Getting AI to Write Like a Human Has Gotten Complicated
Getting AI to stop sounding like a corporate memo has gotten complicated with all the conflicting advice flying around. Try this prompt structure. No, use this one. Add these magic words. I spent about three months obsessively testing different approaches — across ChatGPT, Claude, and Gemini — before anything clicked. Today, I’ll share it all with you.
But what is the actual problem here? In essence, it’s a training data issue. But it’s much more than that.
AI models learn by consuming billions of words — academic papers, published articles, edited content, legal documents. Almost all of it is polished within an inch of its life. When you ask ChatGPT to write something, it doesn’t have a “chill out” button. It pattern-matches to what it saw most during training, which skews heavily toward writing that humans were paid to make sound professional. The model isn’t being difficult. It’s doing exactly what it was optimized to do. Once I understood that, I stopped fighting the tool and started working around its actual limitations.
Start With a Better Prompt
The single biggest lever is your prompt. Most people write something like this:
Write an article about remote work productivity tips.
And they get back something that reads like a consulting whitepaper from 1997. Instead, try this:
Write about remote work productivity tips like you’re texting a friend who just complained about getting distracted at home. Skip the intro. Start with something real. Use “I” and “you.” Assume we’re already on the same page.
The difference is enormous. That second prompt does four specific things:
- Gives the model a concrete social context — texting a friend, not addressing a boardroom
- Cuts the formal scaffolding before it starts (“In today’s increasingly remote landscape” never appears)
- Forces first and second person pronouns throughout
- Signals shared understanding instead of asking it to explain everything from scratch
Here’s another example. Stiff version:
Compose a guide on email management for busy professionals.
Natural version:
Write about how to stop email from eating your entire day. Write like you’re giving advice to someone in your Slack group who’s drowning in messages. Be specific. Use examples. Don’t explain what email is.
That last line — “don’t explain what email is” — is doing more work than it looks like. It tells the model the audience already lives inside this context. That single instruction triggers a less formal, more direct tone almost immediately.
The Five Edits That Loosen Any AI Draft
Even with a solid prompt, you’ll still need to edit. I’ve landed on five specific moves that transform nearly every stiff AI output. Use all of them when you have time. Pick the ones that fit when you don’t.
1. Cut the Throat-Clearing Opener
AI loves starting with “In today’s digital landscape” or “It’s increasingly important to understand that.” Delete it completely. Start somewhere that actually moves.
Before: “In recent years, the importance of maintaining work-life balance has become increasingly recognized by modern professionals.”
After: “Your inbox doesn’t care about balance. It will eat your evening if you let it.”
2. Replace Nominalizations With Verbs
Nominalizations are nouns made from verbs — “implementation” instead of “implement,” “utilization” instead of “use,” “the optimization of workflows” instead of just “optimizing workflows.” They’re probably the fastest route to sounding stiff.
Before: “The implementation of these strategies requires the commitment of team members to the process of continuous improvement.”
After: “You need your team to actually commit to doing this stuff — repeatedly, not just once during an offsite.”
3. Break Up Long Sentences
AI sentences routinely creep past 25 words. Sometimes past 40. Find them. Split them. A short sentence followed by a longer one creates rhythm. That rhythm is what makes prose feel like a person wrote it instead of a content generator.
Before: “When considering the various approaches to remote work management, it is essential to account for the preferences of individual employees while simultaneously maintaining organizational standards and ensuring productivity metrics remain aligned with company objectives.”
After: “Remote work needs rules. But the best rules account for how different people actually work — not just what the employee handbook says.”
4. Swap Formal Words for Plain Ones
Probably should have opened with this section, honestly. It’s the easiest fix and it works every single time. Utilize becomes use. Facilitate becomes help. Leverage becomes use again. Commence becomes start. These swaps take maybe 30 seconds and cut formality by half.
5. Add a Personal Observation
Drop in one sentence that sounds like something you’d actually say out loud. Opinionated. Specific. It doesn’t need to be long — even eight words interrupts the formal pattern.
Before: “Remote work presents various challenges that organizations must address through thoughtful policy development.”
After: “Remote work presents various challenges that organizations must address through thoughtful policy development. I’ve watched teams completely collapse at this because they tried to replicate the office at home instead of building something different.”
That one genuine sentence makes the whole thing sound like a person was involved.
Train Your AI With a Tone Sample
Here’s a technique I stumbled onto by accident. Paste a paragraph of your own casual writing directly into your prompt. The model pattern-matches to your voice instead of defaulting to its formal register. I’m apparently a very blunt Slack communicator and that style works for me while vague instructions like “sound more casual” never quite do.
Template you can copy right now:
Here’s how I naturally write: [paste 3-4 sentences of your own writing]. Now write about [topic] in a similar style. Match the tone, sentence length, and how I use examples. Don’t explain basic concepts. Use “I” and “you.”
This works across ChatGPT, Claude, and Gemini — basically any LLM that accepts written input. The model doesn’t get confused by your voice. It leans into it. A concrete style sample beats vague instructions every time. Don’t make my mistake of spending weeks writing elaborate tone descriptions when three sentences of your actual writing does more.
When the Output Still Sounds Off
Sometimes nothing lands — usually because the topic is technical, or because you’ve asked the AI to write something it treats as inherently official. A contract. A formal announcement. Something that lives in a context where casual tone would actually read wrong.
The move that works here is recontextualizing. Tell it to “rewrite this like an email to a colleague you know well, not a formal report.” Or “write it like a Slack message, not a presentation deck.” That shifts the register without making the content inappropriate for the situation.
You’re not asking it to ignore what the content is. You’re asking it to change the social context it imagines while writing. That changes everything about tone — sentence length, word choice, how much it assumes the reader already knows.
The frustration with stiff AI output is genuinely solvable. The model isn’t being formal on purpose. It’s responding to patterns baked into its training and to exactly how you’re asking for things. Change the prompt. Edit ruthlessly. Add a voice sample. Do all three and you’ll get writing that sounds like something a person actually sat down and produced — instead of something assembled from a database of business jargon and LinkedIn posts.
Leave a Reply