raw results for the Originality.ai 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
- 6
- Method
- Published methodology
- Environment
- Google Chrome on desktop
Results
| ID | Test | Key measurement | Score |
|---|---|---|---|
| T01 | Resistance to Humanisers & Paraphrasers | Rewrites scanned: 0 of 20 | n/a |
| T02 | AI Model Coverage | Models tested: 1 of 6 | n/a |
| T03 | Non-English Text | Passages scanned: 0 of 24 | n/a |
| T04 | Accuracy on AI-Written Text | AI passages caught: 12 of 12 | 5/5 |
| T05 | False Positives on Human Text | Human passages flagged: 0 of 7 | 5/5 |
| T06 | Mixed & Lightly-Edited Text | Mixed passages flagged: 6 of 6 | 4/5 |
| T07 | Document & File Support | Checklist points that worked: 6 of 10 | 3/5 |
| T08 | Score Clarity & Evidence | Checklist points met: 9 of 10 | 4/5 |
| T09 | Scan Speed | Median scan time: 5.5 s | 4/5 |
| T10 | Pricing & Free-Tier Limits | Free allowance: 10,000 words once | 4/5 |
| Overall (in the review) | 4.3/5 | ||
Resistance to Humanisers & Paraphrasers
n/aNot Tested- Method
- Twenty AI passages rewritten by three humanisers.
- Rewrites scanned
- 0 of 20
- Remark
- Not graded: no rewritten passage was scanned.
Raw notes
Twenty AI passages rewritten by three humanisers.
Not run: the three humanisers need their own accounts, and a free plan with 100 credits could not scan the rewrites.
- Run on the next review of this tool.
AI Model Coverage
n/aNot Tested- Method
- The 50 AI passages split across six models.
- Models tested
- 1 of 6
- Remark
- Not graded: a catch rate per model needs more than one model.
Raw notes
The 50 AI passages split across six models.
Not run: every AI passage was written by one model (Claude), so there is no per-model catch rate.
- The other five models were not reachable from the test setup.
Non-English Text
n/aNot Tested- Method
- Twenty-four passages in Spanish, French, German and Chinese.
- Passages scanned
- 0 of 24
- Remark
- Not graded: no non-English passage was scanned.
Raw notes
Twenty-four passages in Spanish, French, German and Chinese.
Not run: the passages were prepared (four AI and four human, in Spanish, French and German) but the credits ran out first.
- Run on the next review of this tool.
Accuracy on AI-Written Text
5/5Tested- Method
- 12 AI-written passages (two from each of six fixed prompts: vaccines, a water bottle, working from home, the First World War, a rainy city and interview advice), 104 to 113 words each, written by Claude on 6 November 2026 and not edited.
- AI passages caught
- 12 of 12
- AI percentage range
- 92 to 97
- Remark
- Scale 5: 47 or more of 50 caught. 12 of 12 is 100 percent, so the 5 band.
Raw notes
12 AI-written passages (two from each of six fixed prompts: vaccines, a water bottle, working from home, the First World War, a rainy city and interview advice), 104 to 113 words each, written by Claude on 6 November 2026 and not edited. Each pasted once and the first verdict taken.
All 12 were marked AI, at 92 to 97 percent AI (median 95).
No passage was marked mixed or original, and no verdict needed a threshold: the tool states 'Likely AI' with a percentage.
- 12 passages, not 50, and all from one model, written by me; a detector is easiest on the model it knows best.
False Positives on Human Text
5/5Tested- Method
- Seven passages from books first published before 1922 (Pride and Prejudice, Walden, the Autobiography of Benjamin Franklin, The Wealth of Nations, The Problems of Philosophy, Emerson's Essays and Great Expectations), 100 to 122 words each, one scan each.
- Human passages flagged
- 0 of 7
- Remark
- Scale 5: at most 1 of 50 flagged. 0 of 7 is 0 percent, so the 5 band. A sample of 7 cannot show a rate of 2 percent, so the grade stands only for these passages.
Raw notes
Seven passages from books first published before 1922 (Pride and Prejudice, Walden, the Autobiography of Benjamin Franklin, The Wealth of Nations, The Problems of Philosophy, Emerson's Essays and Great Expectations), 100 to 122 words each, one scan each.
All seven were marked 100 percent likely original, 0 percent AI. None was flagged.
One of them (The Wealth of Nations passage) was scanned three times and gave the same result every time.
- 7 passages, not 50, all old books: modern human writing, students and non-native writers are the real risk and were not tested.
- The other human scans ran while two copies of my own script shared one tab, and are not counted.
Mixed & Lightly-Edited Text
4/5Tested- Method
- Six passages: three AI drafts that I reworded by hand (sentences changed, facts and a personal detail added) and three where a human paragraph from a book is followed by an AI paragraph.
- Mixed passages flagged
- 6 of 6
- Highlights on the right part
- 2 of 3 joined
- Remark
- Scale 4: 14 to 16 of 20 flagged. All 6 were flagged, but the highlight was wrong on one of the three joined passages, so it falls short of the 5 band's accurate highlights.
Raw notes
Six passages: three AI drafts that I reworded by hand (sentences changed, facts and a personal detail added) and three where a human paragraph from a book is followed by an AI paragraph. Each scanned once.
All 6 were flagged as containing AI. The three edited AI drafts each scored 85 percent AI.
The three joined passages came back as 'Mixed' at 54, 40 and 43 percent AI. In two of them the human paragraph was green and the AI paragraph red. In the third, the human paragraph's last sentence (it has two) and the whole AI paragraph were marked in the same yellow.
- 6 passages, not 20.
Document & File Support
3/5Tested- Method
- The ten-point checklist was done on the free account: paste, PDF, DOCX, text file, batch, web address, a 2,000-word cap, an extension, the API and an export.
- Checklist points that worked
- 6 of 10
- Remark
- Scale 3: 6 or 7 of 10. A point counts only when it worked end to end on this account; the first run gave 4 and the retest gave 6.
Raw notes
The ten-point checklist was done on the free account: paste, PDF, DOCX, text file, batch, web address, a 2,000-word cap, an extension, the API and an export. A retest followed on the same day (the first DOCX was one my script built and the app could not read it): the files were made in LibreOffice and imported as text only, without pressing Scan.
Worked (6): paste; a .txt file through Import from File (the dialog says .txt, .doc, .docx and .pdf up to 10 MB); a real DOCX and a real PDF, both made in LibreOffice, imported as text on the retest; Import from URL loaded the text of a public Wikipedia page; a paste of 2,672 words was taken (a scan of it would cost 27 credits for AI only).
Not counted (4): the export, because Download PDF opened a three-page report page with Chrome's print dialog and saved no file by itself, and a share link is not a file; Bulk Scan opened but was not run; the Chrome extension was not installed; the API is Enterprise only ('Upgrade your plan to access API Keys'). My first DOCX, which my own script built, was refused by the app as unreadable, so that fault may have been mine; the DOCX made in LibreOffice imported without a problem.
- Bulk Scan, the extension and a saved PDF file were not proven; the pricing page lists bulk scan as a Pro feature. Imported files were not scanned, so I did not see the scan result for a Word or PDF file.
Score Clarity & Evidence
4/5Tested- Method
- The ten points were checked on the scans already run, plus the report and history screens, and one passage was scanned three times.
- Checklist points met
- 9 of 10
- Remark
- Scale 4: 8 or 9 of 10.
Raw notes
The ten points were checked on the scans already run, plus the report and history screens, and one passage was scanned three times.
Met (9): a plain call ('Likely AI', 'Likely Original' or 'Mixed'); a percentage; sentence-by-sentence highlighting; a short explanation in words ('This score suggests 96% of your text was created with AI'); a share link and a print-ready report page; a scan history; the same result three times in a row (0 percent AI, 100 percent original, 3 scans); a statement of limits ('AI detectors are highly accurate but they are not perfect') and a 'Think this is False Positive?' link; no upsell over the result.
Not met (1): no margin or confidence band for a result. The gauge shows fixed marks and a setting for how much AI to allow, not how sure the tool is.
- Repeatability was one passage three times, not many.
Scan Speed
4/5Tested- Method
- Each of the 12 AI scans was timed from the click on Scan to the verdict on screen, by script, on one connection.
- Median scan time
- 5.5 s
- Slowest scan
- 6.7 s
- Remark
- Scale 4: a median of 5 to 9.9 seconds. The AI passages were 104 to 113 words; long documents were not timed.
Raw notes
Each of the 12 AI scans was timed from the click on Scan to the verdict on screen, by script, on one connection.
Median 5.5 seconds, average 5.5, fastest 4.5, slowest 6.7.
No scan failed or queued on the free plan.
- 12 scans, not 50. Each time includes the moment my script took to notice the verdict.
Pricing & Free-Tier Limits
4/5Tested- Method
- The free plan was used until I stopped at 84 of 100 credits, and the public pricing page was read in both billing views.
- Free allowance
- 10,000 words once
- Cost of 1,000 words
- 10 credits
- Remark
- Scale 4: a usable free tier and a paid plan from $10 to $19.99 a month.
Raw notes
The free plan was used until I stopped at 84 of 100 credits, and the public pricing page was read in both billing views.
Free: 100 credits once, no daily refill. A credit is 100 words, rounded up per scan, so 1,000 words cost 10 credits and my 100 to 135 word passages cost 1 or 2. Scan History, sharing and file and web address import worked on free; Bulk Scan and the API sit on the paid plans.
Pro is $14.95 a month ($12.95 a month billed yearly) for 2,000 credits, 200,000 words, a month, with bulk scan and more history. Enterprise is $179 a month ($136.58 billed yearly) for 15,000 credits and the API.
- No plan was bought; the pay-as-you-go credit price was not read.
Coverage
Tested
- 12 AI-written passages of 104 to 113 words, one scan each, timed from pressing Scan to the verdict
- 7 human-written passages from books published before 1922, one scan each
- 6 mixed passages: 3 AI drafts with my edits and 3 human paragraphs joined to AI paragraphs
- One passage scanned three times for repeatability
- Import from a text file, a Word file, a PDF and a web address (text import only, not scanned), the 2,672-word paste, Share Report and the Download PDF button, Scan History, and the Bulk Scan and API Keys screens
- The plans page read in both billing views, and the Trustpilot page
Not tested or limited
- The free plan's 100 credits paid for 42 scans before I stopped at 84 credits: the full 50 AI, 50 human, 20 mixed and 24 non-English passages were not run
- Every AI passage was written by one model (Claude), so the model coverage test could not be graded; the humaniser test needs a humaniser account I do not hold
- The other human passages were scanned while two copies of my own script were running against the same browser tab, so their texts could have been mixed together; none of them is counted, which leaves 7 human passages
- The non-English passages were not scanned: the credits ran out
- Bulk Scan, the Chrome extension and the API were not confirmed to work on the free plan; I retested file and web address import after the first run, and imported but did not scan the files, to save credits
- Human passages are old books, which are easier for a detector than modern writing
Still to test
- Humaniser resistance with rewrites from three named tools
- AI model coverage with passages from five more models
- Non-English text in Spanish, French, German and Chinese
- A larger human sample, including modern and non-native writing
- Bulk scans and the Chrome extension, and a scan of an imported Word or PDF file