Every draft produced inside PaperDraft is shown alongside two proprietary scores: a Lexical Naturalness Score and a Writing Rhythm Score. They are the most-asked-about feature on the product and also the most frequently misunderstood, so this page exists to explain exactly what they are, what they are not, and how to use them. The short version: these are writing-quality signals, shown to guide your revision. They are not AI detectors, they are not a compliance check, and they are not a promise about how any third-party tool will treat your paper. They exist to make visible the places where a first draft most often needs your attention before it becomes a submission.
What these scores are (and what they measure)
The Lexical Naturalness Score and the Writing Rhythm Score measure two distinct properties of the draft text. Both are computed locally against reference distributions drawn from human-written academic prose, and both produce a normalized score plus section-level highlights so you can see where the signal is strong and where it dips.
Lexical Naturalness Score
This score measures how natural the draft's word choices and collocations are relative to human-written academic writing. In practice, it flags:
- Over-reliance on a small set of transitional words or connectors.
- Phrases that tend to appear more often in model-generated text than in published academic prose.
- Collocations (word pairs and triples) that drift toward generic register instead of the conventions of your field.
- Sections where hedging or evaluative language is denser than typical human writing in the same genre.
A higher score means the vocabulary profile reads closer to human academic writing. A lower score means the draft is relying on patterns that tend to feel generic to a reader — not wrong, but undistinguished, and a place where your own word choices will likely improve the paper.
Writing Rhythm Score
This score measures the cadence of the prose — how sentences and paragraphs vary in length, structure, and complexity. It flags:
- Long runs of sentences at similar length (a common tell of unrevised model output).
- Paragraphs with uniform internal structure.
- Missing beats — places where a reader expects a short punchy sentence to break a sequence of long ones, or vice versa.
- Sections where connective tissue between ideas has been smoothed away into a single monotonous flow.
A higher score means the rhythm reads more like experienced human academic writing. A lower score means the prose carries a flatness that revision will dramatically improve.
Both scores are computed per draft and broken out per section, so you can see — at a glance — where the text is strong and where it needs you.
What these scores are NOT
These scores are writing-quality signals. They are not, and will never be marketed as, any of the following:
- They are not AI detection. They do not estimate the probability that a passage was produced by a language model, and they do not output a verdict about authorship. Their purpose is to guide revision of your own draft, not to classify it.
- They are not a bypass score. Nothing about the product is designed to help you defeat third-party detection tools, and improving these scores is not a promise about how any other tool will treat your paper.
- They are not a plagiarism check. They do not search any database for matching text. Plagiarism and originality are separate concerns requiring separate tools and, more importantly, honest citation practices.
- They are not a grade predictor. A high score does not guarantee a strong paper, and a low score does not guarantee a weak one. They measure surface features of the writing, not the quality of the argument or the accuracy of the research.
- They are not a finished-paper certification. No score replaces your revision, your own research, or your instructor's evaluation.
If that feels like a lot of "not," that is intentional. The category of AI writing tools has been muddied by products marketing themselves around defeating detection. PaperDraft is explicit that these scores live in a different universe. They help you revise your own draft. That is all, and it is enough.
How to read a low Lexical Naturalness Score
A dip in Lexical Naturalness is almost always a signal that a specific passage has drifted toward generic academic-ish vocabulary rather than the language a human writer in your field would use. A few common patterns:
- Over-used connectors. Stretches of prose where "moreover," "furthermore," "additionally," and "in addition" appear in close succession signal that the draft is gluing ideas together with default connectives rather than reasoning out the actual relationship between them.
- Hedged-to-death claims. Long chains of hedging phrases ("it could be argued that," "one might suggest that," "it is possible to consider that") often indicate a draft that is trying to sound academic without committing to an actual position. Your instructor wants the position.
- Generic abstract nouns. Reliance on high-frequency abstractions ("aspects," "issues," "factors," "considerations") without specific anchoring tends to flatten the score. Replacing them with concrete referents raises it.
- Absence of field-specific language. If a section on a specialized topic does not contain any of the vocabulary a reader in that field would expect, the score notices. Adding the domain's actual terms usually resolves the dip.
The revision move is almost always the same: rewrite the flagged passages in more specific, more committed, more field-aware language. That is also what makes for a better paper.
How to raise a low Writing Rhythm Score
Rhythm is about variation. A flat rhythm score means sentences are marching in formation — usually at a similar length, with similar structure, and without the internal beats that keep a reader engaged. Revision here is mechanical at first and then intuitive.
Tactics that reliably raise the score:
- Break a long sentence with a short one. If three sentences in a row run thirty words, insert a five-word sentence between them. The effect on rhythm (and on readability) is immediate.
- Vary sentence openings. Paragraphs where every sentence starts with the subject tend to read as uniform. Opening some with a dependent clause, a connective, or a concrete example restores movement.
- Let some paragraphs breathe. A two-sentence paragraph placed between two longer ones can do more for rhythm than another tightly packed block.
- Read aloud. This is the oldest advice in editing and still the best. Sentences that feel clipped, droning, or rushed in your mouth will feel the same on the page. Rewriting what your voice trips on almost always improves rhythm.
Rhythm is not about decoration. It is how a reader experiences the paper moving forward. A paper with a strong Writing Rhythm score feels easier to read and more authoritative, because both of those qualities are partly carried by cadence.
How to use the scores during revision
The scores are most useful as a map. They will not tell you what to say, but they will show you where your current draft is working and where it is asking for your attention.
A practical revision pass that uses the scores well:
- Read the draft end-to-end once before looking at any score. Get your own sense of where it is strong and where it is weak. Your sense matters more than any number.
- Open the scores and compare. Note the sections where your instinct and the scores agree — those are the clearest candidates for rewriting first.
- Revise in your own voice. Rework flagged sections by rephrasing in the language you would actually use if you were explaining the idea to a classmate. Specificity is almost always the lever.
- Re-read aloud after each revision pass. You will catch the rhythm dips that the score also catches, and you will catch meaning problems the score cannot see.
- Do not chase perfect numbers. Above a certain threshold, these scores stop correlating with better writing. Pushing a Lexical Naturalness score from the mid-eighties to the high-nineties rarely produces a meaningfully better paper. Spend that time on argument, evidence, and citations instead.
Used this way, the scores function as a second reader — one that sees the surface patterns of your draft without bias, and that lets you focus your revision attention on the places most likely to reward it.
Why we show these scores at all
It would be easier, commercially, to hide them. Tools that generate drafts rarely volunteer any measure of how much work still remains. PaperDraft shows these scores because the first draft is not the final version, and pretending otherwise does the user no favors. A student who submits a draft straight from any AI-assisted tool — ours included — will almost always submit something weaker than what they could have produced with a revision pass guided by good signal.
The scores are the product's way of telling you the truth: here is where your draft is already working, and here is where it still needs you. They are not a verdict, and they are not a shortcut around anyone else's tools. They are an honest picture of the draft in your hands, so you can finish the paper the way the paper deserves to be finished — by you.
Frequently asked questions
Is this the same as an AI detector?
No. AI detectors attempt to classify whether text was produced by a language model. Lexical Naturalness and Writing Rhythm scores measure writing-quality features — vocabulary choices and sentence cadence — against reference distributions of human academic writing, and they surface those measurements so you can revise. They make no claim about authorship, they do not output a verdict, and they are not designed to interact with any third-party detection tool.
What does a low Lexical Naturalness score mean?
It means the draft is relying on patterns — repeated connectors, generic abstractions, heavy hedging, field-vague phrasing — that tend to read as undistinguished in academic writing. The section flagged is a good candidate for rewriting in more specific, more committed, more field-aware language. The score does not say the passage is wrong; it says the passage would probably benefit from your voice.
How do I raise the Writing Rhythm score?
By introducing variation. Mix sentence lengths, vary sentence openings, let paragraph sizes breathe, and read the section aloud to catch runs of uniform cadence. Rhythm responds quickly to revision, so even a single focused pass on a flagged section usually moves the score meaningfully.
Do reviewers see these scores?
No. The scores are shown to you inside the product to guide your revision. They are not attached to your exported document, they are not sent to your instructor or institution, and they are not designed for any external evaluation workflow. They exist between the draft and you.