The Repeated-Structure Tell: Why ‘Not Only… But Also’ Everywhere Looks AI

The clearest sign of machine-written prose usually isn’t a single bad sentence — it’s the same sentence shape used over and over. “Not only does this save time, but it also improves accuracy.” “It’s not just about speed; it’s about trust.” “From onboarding to renewal, the platform handles it all.” Any one of these is fine. Three of them in a row is one of the most overused sentence structures in AI writing, and once you can hear it, you can’t unhear it.

Readers rarely name the problem. They just feel that something is off — the writing is competent, grammatical, even elegant, and somehow it reads like a recording of prose rather than prose. Most of the time, the culprit is structural repetition: a handful of templates the model likes, deployed on loop.

Key takeaways

• The tell is repetition of *structure*, not any single phrase. “Not only… but also” is fine once and grating three times.
• Models lean on symmetrical, balanced constructions because they’re statistically common and safe to predict.
• Heavy structural repetition lowers burstiness — the variation in sentence length and shape — which is part of what reads as AI.
• The fix is to break the pattern, not to ban the construction: keep one instance, rewrite the rest as plain sentences.
• You catch it fastest by searching for the connective words and reading the draft aloud.

Why the machine keeps reaching for the same shape

A language model writes by predicting the next word, over and over, choosing what’s most probable given everything so far. Balanced constructions are catnip for that process. “Not only X but also Y” is a closed frame: once the model commits to “not only,” the rest of the sentence is highly predictable, which is exactly the kind of low-resistance path the prediction machinery favors. Same with “it’s not just A, it’s B” and the travel-brochure “from X to Y” sweep.

Humans use these too. The difference is that we contaminate our writing with everything else going on in our heads — a joke we couldn’t resist, a sentence we cut in half because we got bored, a fragment. One sentence long. Then a rambling one that piles three clauses together because the thought wasn’t finished yet. That noise is what keeps a human paragraph from settling into a single groove. The model has no such noise, so left alone it produces a page where the sentences are individually good and collectively monotonous.

The three usual suspects

Most AI structural repetition comes down to a small cast:

• The correlative pair: “not only… but also,” “both… and,” “either… or” used as a rhythm rather than a real contrast.
• The false-dichotomy pivot: “It’s not about X. It’s about Y.” Punchy once. A tic by the third time.
• The range sweep: “From beginners to experts,” “from strategy to execution,” “from the boardroom to the shop floor.” It sounds comprehensive and says very little.

None of these is wrong. Each is a tool. The problem is a writer — human or machine — who owns one hammer.

Overused sentence structures in AI writing, and what to do instead

Here’s the move. You don’t delete the construction; you ration it. Keep the single best instance, where the balance actually earns its keep, and rewrite the others into flatter, more direct sentences. The goal is variation, not purity.

Take this paragraph, which reads like an untouched draft:

> Our tool not only speeds up your workflow but also reduces errors. It’s not just a time-saver; it’s a quality upgrade. From your first draft to your final export, it supports every step. Not only is it easy to learn, but it also scales with your team.

Four sentences, three templates, and a numbing sameness. Now break the pattern:

> Our tool speeds up your workflow and cuts errors along the way. That second part matters more than the first — a fast draft full of mistakes isn’t actually faster. It supports every step, from your first draft to your final export. And new people pick it up in an afternoon.

The rewrite keeps one “from X to Y” (it’s doing real work there) and lets the rest breathe. The sentences are now different lengths and different shapes. Nothing about it feels chanted.

It’s rhythm, not vocabulary

This is why swapping synonyms does nothing for the problem. You can replace “speeds up” with “accelerates” and “reduces” with “minimizes” and the paragraph will read exactly as robotic, because the *architecture* is untouched. Structure lives above word choice. A reader feels the beat of “not only… but also, not only… but also” no matter which words you pour into the frame.

That’s also why this is one of the higher-value edits you can make to an AI draft. Fixing structure forces you to actually rebuild sentences, and rebuilt sentences carry your decisions, your emphasis, your voice. If you want the paragraph-level version of this idea, the anti-symmetry edit covers how to break the too-balanced pattern across a whole passage, and the sentence-openers audit tackles the related habit of starting every line the same way.

The detection angle, kept honest

There’s a real connection to AI detectors here, and it’s worth stating carefully. Detectors don’t scan for “not only… but also.” They estimate two things: how predictable your word choices are (perplexity) and how much your sentences vary in length and rhythm (burstiness). Structural repetition hits the second one directly. When every sentence is built on the same frame, they tend toward the same length and the same cadence, and that evenness is exactly what low burstiness looks like.

So breaking up repeated structures does tend to nudge a score toward the human side — not because you fooled a pattern-matcher, but because you genuinely added variation. Keep the goal straight, though: detectors return probabilities, not verdicts, and they get it wrong often enough that gaming them is a poor target. OpenAI shut down its own AI-text classifier in July 2023 for weak accuracy, and independent research has documented these tools misfiring on ordinary writing. Fix the repetition because it makes the prose better to read. The detector score is a byproduct.

A two-minute audit for your own drafts

1. Search for the connectives, one at a time. “not only,” “it’s not just,” “it’s not about,” “from,” “whether.” Each search collects every instance in one place.
2. Count them. If a single structure appears more than twice in a short piece, you’ve found your loop.
3. Keep the best one. The instance where the balance is genuinely meaningful stays.
4. Flatten the rest. Rewrite each remaining one as a plain declarative sentence, even a short blunt one.
5. Read it aloud. Repetition is easy to hear and hard to see. If a passage starts to sound like a rhythmic chant, you’re not done.

This works just as well on your own writing, where the repeated structure is usually a favorite phrase you don’t notice, as it does on an AI draft, where it’s baked in. Either way, the cure is the same: notice the pattern, then refuse to let it run.

Closing

Good sentences are not enough. A page of individually strong sentences that all share one skeleton still reads as machine-made, because the tell was never in the words — it was in the repetition. Ration your favorite constructions, let some sentences go short and blunt and others sprawl, and the mechanical hum fades out.

If you want to see where a draft still leans on the same few shapes, run the draft through PaperBleach and watch which sentences the heatmap keeps flagging. You can also browse the rest of the PaperBleach blog for the other craft edits that pair with this one.

Frequently asked questions

Is ‘not only… but also’ grammatically wrong?

No, it’s perfectly correct, and used once in a while it’s a fine construction. The problem isn’t the grammar, it’s the frequency. When the same correlative pair shows up three times in a 600-word piece, a reader starts to feel the template underneath the sentences. The fix isn’t to ban it; it’s to notice when you’re leaning on it and let some sentences take a different shape.

Why does AI reach for the same sentence structures so often?

A language model predicts the most probable next word given everything before it, and balanced, symmetrical constructions are statistically common in the training data. “Not only X but also Y,” “It’s not just A, it’s B,” and “From X to Y” are safe, high-frequency patterns, so the model returns to them. Humans have the same habits, but we also have moods, deadlines, and pet phrases that break the pattern up.

Do AI detectors specifically look for repeated sentence structures?

Not by name. Detectors measure statistical texture, mainly perplexity (how predictable your word choices are) and burstiness (how much your sentence length and shape vary). Repeated templates lower burstiness because every sentence marches to the same rhythm, so heavy structural repetition does tend to push a score in the AI direction. But it’s a side effect of the measurement, not a blocklist.

How many times can I use the same structure before it’s a problem?

There’s no exact number, but a useful rule of thumb: if you can feel the pattern as a reader, it’s too much. Two “not only… but also” sentences in a long essay will pass unnoticed. Three in one paragraph will not. Read the draft aloud, and any structure that starts to sound like a chant is the one to break up.

What’s the fastest way to catch my own repeated structures?

Search the document for the specific connective words, one at a time: “not only,” “it’s not just,” “from,” “whether.” Each search jumps you to every instance, and seeing them stacked makes the overuse obvious. Then rewrite every second or third one into a plain declarative sentence. The rhythm loosens up immediately.