Reading Level and AI Detection: Why Simpler ESL Prose Sometimes Scores More AI, Not Less

You cleaned up your essay. You cut the long, tangled sentences, chose plainer words, and kept every point short and clear. It reads better than your first draft did. Then the AI detector lit it up, and now you’re staring at a high score wondering how making your writing simpler made it look more like a machine.

For a lot of ESL students the honest answer is uncomfortable: reading level and AI detection are linked, and the clean, simple, low-variation prose you write when you’re careful in a second language shares a statistical fingerprint with the text an AI model produces. You didn’t do anything wrong. You wrote in the safest, clearest way you know, and that very safety is what the detector reacts to.

This guide explains why that happens, and what to change so your writing stays clear and correct but stops setting off the alarm.

Key takeaways

• Simpler prose is not automatically more human. Detectors read predictability and rhythm, not clarity, and simple writing can be very predictable.
• Careful second-language writing tends toward common words and even, similar-length sentences, which is exactly the low-variation pattern detectors associate with AI.
• The fix is variation, not complexity. You can raise variation without adding a single error.
• Mix sentence lengths, add detail only you could write, and let transitions follow your argument instead of sitting on top of it.
• A detector gives a probability, not proof. These tools are documented to misfire on non-native English, so treat a high score as a prompt to revise, not a verdict.

How reading level and AI detection get tangled

Most detectors are built on two measurements. Understanding them is the whole game, and we go deeper in our piece on perplexity and burstiness, but here’s what matters for reading level.

Perplexity: how predictable your words are

Perplexity is a measure of surprise. Given the words you’ve already written, how expected is the next one? Language models are trained to pick the most likely next word, so their output is low-perplexity by design. It flows smoothly because it’s predictable.

Now think about what you do when you write carefully in a second language. You reach for the common word instead of the rare one. You choose the phrasing you’re sure is correct over the one that’s more interesting but riskier. That’s smart, and it protects you from mistakes. It also lowers your perplexity, because common, safe words are the predictable ones. The detector can’t tell the difference between “this person chose safe words to avoid errors” and “a model chose safe words because it always does.”

Burstiness: how much your rhythm varies

Burstiness is variation in sentence length and structure. Human writing tends to be bursty. A long, winding sentence gets followed by a short one. A person changes pace, pauses, hammers a point, then explains it slowly.

Simpler writing flattens that out. When you keep every sentence short and clean, they end up roughly the same length, built the same way, one after another. Low burstiness. And low burstiness is one of the loudest signals a detector has, because models also produce very even, uniform text. So the more you smooth your prose into tidy, equal pieces, the closer its rhythm gets to machine output.

Put the two together and you get the trap. Simple word choice pushes perplexity down. Uniform short sentences push burstiness down. Both moves feel like good, clear writing, and both moves point the detector toward “AI.”

Why careful second-language writing drifts this way

This isn’t a flaw in you. It’s a natural result of how people write in a language they’re still mastering, and it’s well documented.

When you’re not fully confident in English, caution is the sensible strategy. You stick to vocabulary you’ve tested. You build sentences on patterns you know are grammatical. You avoid the idioms and the odd turns of phrase that might come out wrong. The result is prose that is correct, clear, and remarkably even, because you’ve filtered out the risky, surprising parts that would have raised your variation.

Researchers have found that this shows up as detector bias. A 2023 study by Liang and colleagues found that several popular detectors flagged writing by non-native English speakers far more often than writing by native speakers, in part because non-native writing tended to use a narrower, more predictable band of vocabulary. When the researchers had the non-native samples rewritten with richer language, the false alarms dropped. The bias wasn’t about cheating. It was about the statistics of cautious word choice.

It’s worth remembering how rough these tools are. OpenAI launched its own AI text classifier and then quietly shut it down in 2023, citing a low rate of accuracy. The companies building detectors know they guess. A high score is a signal to look again, not a confession.

How to add variation without adding errors

Here’s the good news. Everything that pushes your writing toward “AI” can be fixed by raising variation, and variation does not require harder English. You can keep every clear word and safe structure you already trust, and still break the uniform pattern.

Mix sentence lengths on purpose

This is the single highest-leverage change. Most careful ESL drafts run a chain of similar medium sentences, all in the 15-to-20-word range. Go find your longest run of same-length sentences and break the pattern by hand.

Split one long sentence into two short ones. Take two short sentences and join them with a clause. Then add one very short sentence to land a point. Like this. You haven’t changed your vocabulary at all. You’ve only changed the rhythm, and that alone raises burstiness and reads as more confident.

Add detail only you could write

Predictable claims come from generic language. Specific claims don’t. When you name the actual study, the real number, the exact example from your own reading, you force in words a model wouldn’t have guessed, and you raise perplexity in the honest way.

• Generic: *Many studies show that this method is effective.*
• Specific: *The one trial I found tested it on 40 students over six weeks, which is too small to trust much.*

The specific version is clearer to a reader, harder for a detector to predict, and it shows an examiner you actually engaged with the source. You don’t need fancier English to write it. You need your real detail.

Let transitions follow your argument

Careful writers often bolt the same connectors onto every paragraph, because a phrase bank taught them that’s what academic English does. Rows of *moreover*, *furthermore*, and *in addition* create a predictable rhythm the detector keys on.

You don’t have to ban those words. Just vary how you link ideas, and sometimes carry the link inside the sentence instead of announcing it up front. “The policy also raised costs” is cleaner than “Furthermore, the policy raised costs,” and it’s less templated.

Check your own draft before you submit

You can catch most of this yourself in ten minutes.

1. Read the draft aloud. If it sounds like one flat tone all the way through, your rhythm is too even.
2. Mark sentence lengths in a rough way. If five sentences in a row are almost the same length, break at least two of them.
3. Circle repeated openers. Three paragraphs starting with the same connector is a pattern worth changing.
4. Find one generic claim per section and replace it with a specific detail from your source.

If you want a second opinion, you can check a draft and humanize it to see which sentences a detector still finds predictable, then fix only those instead of rewriting the whole thing. Keep the parts that already read as human.

Frequently Asked Questions

Why does my simple, clear English get flagged as AI? AI detectors mostly measure two things: how predictable your words are (perplexity) and how much your sentence length varies (burstiness). Simple, careful writing is often very predictable and very even, because you’re reaching for common words and safe, similar-length sentences. That low-variation pattern is close to what a language model produces, so the tool can rate it high-probability AI even when you wrote every word. It’s reading the statistics of your phrasing, not your honesty.

I lowered my reading level to be clearer. Why did my AI score go up? Lowering reading level usually means shorter words and shorter, more uniform sentences. That raises predictability and flattens rhythm, which are the exact signals detectors lean on. Clearer for a human reader and more AI-like to a detector can happen at the same time. The fix isn’t to write harder; it’s to keep the clear words but vary how your sentences are built.

Should I write more complex sentences to look human? No. Complexity for its own sake adds errors, especially in a second language, and errors don’t lower an AI score in any reliable way. What helps is variation, not difficulty. Mix a short sentence with a longer one, add a concrete detail only you would know, and let your transitions follow your argument. You can stay clear and correct while still sounding like a person.

Is a high AI score proof my writing is bad or that I cheated? No. A detector outputs a probability, not a verdict, and these tools are known to misfire on non-native English. A 2023 study by Liang and colleagues found several detectors flagged writing by non-native English speakers far more often than native writing, and OpenAI shut down its own classifier in 2023 for low accuracy. A high score is a guess about patterns, not evidence about you.

What’s the fastest fix if I write in a second language? Vary your sentence length. Most careful ESL drafts run a chain of similar medium sentences, so break one in two, let one run longer with a clause, and add a short blunt one to land the point. It raises burstiness, reads as more confident, and doesn’t require any risky new vocabulary. Do that one thing and re-check the draft.

The point isn’t a lower score

Simpler is not the same as more human, and that’s the whole misunderstanding. What reads as human is writing that varies, that carries a specific detail, that changes pace because a person with something to say wrote it. You can get there without reaching past your English, because none of it depends on harder words.

Write with variation because it’s better writing, and the detector problem mostly fades on its own. When you want to keep sharpening the habit, browse more guides on writing and detection, or see which plan fits if you’ll be checking drafts often.