How to Use AI Without It Sounding Like AI

You've seen it. The email that's slightly too polished. The LinkedIn post with that weird cadence. The document that uses "delve" and "leverage" three times each. Something feels off, even if you can't immediately say why.

AI-generated content has a smell. People are getting better at detecting it. And when they do, it undermines the whole point of the communication.

I use Copilot in Office 365 every day. It drafts emails, summarizes meetings, helps with documents. It saves me hours each week. But I've learned that the default output is almost never ready to send. The AI gets you 70% of the way there. The last 30% is what makes it actually sound like you.

Here's what I've learned about using AI as a starting point without ending up with content that screams "a robot wrote this."

Why AI Content Sounds Like AI Content

Before fixing the problem, it helps to understand why AI output has that distinctive flavor.

AI writes for the average case. Large language models are trained on massive amounts of text. They learn to produce output that's statistically likely to be acceptable. The result is prose that's competent but generic. It lacks the specific voice, quirks, and preferences that make writing feel human.

AI hedges constantly. Notice how often AI uses phrases like "it's important to note" or "there are several factors to consider." It's covering its bases, trying not to be wrong. Humans with actual opinions don't write like this.

AI loves certain words. Every model has its favorites. "Delve." "Leverage." "Facilitate." "Comprehensive." "Robust." These words aren't wrong, but they're overused to the point of becoming AI fingerprints.

AI structures things predictably. Introduction, three main points, conclusion. Bullet points for everything. Headers that telegraph exactly what's coming. It's organized to a fault.

AI is relentlessly positive and balanced. It rarely takes strong positions. It acknowledges multiple perspectives even when one is clearly right. It avoids anything that might offend anyone. Real humans have edges. AI has been sanded smooth.

The Copilot Workflow That Works for Me

I use Copilot in Outlook, Word, and Teams daily. Here's my actual process for getting useful output that doesn't sound artificial.

Email Drafting

What I do: I use Copilot to generate a first draft, then rewrite at least half of it.

The prompt matters. Instead of asking Copilot to "write an email to follow up on our meeting," I give it context and constraints:

"Draft a short follow-up email to Erik after our meeting about their storage infrastructure. Mention that I'll send the TCO comparison by Friday. Keep it under 100 words. Casual but professional tone."

The more specific the prompt, the less generic the output.

What I change:

The opening line. Copilot loves "I hope this email finds you well." I delete this every time.
Filler phrases. "I wanted to reach out to" becomes "I'm." "Please don't hesitate to" becomes "Let me know."
The sign-off. Copilot is overly formal. I adjust to match how I actually communicate with that person.
Anything that doesn't sound like me. If I wouldn't say it out loud, I rewrite it.

Example transformation:

Copilot draft:

"I hope this email finds you well. I wanted to reach out following our productive meeting yesterday to express my appreciation for your time. As discussed, I will be sending over the comprehensive TCO comparison by Friday. Please don't hesitate to reach out if you have any questions in the meantime."

My version:

"Good talking yesterday. I'll have the TCO comparison over to you by Friday. Give me a shout if anything comes up before then."

Same information. Half the words. Actually sounds like a person.

Meeting Summaries in Teams

Copilot's meeting recaps are genuinely useful. But they read like meeting minutes from a court reporter.

What I do: I use Copilot to capture the facts, then rewrite the summary for whoever needs to read it.

What I keep: Action items, decisions made, key numbers or commitments. The factual stuff Copilot captures well.

What I rewrite: The narrative summary. Copilot produces something like "The team discussed various approaches to the implementation timeline and reached consensus on a phased approach." I'd rewrite this as "We agreed to do the rollout in two phases. Pilot in Q1, full deployment in Q2."

Shorter. Clearer. Human.

Documents in Word

For longer documents, Copilot is useful for generating structure and getting past the blank page. But the output needs significant editing.

What I do: I ask Copilot to outline or draft sections, then rewrite for voice and cut aggressively.

Specific tactics:

Remove every instance of "it's worth noting that." Just state the thing.
Cut the warm-up paragraphs. AI loves to introduce what it's about to say before saying it.
Add specifics. AI writes in generalities. I add concrete examples, numbers, and names.
Vary sentence length. AI defaults to medium-length sentences. I mix in short punchy ones. And occasionally a longer one that builds.
Delete duplicate ideas. AI often says the same thing multiple ways to fill space.

The Words and Phrases to Hunt Down

When editing AI output, I search for these and usually delete or replace them:

"It's important to note" and "It's worth mentioning" (just say the thing)
"In order to" (use "to")
"Delve," "leverage," "facilitate," "comprehensive," "robust"
"Utilize" (use "use")
"Excited to" and "Thrilled to"
"I hope this email finds you well"

You get the idea. Once you start noticing the patterns, you'll spot them everywhere.

Voice Matching: Teaching AI How You Write

One technique I've found useful is giving Copilot examples of my own writing.

In Word, I'll sometimes paste a previous email or document I've written and ask Copilot to "match this tone and style" for new content. It's not perfect, but it gets closer to my voice than starting from nothing.

For recurring content types, I keep a few examples handy:

A follow-up email I liked how I wrote
A proposal intro that felt right
A LinkedIn post that performed well

When I need similar content, I reference these as style guides.

When to Skip AI Entirely

Sometimes AI isn't the right tool. I've learned to recognize these situations.

High-stakes personal messages. If I'm writing to congratulate someone on a promotion, express condolences, or navigate a sensitive situation, I write it myself. These moments require genuine human attention. Using AI would be disrespectful even if no one could tell.

When I need to think, not just write. Sometimes the value of writing is working through ideas. Outsourcing the writing to AI shortcuts the thinking. If I'm trying to figure out what I actually believe about something, I need to write it myself.

Very short messages. For a two-line email, opening Copilot and prompting it takes longer than just typing. AI shines for longer content where the time savings are real.

When authenticity is the point. Some content needs to be visibly human. A personal blog post. A heartfelt recommendation. The human fingerprints aren't bugs, they're features.

The 70/30 Rule

Here's the mental model I've landed on: AI gives me 70%, I provide 30%.

That 70% is valuable. It's structure, first-draft language, captured information, and time saved. I'm not dismissing it.

But the 30% is what makes the content actually work. It's voice, judgment, editing, and the specific details that make writing feel real. That 30% is also where the thinking happens.

If you're using AI output at 100%, sending whatever it generates, people will notice. Maybe not consciously. But something will feel off.

If you're providing that 30%, shaping the output, adding your voice, cutting the fluff, the AI becomes invisible. It's just a tool that helped you work faster. Which is exactly what it should be.

The Real Skill

Using AI well isn't about prompting. It's about editing.

Anyone can ask Copilot to write an email. The skill is knowing what to change, what to cut, and what to add to make it sound like it came from a human with actual opinions and a distinct voice.

That skill is worth developing. As AI gets more common, the gap between people who use it thoughtfully and people who use it lazily will become more obvious. The robot-sounding emails will pile up in inboxes, ignored or deleted. The ones with human fingerprints will stand out.

Be the person whose AI-assisted content doesn't smell like AI. It's not that hard. It just takes the willingness to treat AI as a starting point rather than a finish line.

What are your tells for AI-generated content? I'm always looking for new patterns to watch for.