Contraction rate: the small habit that separates registers
Contraction rate is the proportion of contractible phrases a writer actually contracts: "don't" rather than "do not", "it's" rather than "it is", "we've" rather than "we have". It is one of the smallest measurable habits in English and one of the most stable. People settle into a rate for each situation and keep it. Humanize 360 measures contraction rate as one of the eleven signals behind the Human Score and as one of the three numbers a custom style profile records from your writing.
Read these two. "We have not decided yet, but it is likely that we will not renew the contract." Zero contractions out of four chances. "We haven't decided yet, but we probably won't renew." Two out of two. The first is fine in a board minute and odd in a text message; the second is the reverse. A language model in its default register lands almost always on the first, whatever the setting, and that consistency is what a detector notices.
Why it is a good signal
Most stylistic habits are hard to count. Contractions are not: there is a finite list of them and a clear set of phrases that could have been contracted. So the rate can be measured on a short passage without a reference model or a training set. It is also honest about register. Formal writing legitimately has a rate near zero. The tell is not a low rate; it is a rate that does not match the rest of the text, such as casual vocabulary and exclamation-style warmth with every "do not" spelled out.
How models handle it
- Default output rarely contracts, because published formal prose dominates training text.
- Asked to be casual, a model sprinkles contractions unevenly, often in the first paragraph and then forgets.
- Contractions in reported speech and quotations are copied faithfully, so dialogue can look human while the surrounding prose does not.
- Some models contract "it's" and "don't" readily but avoid "would've" or "there'd", giving a narrow, repetitive set.
How Humanize 360 sets it
Each mode has a target. Academic and Formal hold the rate at zero, which keeps them acceptable for coursework and official documents. Casual raises it and spreads contractions across the whole piece rather than the first few lines. Standard and Flowing keep the input's rate unless a style profile says otherwise. With a profile, the engine measures your own rate from 300 words and aims for it, so a writer who never contracts is not forced to and a writer who always does gets their voice back. The Changes view shows each contraction added or removed.
Reading the signal in the breakdown
A low contraction reading alongside a formal register and varied sentence length is not a problem. A low reading beside casual vocabulary, direct address and stock warmth is. Look at the signal together with tell-word ratio and transition-word overuse; when all three lean the same way, the text is in the mid-formal register that models default to, and choosing a real register in either direction will help.
Common questions
Should I add contractions to sound more human?
Only if the register calls for them. In an academic essay, no. In an email, a blog post or a product page, usually yes, and consistently rather than in the first paragraph only.
Does Academic mode ever add contractions?
No. Academic and Formal hold the rate at zero. If you want a formal piece with a light conversational touch, use Standard with a style profile built from your own writing.
Is a zero contraction rate an AI tell?
Not by itself. Plenty of human formal writing has none. It becomes a tell when it sits beside casual vocabulary or the stock hedges models favour. The breakdown shows the neighbouring signals so you can judge.
How does the style profile measure my rate?
It counts contractible phrases in your 300-word sample and the share you contracted, then stores that number as a target alongside sentence length, variance and vocabulary spread.