Write Like a Human Using AI: 7 Strategies That Actually Work

AI writing has a tell, and most readers can detect if something was written by AI within a few seconds, even if they can’t explain why that is exactly. You might be wondering why I speak with so much confidence, right?

I run Undetectable AI, which means generating and humanizing Contenu de l'IA is not new to me, nor is it a side experiment; it is what we do on a large scale, every day at Undetectable AI.

Due to my experience, I have learned that the problem with AI writing is usually not with the writing itself; it has more to do with the predictability of the patterns. 

Over time, my team and I at Undetectable AI have tested several strategies that have claimed to fix AI writing. The kind of strategies that claim to help you achieve 100% human written scores. So I can boldly say we have seen firsthand what works and what doesn’t.  

In this article, I will focus on seven tested and trusted strategies that actually work. I have put these strategies through the same 400-word baseline draft, measuring how each change in strategy affected Détecteur d'IA scores each time.

You will be getting before and after examples with the detector scores after each strategy is applied. 

This article will also help you understand that AI doesn’t write badly; it just writes uniformly. The fix is variation in rhythm, using specific examples, and the kind of voice that an artificial intelligence model cannot produce due to lack of human experience.

---

Principaux enseignements

• AI writing has 5 recognizable tells: predictable sentence rhythm, tricolon overuse, em-dashes, “in today’s fast-paced world” phrasing, and generic examples that don’t have a real source. 

• The 7 strategies in this article were applied to the same 400-word baseline draft, and they moved the detector score from 100% AI to 45% AI on the Undetectable AI Detector.

• The single highest-leverage strategy is varying sentence length. Varying sentence length moves the detector score more than any other strategy.

• Always run a detector check before publishing. The same patterns that detectors’ flags are the same ones readers notice. The Undetectable AI Detector is a free-to-use option, but any reputable detector works too

---

Why AI Writing Has a Tell (And What It Actually Sounds Like)

Most people see AI writing and think low quality, but it is not. AI doesn’t write as badly as you think; it only writes uniformly. Artificial Intelligence writes too consistently.

The sentences tend to have the same range, and the vocabulary is more formal than usual. For instance, where a person would ordinarily say ‘use,’ AI would say ‘utilize.’

While it is possible for readers to tell if the content is AI-written, they find it hard to pinpoint the exact characteristics that make it artificial.

So when they can’t identify the problem, their instinct is to disengage from the content and find something else to satisfy their curiosity. That is why AI-generated content underperforms compared to human-written content. 

The 5 Most Common AI Writing Tells 

It is easy to detect AI writing once you know the exact things to look out for.

Below are five common AI writing tells: 

• Predictable Sentence Rhythm: AI sentences have a consistent length, ranging from 15 to 20 words. They also use the SVO pattern. You can also predict the rhythm because these sentences tend to follow the same cadence. In a way, you end up getting a smooth but overly uniform prose lacking in variation.
  
• Tricolon Overuse: Artificial intelligence has a way of using a series of three words that are parallel in structure, length, or rhythm. That is why you would usually find phrases like “clear, concise, and compelling” appearing too often in an AI text. Humans use these words, too, no doubt, but not as consistently as AI.
  
• Em-dashes for Drama: It is not really popular for humans to use em-dashes more than twice in an essai, especially if it is not necessary. However, AI uses em-dashes to add drama to a text instead of using actual sentence variety.
  
• Cliché Openings: AI writing tends to start with phrases like ‘in a fast-paced world.’ These openings are generic and add no specificity or insight to the text.
  
• Generic Examples: Rather than referencing specific evidence, AI may say “many studies show.” A human writer in this case is more likely to name the exact study, the company, the year, and the finding. In a way, you can say that specificity is a clear marker in differentiating AI writing from human writing. 

Research Methodology 

To prove that it is possible to write like a human using AI, I ran a controlled test using a 400-word baseline draft on the benefit of remote work generated by ChatGPT.

I made no edits to the draft before running it through three detectors, including the Undetectable AI Detector, GPTZero, and Originality.ai on Friday, ‎12 ‎June ‎2026, ‏‎13:47. Here are the results:

Baseline Draft

1. IA indétectable

2. GPTZero

3. Originality.ai

After the initial testing confirmed that the 400-word text generated by ChatGPT is AI-generated, I applied each step of the strategy one at a time to that same draft, rescoring it on all three detectors after every strategy was applied one at a time. 

The 7 Strategies to Write Like a Human When Using AI

The following are strategies that can help you if you have been searching for how to make ChatGPT write like a human: 

Strategy 1: Vary Sentence Length Dramatically 

This is the single highest leverage edit you can do to an AI draft, because AI gives you about 15 to 20 words per sentence.

Human writing lets you have short sentences that can stretch into long sentences and later break into fragments while still maintaining the same idea. This strategy moved the draft from 100% AI to 92% AI on Détecteur d'IA indétectable. 

Avant : 

“The popularity of remote work has grown in recent years. It has a lot of advantages for both the employees and the companies. Research has shown that remote workers are more productive than those working in an office.” 

Après :

“Remote work exploded after 2020. The reason? Nobody wants 90 minutes of traffic anymore. And once companies saw that output didn’t drop, in many cases it went up, the old argument for five days in the office started falling apart fast.”

Strategy 2: Replace Generic Examples With Real Specifics

The default phrase for AI when it needs to be specific is to say things like “many studies show,” “businesses are finding success,” “experts agree.” AI uses these phrases to say things without actually saying anything meaningful. 

For human beings, we name the study, the company, the year, and the findings. Specific examples are more credible and useful to the reader. When we made this swap on our baseline draft, the detector score improved.

Avant :

 “Studies have shown that remote workers are often more productive than those working in traditional office settings.”

Après :

 “A 2023 Stanford study by Nicholas Bloom found a 13% productivity increase among remote workers at a Chinese call centre, largely driven by fewer breaks and sick days.”

Strategy 3: Kill the AI Phrase Library

There are some phrases that are already associated with AI output, and just one appearance flags the entire paragraph.

We went through the draft and removed the following:

• “In today’s fast-paced world.” It is just better to replace it with the actual point.
• “It is important to note that.” You should just go straight to the point because the reader already knows it is important 
• “In conclusion” / “To wrap up.” It is better to use a real closing thought.
• “Undeniably,” “certainly,” “absolutely.” State the truth or fact itself 
• “Tapestry,” “delve into,” “navigate the complexities.” 

Strategy 4: Add Personal Experience or Real Observation

AI is incapable of having a personal experience, so it cannot fake firsthand experience. So, including at least one sentence of real human experience changes the entire authority of a paragraph. 

This strategy is also known as the E-E-A-T strategy. That is, Experience, Expertise, Authoritativeness, and Trustworthiness. It is the kind of strategy that Google rewards the most in 2026 because it helps your search credibility. 

Avant :

“Remote work can have a positive impact on employee wellbeing. Many workers report feeling less stressed when they can work from home. This can lead to improved mental health outcomes over time.”

Après :

 “When we surveyed our own team after 18 months fully remote, the number that stood out wasn’t productivity, it was the drop in reported burnout. Sixty percent said they felt less overwhelmed than they had in the office. That wasn’t a study. That was Tuesday.”

Strategy 5: Use Contractions and Causal Phrasing

AI’s default is to write formally. It uses formal English in a way no actual person writes. Rather than say ‘Don’t, ‘ it says ‘Do not.’ ‘It is’ instead of ‘it’s.’

To make your work sound more human, you should use contractions and more casual phrasing without changing any of the substance. 

You can also add occasional fragments like starting a sentence with ‘And’ or ‘But.’ Breaking the rules of AI writing is what actually signals that your text is a product of your effort.  

Avant :
“Remote work does not suit every type of employee. It is important to consider individual working styles before implementing a remote work policy. Some employees may find that they do not perform as well outside of a structured office environment.”

Après :
“Remote work doesn’t suit everyone. Some people genuinely need the structure of an office, and that’s fine. But the policy shouldn’t be built around the exception.”

Strategy 6: Take a Stance

AI by default never takes a stand. It is always trying to make a balanced argument. It would say something like “This approach has both benefits and drawbacks” rather than picking a side.

So it now falls on you to ask it to pick a side and give reasons, just like how humans do. Force a clear position in your prompt and see the kind of results that you would get. 

Avant :
“Remote work has both advantages and disadvantages. While it offers flexibility and can improve productivity for some employees, it can also lead to feelings of isolation and make collaboration more difficult. The right approach will depend on the needs of each organisation.”

Après :
“Remote work is better for most knowledge workers. The isolation argument is real, but it’s solvable. The collaboration argument is mostly a management failure rather than a structural one. Companies that defaulted back to five days in-office didn’t do so because the data told them to.”

Strategy 7: Run a Detector Check + Humanize Where Needed

After applying the six strategies above, you can run the results before you publish.

If the score is high, you then have two options: do more human editing or use a humanization tool like Humanisateur d'IA indétectable. Both options will give you good results, but nothing beats human editing, though you might just spend more time doing it on your own. 

At Undetectable AI, using a detector is part of our internal workflow. No content goes out without verification, especially AI-generated content. 

Real Before & After: One Draft, 7 Strategies, Detector Scores

This is what “how to make ChatGPT write like a human” looks like in practice. Same 400-word draft. 7 strategies. Real detector progression.

Stade	IA indétectable Détecteur	GPTZero	Originalité.ai
Baseline AI draft	100% AI	100% AI	100% AI
+ Strategy 1 (sentence variation)	92% AI	85% IA	94% IA
+ Strategy 2 (specific examples)	80% AI	86% AI	89% IA
+ Strategy 3 (kill AI phrases)	75% IA	80% AI	80% AI
+ Strategy 4 (real observation)	71% AI	75% IA	72% AI
+ Strategy 5 (contractions)	68% AI	72% AI	71% AI
+ Strategy 6 (take a stance)	65% IA	50% AI	35% AI
+ Strategy 7 (humanize remaining)	45% AI	4% AI	1% AI

Final Draft (after the seven strategies)

1. IA indétectable

2. GPTZero

3. Originality.ai

Prompt Patterns That Produce Human-Sounding Drafts 

While the 7 strategies above work as post-draft edits, you can still cut down on your editing time by using the right prompts.

These are three patterns that you can use to create human-sounding drafts: 

1. The “Specific Persona” Prompt 

One thing I have realised about people and the way they use AI is that they believe that all they need is just a topic and word count, and the AI model will do the bulk of the work.

However, what they fail to realise is that by doing this, they have given AI leverage to use its default voice. 

That is, they let Artificial Intelligence use the exact uniform, formal rhythm we have been talking about. So, what you need to do when prompting AI is to be specific about the writer persona that you need before making any content request. 

You can use this as your opening line before every prompt:

“You are Beth, a financial analyst writing for finance enthusiasts. Your writing voice is conversational, direct, and practical. You use contractions, vary sentence length, and avoid generic AI phrases like “in today’s fast-paced world.”

After this prompt, you can then input your content request.

2. The “Take a Stance” Prompt 

AI tools are trained in a way that prevents them from taking a definite stance. These tools try to avoid controversy by producing results that tend to present both sides.

This instinct allows the tool to produce writings that say a lot without saying anything. So, to get the best result, you have to ask the tool to take a stance, 

To do this, you can use this prompt:

 “Take a clear stance on financial responsibility. Explain your reasoning and the trade-offs. Use first person if it helps. Be direct. No hedging language. Keep paragraphs ≤4 lines. Avoid tricolons and clichés.” 

3. The “Avoid List” Prompt 

To avoid wasting your time, it’s best to tell your tool exactly what not to do. You can make an ‘avoid list’ that gives the tool an idea of what is okay to include and what is not. Over time, you can even build a list as you notice which patterns keep showing up in your draft. 

For example, you can use a prompt like:

“Write [the request]. Avoid: in today’s fast-paced world, it is important to

note, em-dashes for drama, tricolons (X, Y, and Z), tapestry, delve into, navigate the complexities, in conclusion.” 

Questions fréquemment posées 

Réflexions finales

The takeaway from this article should be that AI does not write badly, just uniformly. That same uniformity is what readers feel before they can figure out what makes any text awkward.

When we applied the seven strategies to the same 400-word draft, the detector score moved from 100% AI to 45% on Undetectable AI.

However, the most important part is how the writing itself moved from something that was technically correct to something that was actually worth reading. It became more specific, the sentences became more varied, and more humanly written. 

After following the seven strategies, if you still want to use a detector to make your work airtight, the Détecteur d'IA indétectable is free to use.

You can get an analysis of where your content stands with the free quota we offer before you add your own edits and perspective.