*% Versus 21%: Do Not Over-Read the First Visible AI Number

The AI writing indicator on a Turnitin-linked workflow does not always show a tidy percentage. Sometimes it shows an asterisk (*%). Sometimes it shows a number in the low twenties. Students often treat that jump as the moment the paper “became a problem.” The documentation points somewhere quieter: a display band, chosen in part to reduce over-reading of low estimates — not a courtroom gavel.

This article focuses on that display rule, how it sits next to a low similarity score (around 2%), and how to keep both numbers in proportion.

Short answer

Turnitin’s documentation describes Similarity and AI writing detection as completely independent. Low similarity and a visible AI percentage can both be true. Estimates in the band above 0% and below 20% are often shown as *% rather than a specific digit with the same highlight treatment. When the estimate reaches the display threshold, a number such as ~21% can appear. That crossing is a visibility threshold. It is not proof the work suddenly doubled in culpability. FAQ-style guidance also says the AI indicator should not be the sole basis for action or a definitive grading measure. Edit flagged qualifying prose if you still can. No workflow guarantees the next score.

What *% is doing on the screen

Read the asterisk as a deliberate UI choice for a low band, not as a mysterious error code. According to Turnitin’s documentation, scores above 0% and below 20% are not shown as a specific percentage in the same way higher estimates are; the indicator can appear as *%. Public materials describe this as helping reduce misinterpretation of false positives in that low range.

Practical translation:

• *% does not mean “zero concern forever.” It means the estimate sits in a band the product chooses not to spell out as a digit with the usual highlight package.

• Moving from *% to ~21% means the estimate crossed the display threshold. You can often see a number and passage highlights.

• It does not mean the paper became twice as “AI” overnight. A slightly lower estimate might still have stayed behind the asterisk.

When your screenshot shows ~21%, you are often looking at the first visible number in that design — a prompt to open the full AI Writing Report, not a finished judgment.

Put the low similarity score back in its lane

Similarity near 2% asks whether wording overlaps material in Turnitin’s comparison databases. It does not measure authorship. It does not cancel an AI classification. Side-by-side display is layout, not shared math.

Useful reading: prose may be largely unmatched in indexed sources and some qualifying prose may still be assigned to AI-like categories. Formal human academic English can sit in that neighbourhood too. A high similarity score would not prove AI use either — the axes stay separate both ways.

Useless reading: averaging 2% and 21%; treating low similarity as proof the AI number is broken; treating the first visible AI digit as proof of misconduct.

Enough mechanism to avoid over-reading

Qualifying prose is segmented (with overlaps), scored, and aggregated into one document percentage. Read three consequences, then stop:

• Denominator discipline. The percentage applies to qualifying text — not automatically every character in the file. Lists and many table cells are not treated like long-form prose.

• Highlights beat headlines. The AI Writing Report PDF shows where classification concentrated. A digit without passages is a weak object to argue about.

• Short documents swing harder. With less room for overlapping segments, predictions can look closer to all-or-nothing. Still inspect highlights when they exist.

How to respond without mythologizing 21%

Ask for the full AI Writing Report (and Similarity if attribution matters). Separate jobs: overlap versus flagged prose. Rewrite the largest highlighted blocks first, from your notes and sources, in language you can explain aloud. Freeze quotations, numbers, citation fields, tables, equations, and required templates. Re-check meaning after each block. Keep dated drafts and both PDFs. If you speak with an instructor, name the independence of the two systems and the asterisk display band — and remember the indicator is not meant to be the sole basis for action.

Do not plant typos or degrade prose to chase a quieter meter. Cross-tool scores are not interchangeable. No rewrite path can promise the next Turnitin percentage.

If revision is already scoped to the highlights

When the AI Writing Report is in hand and the job is to revise only the flagged prose while leaving citations, tables, and layout alone, a document-level pass that imports the report can keep the scope honest: change what was marked, freeze the rest, and re-check facts yourself afterward. That is the narrow use case HumanPen is built for — not a forecast of any future score.

Keep the display rule next to the digit: *% and ~21% are chapters in one visibility design. Read them that way, then edit the passages — not the mythology of the first number you can see.