The Transition-Word Problem: ESL Overuse of Furthermore, Moreover, and Additionally

Open almost any essay written by a strong ESL student and you’ll spot a pattern fast. Furthermore at the top of one paragraph. Moreover at the next. Additionally a few lines down. Each word is correct. Each one was taught as a sign of good academic writing. And together they can make an AI detector light up like a slot machine.

That’s the cruel part. The habit that earned top marks in an English program in Seoul, São Paulo, or Riyadh is the same habit that gets a paper flagged as machine-written at a university in the US. Let’s unpack why that happens — and how to fix the rhythm without throwing away your voice.

Key takeaways

• ESL writers often overuse *furthermore*, *moreover*, and *additionally* because that’s exactly how academic English is taught in many countries. It’s a learned habit, not a flaw.
• AI detectors don’t read for meaning. They measure predictability, and a paragraph that opens with a formal connector every time looks statistically like machine output.
• This creates a double bind: the same writing your teacher praised abroad can get you flagged at a Western university.
• The fix isn’t deleting every connector. It’s varying *how* you link ideas so your rhythm stops being predictable.
• A connector should earn its place by signaling a real logical turn, not act as default glue between every sentence.

Where the habit comes from

This isn’t carelessness. In a huge number of English-language curricula outside the US and UK, transition words — the textbooks often call them “cohesive devices” or “linking words” — are taught as the backbone of academic style. Students memorize lists. Exam rubrics literally award points for using a range of connectives. IELTS and TOEFL preparation drills them hard.

So a student spends years being rewarded for writing “Furthermore, the data suggests…” and “Moreover, this finding indicates…” By the time they reach a graduate program abroad, the habit is automatic. They’re not padding. They’re doing exactly what every teacher told them was excellent.

The disconnect is that native academic writers, especially in informal-leaning American prose, use these words far more sparingly. They reach for shorter joins, or none at all. So the very feature that signaled mastery in one context starts to signal something else entirely in another.

Why detectors react to it

Here’s the mechanical reason this matters for AI detection. Detectors don’t understand your argument. They estimate how *predictable* your text is — how closely your word choices match what a language model would have produced in the same spot.

Large language models trained on mountains of formal text learned that *furthermore* and *moreover* are extremely common connectors. So when a model writes, it gravitates toward them too. Three paragraphs opening with *Additionally* is a very LLM thing to do.

When your human writing happens to lean on the same connectors with the same regularity, you land in the same statistical neighborhood as the machine. Low surprise, steady rhythm, predictable openers. That’s the territory detectors measure with perplexity and burstiness. A human who varies wildly between a four-word sentence and a thirty-word one, and links ideas a different way each time, reads as “bursty” and surprising. A page of evenly weighted sentences each glued with a formal connector reads as flat and predictable — which is what the detector treats as a red flag.

None of this means the connectors are wrong. It means your pattern got predictable, and predictability is the thing being scored.

The double bind for ESL writers

This is where it gets genuinely unfair. A 2023 study from Stanford researchers (Liang and colleagues, with James Zou as senior author) ran 91 TOEFL essays written by non-native English speakers through seven commercial detectors. On average, 61.3% of those human-written essays were misclassified as AI-generated. Essays by native English speakers were flagged only about 5.1% of the time. Part of the reason: non-native writers tend to use a narrower, more predictable band of vocabulary and phrasing, which reads as machine-like to a system trained to spot low surprise.

The transition-word habit feeds straight into that. You were taught to use a consistent set of formal connectors. The detector reads consistency as a signal. So you can write every word yourself, honestly, and still get a high AI score — for doing the thing your education trained you to do well.

It’s worth remembering that detectors output probabilities, not proof. A score is a guess about likelihood, not evidence of cheating. OpenAI even pulled its own AI-text classifier in July 2023, citing its “low rate of accuracy.” Keep that in mind if you’re ever accused: the tool flagging you is fallible, and the bias against non-native writers is documented.

A mini-scenario

Picture Mai, a Vietnamese master’s student. Her literature review reads:

> *Furthermore, the study by Chen (2021) supports this view. Moreover, similar results were reported by Garcia (2022). Additionally, a longitudinal analysis confirmed the trend. In conclusion, the evidence is consistent.*

Every sentence is true and grammatical. But each one is roughly the same length, each opens with a formal connector, and the rhythm never changes. A detector reads that as low-surprise, evenly paced — machine-shaped.

Now the same content, loosened up:

> *Chen (2021) backs this view, and Garcia (2022) found much the same thing a year later. A longitudinal analysis confirmed the trend too. The evidence lines up.*

Same facts. Same formality of argument. But the sentences vary in length, the connectors changed or vanished, and the last line lands short and plain. It reads like a person made choices. That variation is what both human readers and detectors register as natural.

How to fix the pattern without losing your voice

You don’t need to purge every connector. You need variety. A few practical moves:

1. Vary the *type* of link, not just the word

Swapping *furthermore* for *in addition* doesn’t help — it’s the same move in a new coat. Instead, rotate between strategies. Sometimes use a plain *and* or *but*. Sometimes drop the connector and let two sentences sit side by side. Sometimes carry the reader forward with a detail instead of a signpost.

2. Use the “does it survive deletion?” test

Read each connector and mentally delete it. If the paragraph still makes sense — and it usually does — you probably didn’t need the word. Keep the connectors that genuinely mark a turn in logic (a real contrast, a real consequence) and cut the ones doing nothing but filling the gap.

3. Break up the metronome

The biggest tell isn’t any single word. It’s the steady beat: similar-length sentences, each opening the same way. Throw in a short sentence. Then a longer one. Start one sentence with the subject, the next with a clause. The connectors will thin out naturally once the rhythm stops being uniform.

4. Read it aloud

If you’d never say “moreover” out loud to a friend explaining your point, that’s a hint it’s doing ceremonial work rather than real work. Spoken rhythm is naturally varied. Borrow some of it.

5. Check, don’t guess

After revising, run the draft through a detector and look at the sentence-level highlights. You’ll often find the flagged lines are exactly the ones starting with *Moreover* or *Additionally*. That tells you where to keep loosening. For more on the writing side of all this, there are more guides on writing that reads human.

Frequently asked questions

Why do ESL writers use furthermore and moreover so much?

In a lot of English-language programs outside the US and UK, these connectives are taught as the markers of good academic writing. Students get checklists of “cohesive devices” and earn points for using them. The habit is rewarded for years before anyone says it can backfire. It isn’t sloppiness — it’s training. The trouble is that the same habit reads as formulaic to a Western reader and statistically predictable to a detector.

Will using fewer transition words make my essay pass an AI detector?

Not by itself. Connectors are one signal among many. Detectors look at overall predictability — sentence length, word choice, rhythm — and they report probabilities, not proof. Varying your transitions helps, but it’s the full picture that moves a score. The most reliable move is to check a draft, see which sentences light up, and revise those specifically.

Should I just delete every furthermore and moreover from my writing?

No. They’re real words with real jobs, especially in formal or legal prose. The problem is frequency and predictability, not the words themselves. Keep them where they mark a genuine logical step, and cut the ones that are just glue. A good test: if the sentence still makes sense after you remove the connector, you probably didn’t need it.

Is it fair that ESL writers get flagged more often for this?

It isn’t, really. In the 2023 Stanford study, detectors flagged 61.3% of non-native English essays as AI on average, versus about 5.1% of native-speaker essays — partly because more predictable word choices read as machine-like. The connector habit is one piece of that. Knowing the bias exists is the first step to writing around it and defending your work if you’re wrongly accused.

What can I use instead of always starting with Additionally?

Mix it up. Sometimes a plain *And* or *But* does the job. Sometimes you can drop the connector entirely and let the ideas sit next to each other. Sometimes a concrete detail or a short sentence carries the reader forward better than any formal word. The goal is variety, not a banned-word list.

The bottom line

If your writing leans hard on *furthermore*, *moreover*, and *additionally*, you’re not a bad writer. You’re a well-trained one who learned a style that happens to overlap with how machines write. The fix is rhythm, not erasure: vary how you connect ideas, keep the connectors that earn their keep, and let some sentences run short and plain.

Want to see where your own habit is showing? Check your draft and see which sentences get flagged — the highlights usually point straight at the lines you can loosen first.