Does Cutting Words Change Your Turnitin AI Score?

The number on your AI report is not a percentage of your entire paper. It is a percentage of what Turnitin calls "qualifying text," a subset of your submission, not the whole thing. Whether cutting words moves that number depends on which words you cut and whether they belong to that subset.

The displayed percentage indicates the amount of qualifying text within the submission that Turnitin's AI writing detection model determines was generated by AI.

That sentence is from Turnitin's AI writing detection capabilities FAQ page. The phrase to hold onto is "qualifying text." The percentage is a share of that pool, not of everything you uploaded. So before you delete a single paragraph to get under a word limit, you need to know what counts as qualifying and what does not.

Turnitin defines "qualifying text" on that same FAQ page, and the definition is narrower than most students assume when they first see a report.

This qualifying text includes only prose sentences, meaning that we only analyze blocks of text that are written in standard grammatical sentences and do not include other types of writing such as lists, bullet points (short non-sentence structures), or other non-sentence structures.

So the denominator, the base the percentage is calculated from, is prose sentences. Everything that isn't a prose sentence sits outside that base. For a fuller breakdown of what falls inside and outside that category, see what qualifying text means in Turnitin. Based on the definition above, here is the short version:

• Prose sentences (standard grammatical sentences): counted as qualifying text
• Lists and bullet points: excluded, described as "short non-sentence structures"
• Other non-sentence structures: excluded by the same wording

This is where it becomes practical. Because the denominator is prose only, cutting different parts of your paper has different effects on what the percentage is measured against.

• Cut prose sentences: the denominator changes
• Cut lists or bullet points: the denominator stays the same. The definition above names them as excluded
• Cut a table: it depends on what is in it. Turnitin's August 2023 release notes say "We are now able to process long-form prose text in tables." Short entries in a table are not prose sentences; a paragraph sitting inside a table cell is

Here is where the documentation stops helping. Turnitin does not say whether removing prose will push the percentage up, down, or leave it where it is. The outcome depends on which prose you remove: flagged sentences or unflagged ones. You cannot know that in advance unless you already have a report. Even with a report in hand, the highlights show what the model flagged in the current version, not what it will flag after you edit. What the documentation does let you say is what moves the denominator and what doesn't. It does not let you say which direction the number goes when the denominator moves.

There is a second issue that appears when you cut a lot. Turnitin's FAQ page acknowledges that short documents behave differently.

In shorter documents where there are only a few hundred words, the prediction will be mostly 'all or nothing' because we're predicting on a single segment without the opportunity to overlap.

Once a document shrinks to a few hundred words, the model makes what is essentially one call on the whole text. The percentage stops being a fine-grained reading and becomes closer to a binary judgment. That is our inference from the quote, not Turnitin's wording. The quote itself says "all or nothing" and explains the reason: a single segment with no overlap.

Results show that our accuracy increases with a little more text so submissions that contain less than 300 words may result in an AI writing score that is likely less accurate.

That statement is not on the FAQ page. It appears in a December 2023 release note on Turnitin's AI writing detection model page. Read it carefully: it says "less accurate," not "higher" or "lower." Accuracy is about whether the score reflects reality, not about which direction the number moves. A less accurate score could overshoot or undershoot. The documentation does not say which. What this means for you is that hitting your word limit and staying in the range where the reading is reliable can be two different, sometimes conflicting goals. If you want to understand why short papers often show extreme readings, see why Turnitin flags short papers as 100% AI.

Given everything above, here is what you can actually do when you need to shorten a paper and the AI percentage is on your mind.

• If you already have a report, check which prose sections are highlighted before you cut. Flagged and unflagged prose affect the percentage differently, and you want to know which is which before you start deleting.
• Trim lists and bullet points first. Those are named as excluded, so cutting them reduces your word count without touching the denominator. Tables are not a safe assumption: check whether the cells hold short entries or full paragraphs before you treat them as free.
• Watch the length. Turnitin's FAQ says that in documents of "only a few hundred words" the prediction becomes mostly "all or nothing", and its release note says accuracy increases with a little more text.
• If the submission is under 300 words, treat the percentage with skepticism. Turnitin itself says the score may be less accurate at that length. That is about the reading's trustworthiness, not about a direction the number will move.

None of this tells you the percentage will go down after you cut. The denominator may shift, the accuracy may drop, but the direction is not something the documentation predicts. If you want to work on the prose itself rather than just trimming around it, lowering your AI score without breaking citations walks through that separately. HumanPen's condense tool trims wordiness toward a target word count you set, for the whole .docx or for the paragraphs you select, and returns an editable file. It does not promise a specific AI score, because for the reasons this article just laid out, nobody can.