One line in my research notes said 5,000 words per process. The next said 10,000 characters in one go. A third recorded paid allowances in words per month.
I was checking whether "unlimited words" could be treated as a comparable feature across AI humanizers.
It cannot.
Phrasly's pricing page makes the mismatch easiest to see: it puts Unlimited on a plan that also says 5,000 words per process.
Both statements on that plan can be true. Unlimited describes monthly humanizations. The 5,000-word number describes one process. For 12,000 words, that published boundary turns one input into at least three processes, before anyone asks whether the text reads well afterward.
That is simple arithmetic, but it changes the buying question. A monthly allowance tells you how much work the account may cover over time. A per-process cap tells you how many pieces the document must become today. For a long paper, those are different costs.
Every split creates a boundary to manage. You need to decide where one piece ends, submit it, collect the result, put it back in order, and check the join. If headings, citations, lists, or transitions cross that boundary, the work is more involved than dividing a word count by 5,000.
The word "unlimited" does not identify which limit disappeared. It may refer to the monthly allowance, input size, number of requests, or usage frequency. A trial, free tier, and paid tier on the same site can each use different rules.
QuillBot's official AI Humanizer page shows how much the access tier matters. The Free block stops at 125 words. Premium carries the separate phrase No word or usage limit. Applying the Free number to Premium is wrong. Applying the Premium claim to Free is wrong too. The page needs both statements because they belong to different accounts.
Undetectable.ai shows a different mismatch. Its humanizer page sets a 10,000-character maximum for one submission. The pricing page switches units and describes monthly paid allowances in words. Characters per run and words per month answer different questions.
I would leave the character figure as characters. The ratio changes with language, word length, spacing, and punctuation. The useful check is to measure the actual document in the unit the interface uses. A 10,000-character field either accepts that exact passage or it does not.
This leaves four numbers to write down before paying for a long-document job.
First is the account allowance for the billing period. Second is the amount accepted in one request. Third is the unit, because words and characters cannot be swapped casually. Fourth is the access tier that makes each statement true.
There may also be a fifth limit that has no word count at all: the interface.
Before comparing plans, I would record both the document's word count and its character count. Then I would mark the section that needs rewriting. A 40,000-word thesis with 1,500 words in scope creates a different job from a service that has to process all 40,000.
A text box and a file workflow solve different problems. A paste field can have a generous input limit while still returning plain text that must be rebuilt inside a DOCX. A file service can accept a much larger source document yet charge or process only selected passages. File type, file size, page handling, output format, and selection controls matter when the deliverable is a document rather than a paragraph.
This is where roundups often become misleading without making any individually false statement. One row says "unlimited." Another says "10,000 characters." A third says "125 words." The table looks comparable because every value sits under a column called limits. In practice, one may describe a paid monthly allowance, one a per-run field, and one a free tier.
The smallest applicable boundary decides what happens to the paper.
Suppose a plan has unlimited monthly usage and a 5,000-word process cap. The monthly label means a 12,000-word paper does not exhaust a stated word allowance. The process cap still splits it into at least three submissions. If the interface only accepts pasted text, the document also has to leave and re-enter its file structure.
Now suppose another plan has a monthly word allowance and a 10,000-character input field. You have to satisfy both. Enough words remaining in the account does not make an oversized passage fit in the field.
None of these capacity facts predicts detector results. The allowance governs accepted or billable volume. Detector classification, preserved meaning, and document usability sit outside that number. Those questions need their own test and manual review.
My notes now use a sentence that is less attractive than an "unlimited" badge but much more useful: identify the exact plan, the noun being limited, the time period, the per-request boundary, and what comes back after processing.
For a long paper, add the real word and character counts beside those rules. Then count the submissions and the reconstruction work. That tells you what the plan can do with the document already on your screen.
I kept the dated limits and the full reading method in this longer guide.
