Japanese Speakers Writing English Essays: The Politeness Patterns Detectors Misjudge

A student in Osaka writes an English essay the way she was taught: courteous, measured, never overstating. She softens her claims, leans on the passive voice, opens with a careful framing instead of a blunt thesis. She wrote every word. The AI detector flags it anyway.

This happens to Japanese speakers writing in English more than it happens to native writers, and it isn’t bad luck. The same politeness habits that make Japanese-influenced English feel respectful and well-mannered are the habits a detector reads as “machine.” Here’s why, and what to do about it.

Key takeaways

• A 2023 Stanford study (Liang et al., published in *Patterns*) found that seven popular AI detectors wrongly flagged a majority of non-native TOEFL essays as AI-generated, with an average false-positive rate of 61.3%, while staying near-perfect on native US student essays.
• The trigger is statistical, not a judgment about quality: detectors reward surprising, uneven wording, and polite Japanese-influenced English tends to be smooth and predictable.
• Keigo-shaped habits, heavy hedging, indirect openings, the passive voice, and set formulaic phrases, look “AI-like” because they keep word choice safe and the rhythm even.
• A flag is a probability, not proof. The same essay can score very differently across tools and across days.
• You lower false positives by adding concrete detail and varying your rhythm, not by writing ruder or sloppier English.

What the research actually found

In 2023, a Stanford team led by Weixin Liang published a study in the journal *Patterns* with a flat, unmissable title: “GPT detectors are biased against non-native English writers.” They ran real essays through seven popular detectors. On essays by native-English US eighth graders, the tools were nearly flawless. On TOEFL essays by non-native speakers, those same detectors averaged a 61.3% false-positive rate. More than half of the honest, human-written essays got tagged as machine output.

The researchers traced the cause to one technical idea: perplexity. Low perplexity means the next word is easy to predict. Non-native writers tend to stay inside standard, “safe” phrasing, which keeps perplexity low, and detectors read low perplexity as a fingerprint of a language model. The tools weren’t reacting to who wrote the text. They were reacting to how predictable the words were.

That blind spot hits Japanese writers in a specific, recognizable way.

Why politeness in English reads as “AI”

Japanese has an entire grammar of deference, *keigo*, built around softening, indirectness, and avoiding flat assertions. When that instinct carries into English, it produces prose that is courteous and cautious. Most of it is grammatically perfect. It just happens to overlap with what AI text looks like statistically.

Hedging on every claim

Japanese rhetoric treats a bare assertion as a little rude. So the careful English writer cushions everything: “it may be said that,” “it is generally considered,” “I think that perhaps,” “to some extent.” These are high-frequency, low-surprise phrases. A model reaching for the most probable next word produces the same hedges constantly. Stack three of them in a paragraph and a detector sees a textbook low-perplexity signature, even though a real person wrote it out of plain good manners.

Indirect, front-loaded openings

Instead of leading with the point, Japanese-influenced English often warms up to it: “In recent years, there has been a growing discussion about…” or “It is widely believed that this issue is important.” That delayed, generalized opening is exactly the cadence detectors associate with generated prose, partly because AI loves to open the same way.

The passive voice as politeness

Japanese frequently drops the agent and lets the action carry the sentence. Translated over, you get “mistakes were made,” “the policy was decided,” “improvements can be expected.” Grammatically fine. But agentless, evenly weighted sentences are smooth and predictable, the opposite of the choppy, who-did-what-to-whom rhythm detectors read as “human.”

Set formulaic phrases

English education in Japan rewards reliable templates, “In conclusion, it can be said that,” “There are several reasons for this.” They’re safe and easy to grade. They’re also the exact connective tissue language models default to. A real student using a phrase she practiced fifty times looks, to the tool, identical to a machine using its most probable transition.

If you want the underlying math, we break down perplexity and burstiness, the two statistics behind most detectors in plain language. The short version: detectors don’t grade your ideas. They grade how surprising and how uneven your text is.

A mini-scenario: two versions, same meaning

Say Haruki wants to argue that his university should keep its library open later. His careful draft:

> “It is generally considered that library hours are important for students. It may be said that longer hours can be beneficial. Therefore, it is hoped that the opening time will be extended.”

Every word is his. It’s also a near-perfect low-perplexity storm: three hedges, no agent, no specifics, and three sentences of almost identical length and weight. A detector will likely flag it.

Now the same point, written concretely:

> “Our library closes at 9 p.m., which is brutal during exam week. Last January I got kicked out at closing with two chapters left and ended up reading on the cold floor of the 24-hour convenience store. Push closing to midnight in December and January, and that stops happening.”

Same student, same honesty, still perfectly polite in tone. But now there’s a specific time, a season, a real moment, a number, and the sentence lengths jump around. The writing got *better*, and the perplexity and burstiness went up as a side effect, which pulls the AI score down. You’re not tricking the tool. You’re writing like a specific person instead of a template.

What to do if you’re getting flagged

First, a flag is not a verdict. Detectors output a probability, not proof. OpenAI quietly retired its own AI-text classifier on July 20, 2023, citing a “low rate of accuracy.” Run your essay through three tools and you’ll often get three different scores. Run it twice through the same tool after small edits and the number moves. Anyone treating that number as evidence of cheating is misusing it.

Second, protect yourself with process. If you’re a student, keep your drafts, your outline, your notes, the version history in Google Docs or Word. A false positive is far easier to push back on when you can show the essay grew over time instead of arriving fully formed.

Third, improve the writing rather than sabotage it. Don’t add typos or weird words to fool the detector, that backfires with human readers and can look manipulated. Do this instead:

• Cut one hedge per paragraph. You rarely need “it may be said that” *and* “perhaps” *and* “to some extent” in the same sentence.
• Turn one passive sentence active per paragraph. Name who did what.
• Replace a vague claim with a concrete fact, date, place, or number.
• Vary sentence length on purpose. Put a five-word sentence next to a thirty-word one.
• Read it out loud. Where it sounds recited, break the pattern.

None of that requires you to be blunt or impolite. You can keep a respectful tone and still sound like a living person with specific experiences.

Where a humanizer fits, and where it doesn’t

If the words and reasoning are genuinely yours and a detector keeps mislabeling them, smoothing out the over-hedged, over-passive patterns is editing, not cheating. That’s the honest use of a humanizer: take human writing that scores “AI” because of second-language politeness habits, and nudge its rhythm and word choice toward how a specific person actually talks.

The line you shouldn’t cross is generating an essay and passing the machine’s thinking off as your own. A tool can fix how predictable your prose is. It can’t, and shouldn’t, supply your ideas. For Japanese writers especially, the goal is to sound like *you*, not to chase a number.

If you want to dig in further first, browse the rest of our writing on detection, or see the plans when you’re ready.

Frequently asked questions

Why does my English essay get flagged as AI when I wrote every word? Most detectors don’t read for meaning. They estimate how predictable your wording is (perplexity) and how much your sentence lengths vary (burstiness). Polite, hedged, evenly paced English fits the same pattern AI produces. The 2023 study by Liang and colleagues documented this directly, with seven detectors misclassifying most non-native TOEFL essays.

Is the hedging from Japanese politeness really what’s triggering the flag? It’s a big part of it. “It may be said that” and “it is generally considered” are high-frequency, low-surprise phrases, and AI hedges in very similar ways. Detectors read predictable phrasing as machine-like, so the caution Japanese English education rewards is exactly what the tool punishes.

If a detector flags my essay, does that prove I used AI? No. A detector gives a probability, not evidence. OpenAI retired its own classifier on July 20, 2023, for low accuracy, and scores swing between tools and between runs. A flag is a reason to look closer, never a verdict. Keep your drafts and version history if you ever need to defend your work.

Should I write less politely to avoid being flagged? Not exactly. The fix is to be specific and varied, not blunt. Cut a redundant softener, add a concrete detail, and let sentence lengths change. That raises perplexity naturally, as a side effect of clearer writing. You can stay courteous and still sound like a specific human.

Does running my own honest writing through a humanizer count as cheating? If the words and ideas are yours, smoothing the formulaic patterns that trip detectors is editing. The line is passing off machine-generated thinking as your own. For human writing that’s being false-flagged, the goal is to sound like you.

—

If a detector has mislabeled your honest, hand-written English, you’re not imagining it, and you’re not alone. The fix isn’t to write worse or ruder. It’s to write more specifically, in your own rhythm, while knowing what the tool is really measuring. When you’re ready to see exactly which lines are tripping the score, run it through PaperBleach.