To check a PDF for AI writing, the short version is this: upload the file rather than pasting its text. A PDF carries structure that a paste destroys, and the structure is part of what you are reading. One upload a day with no account, up to 10MB, PDF or Word.
The longer version is about how to read the result, because a single percentage for a forty-page document is close to useless. A report is not uniformly anything: the methods section and the introduction were usually written under different conditions, and a document-level average hides exactly that.
The first step is the one people get wrong, and it costs the most.
1. Upload the actual file. PDF or .docx, up to 10MB. The extraction keeps paragraph boundaries, which is what the scoring needs: paste a document into a text box and you typically lose paragraph breaks, and a wall of text reads differently from the same words in paragraphs.
One practical note on what is checked: the upload is verified by content rather than by file extension. An image renamed to .pdf is rejected, and so is a spreadsheet or a slide deck renamed to .docx, with a message saying so. That is deliberate, because a scanned page is a picture of text and there is no prose in it to read.
2. Expect scanned PDFs to fail. This is the most common disappointment. If the PDF came from a scanner or a photograph, it contains images, not text, and there is nothing to extract. Run it through optical character recognition first and check the output, or retype the section you care about.
3. Read it by section, not as one number. More on that below.
4. Keep the original. If the document's authorship might be questioned, the file history matters more than any score. Do not overwrite it with an edited version.
Why a single percentage for a long document is the least useful form of the result.
Think about what a forty-page report actually contains. A literature review assembled from sources. A methods section written to a template. A discussion written at speed the night before. Those are different kinds of writing and they score differently, and averaging them produces a number that describes none of them.
Two consequences worth acting on. A moderate overall score on a long document often means a couple of sections read high and the rest read low, which is a completely different situation from the whole document reading moderately. And a low overall score can still contain a section that reads high, which is the one you would have wanted to know about.
So where the result offers per-section or per-sentence detail, that detail is the product and the headline percentage is the packaging. On our tool, sentence-level detail inside an uploaded document is Pro and above; the free and Starter tiers return the document-level reading. That is a real boundary rather than a soft one, and it is worth knowing before you plan a workflow around it.
Academic and technical prose is structurally regular: uniform sentence shapes, heavy hedging, stock transitions, a template. Those are the properties detectors read as machine-made, which is why human-written abstracts, literature reviews and methods sections routinely score high. If you are checking your own report and the methods chapter reads worst, the likeliest explanation is that methods chapters are written to a formula. The false-positive evidence is here, and it matters most for exactly this kind of document.
Two different jobs with different standards, and one of them has consequences for a person.
Reviewing a submission, as an editor or a client: a score is a reason to read more closely, and that is all it is. Ask for the working file, the draft history, or a conversation about the argument. Any of those tells you more than a percentage, and a percentage is not a basis for an accusation.
Assessing a student's work: the same, with a much higher duty of care. Detectors produce false positives on human writing, and the bias is not evenly distributed: writers working in English as a second language are flagged more often, which makes a score-based accusation unfair in a patterned way rather than a random one. Your institution's process exists for this and a tool output is not a substitute for it.
In both cases, if the person's answer is "I wrote it", the productive next step is evidence of how it was written: version history, dated drafts, notes, the restructured chapter. What that evidence looks like is here. And tell them not to rewrite the document, because a rewrite destroys the record that would have supported them.
One last thing on what leaves your machine: an uploaded document is processed on our servers. For confidential or third-party material, that is a decision to take against your own obligations before uploading rather than after. Our privacy policy governs retention.
The daily count gates text scans and rewrites; document uploads have their own, tighter count.
| Free | Starter | Pro | |
|---|---|---|---|
| Price | $0 | $9.99/mo · $7.49 yearly | $19.99/mo · $14.99 yearly |
| Actions a day (shared) | 3 | 20 | Unlimited |
| Words per scan | 1,500 | 5,000 | 7,500 |
| Words per humanization | 260 | 5,000 | 10,000 |
| Detection a month | No cap | No cap | No cap |
| Humanizer a month | No cap | 40,000 words | 80,000 words |
One action is one scan, one file upload or one rewrite, from a single shared daily count. That daily number is the gate that matters: there is no monthly ceiling on any plan, so what separates the tiers is how often you can work and how long a single piece can be, not how much you get through in a month.
Read the free row together rather than line by line. No monthly cap sounds unlimited and is not: three actions a day at up to 260 words a rewrite is roughly 780 rewritten words in a day. It is a real allowance for checking and fixing specific passages, and it is not a plan for processing a backlog.
Current prices, the Business and Enterprise tiers and regional pricing are on the plans page, which is the page we keep up to date.
Document uploads have their own count, and it is the tightest one here. One upload a day with no account, 5 on a free account, 25 on Starter and 100 on Pro. The file size cap is 10MB on every tier, including Business, so paying more buys more uploads and not bigger files. Sentence-level detail inside the document is Pro and above. None of this spends from the shared daily actions in the row above.
Upload the file rather than pasting its text. The extraction keeps paragraph boundaries, which the scoring needs, and pasting usually loses them. You get one upload a day with no account, up to 10MB, for PDF or .docx. Then read the result by section rather than as one number, because a long document contains several kinds of writing and an average describes none of them.
Most often because it is a scan. A PDF produced by a scanner or a photograph contains images rather than text, so there is nothing to extract and nothing to score. Run it through optical character recognition first and check the output is readable. Uploads are also verified by content rather than file extension, so an image renamed to .pdf is rejected, as is a spreadsheet or slide deck renamed to .docx.
Not very. A forty-page report contains a literature review, a methods section written to a template and a discussion written at speed, and those score differently. A moderate overall number usually means a couple of sections read high and the rest low, which is a different situation from the whole document reading moderately. Where per-section or per-sentence detail is available, that detail is the useful part. On our tool sentence-level detail inside an upload is Pro and above.
Because formal academic prose is structurally regular: uniform sentence shapes, heavy hedging, stock transitions and a template to follow. Those are exactly the properties detectors read as machine-generated, which is why human-written abstracts, literature reviews and methods sections routinely score high. If your methods chapter reads worst, the likeliest reason is that methods chapters are written to a formula.
No, and this is the part to be careful about. Detectors produce false positives on human writing and the bias is patterned rather than random: writers using English as a second language are flagged more often, which makes score-based accusations unfair in a systematic way. Treat a result as a reason to look more closely and ask for evidence of how the work was written, such as version history and dated drafts. Your institution process exists for this.
One upload a day with no account, five on a free account, twenty-five on Starter and a hundred on Pro. The 10MB file cap is the same on every tier including Business, so a higher plan buys more uploads rather than bigger files. Document uploads are counted separately from the shared daily actions that gate text scans and rewrites.
PDF or .docx, one upload a day with no account.
Open the tool →Why formal academic prose reads high, and who it affects.
Read the evidence →The evidence that outweighs any score.
See the steps →What happens when a document is longer than one pass.
Read the guide →One upload a day with no account, 10MB, PDF or Word. Then read it section by section.