unslop

Structured writing habits and AI detection

Detectors reward irregularity. Some people write in a naturally regular, systematic way, and that style scores higher for reasons that have nothing to do with how the text was produced.

3 min read

Wrong flags concentrate in writing with specific measurable properties, and one of those properties is consistency. People whose writing is naturally systematic, methodical or highly structured sit closer to the flagged end of the scale than people who write loosely.

This page is about writing habits and what the software measures. It makes no claim about anybody, only about how particular writing patterns are scored.

The habits that raise a score

Working from a plan. Writing to an outline produces even paragraph lengths and a consistent structure down the page. A classifier reads that regularity, and generated text has exactly the same property.

Consistent sentence construction. Some people find one clear construction and use it. It is precise and it is easy to read. It also flattens the variation that detectors treat as the strongest human signal.

Human academic writing in our corpus varies sentence length with a standard deviation of 8.7 words, generated text with 6.4.

Two overlapping histograms of within-document sentence length variation. Human writing centres on a standard deviation of 8.7 words, generated text on 6.4.
Two overlapping histograms of within-document sentence length variation. Human writing centres on a standard deviation of 8.7 words, generated text on 6.4.

Explicit signposting. Marking every logical step out loud is careful writing, and it means more sentences opening with "Moreover", "Furthermore", "Additionally". Those open 2.4% of human sentences and 5.8% of generated ones.

Bar chart comparing how often sentences open with a connective such as moreover or furthermore. Human writing 2.4 percent, generated writing 5.8 percent.
Bar chart comparing how often sentences open with a connective such as moreover or furthermore. Human writing 2.4 percent, generated writing 5.8 percent.

Precise, repeated terminology. Using the same exact term for the same exact thing rather than varying it for style. This is good technical practice and it lowers word-choice unpredictability, which is the other thing detectors measure.

Thorough revision. Every pass smooths, and smoothing removes variation. See why good writing gets flagged more.

Why this matters practically

None of these are errors. Several of them make the writing better and all of them are stable features of how somebody writes rather than choices made on one document.

The consequence is that the score is not a one-off event to be waited out. If your natural style sits in this territory, it sits there on every document, and you will see it again.

That is an argument for knowing your own number rather than reacting to each result.

Find your baseline

Run three or four pieces you wrote before any of this was a question. If they all come back in the same range, that range is your writing, and a new document inside it is not a signal about that document.

Half an hour, once, and every score afterwards is a comparison instead of a verdict. See find your own baseline score.

What moves it without changing how you work

Vary sentence length on purpose, at the end. Treat it as a mechanical pass over the finished draft rather than something to think about while writing. Find three sentences of similar length, split one, merge another pair.

Delete connective openings in the same pass. If the logical step is real, the sentence order carries it.

Keep the specific detail. This one usually needs no work, since precise writers already have it.

Full list in what makes text read as machine written.

Ours is free and unlimited with no account, scores every paragraph separately, has no word cap, and never stores your text. Wrong-flag rate 0.4%, which is 99.6% specificity, against Turnitin's 4%.

Check any text with our detector, free and unlimited →