How to humanize a ChatGPT essay before submitting.
AI detectors are getting better. Some schools quietly flag essays that score above 70% AI-likely — and the worst part is that humans get false-positived for using one wrong word. Here are the 11 specific patterns to fix, in order of how badly they betray you.
Two facts that change everything about this:
- AI detectors don’t actually read your essay. They look at sentence rhythm, word choice patterns, and over-balanced structure. A real human can trip them.
- ChatGPT writes in the same voice over and over. If you fix that voice, it doesn’t matter who wrote the first draft — the final draft passes.
So this isn’t about cheating. It’s about the fact that the same patterns that read as “machine-written” also read as generic — the kind of essay 10,000 other applicants are submitting. Fix one, you fix both.
The 11 patterns, ranked by detector weight
1. Robotic transitions (high signal)
Words like “Furthermore”, “Moreover”, “In conclusion”, “Additionally”. These read as machine-written because LLMs use them to glue paragraphs that don’t actually connect. Replace with: nothing (just delete), or “And”, or “But”, or “Here’s the thing”.
2. The LLM verb cluster
Strip every instance of: delve, navigate, leverage, embark on, underscore, showcase. ChatGPT loves these. No real 17-year-old says “I delved into chemistry.” Use: dig into, figure out, use, start, show.
3. Pretentious metaphors
Tapestry. Mosaic. Symphony of. Kaleidoscope. Weave together. These are wedding-toast words masquerading as literary. Cut them entirely. If you really meant something specific, replace with what you actually meant.
4. Em-dashes used as commas (very high signal)
LLMs over-use em-dashes inside long sentences. The pattern “...the result of which — and this is what I want to focus on — was a shift in perspective” trips detectors hard. Em-dashes are fine for actual asides (like that one), but never as commas in a long, balanced sentence.
5. The triple-parallel structure
“Not just X, but Y, and ultimately Z.”ChatGPT writes this so often it’s basically a tell. Pick one element and cut the other two.
6. Over-balanced sentence rhythm
AI prose has even rhythm — every sentence is roughly the same length. Real teen writing jerks: 3 words. Then a 25-word run-on with a tangent in the middle. Then nothing. If three sentences in a row are the same length, break the pattern.
7. Abstract nouns where verbs should be
“The realization that” → “I realized”. “The development of my interest” → “I got into”. AI nominalizes everything because it sounds academic. Humans verb.
8. Hedging modifier stacks
“Perhaps somewhat in many ways this experience...”Each hedge alone is fine. Stacked, they’re LLM tells. Pick one or none.
9. The over-confident adverbs
Ultimately. Truly. Remarkably. Fundamentally.These don’t add meaning, they add “machine-trying-to-sound-smart.” Strip them.
10. Missing weird-but-true details
AI cannot invent the brand of cereal your grandmother bought, or the specific street sign you walked past every day, or the off-handed thing your coach said in 2022. Add ONE weird-but-true detail per paragraph. Detectors can’t fake them. Readers love them.
11. No contractions
Real teens write “I’m,” “don’t,” “can’t.” AI tends toward “I am,” “do not,” “cannot.” Use contractions unless the formality is genuinely warranted.
The 5-minute self-fix
Run this on every essay before you submit:
- Cmd-F “Furthermore”, “Moreover”, “Additionally”, “In conclusion”. Delete every one.
- Cmd-F “delve”, “navigate”, “leverage”, “embark”, “showcase”. Replace each.
- Read aloud. Anywhere two sentences in a row are the same length, break one.
- Add ONE weird-but-true detail per paragraph.
- Replace 3 abstract nouns with verbs (“the realization that” → “I realized”).
That’s 80% of the value. The other 20% is doing this with an AI that flags everything line-by-line and shows you the swap — which is what we built ApplyAI’s humanizer to do.