You’ve probably read something recently and thought, “Yep. AI wrote that.”
Maybe you were right. Maybe you weren’t.
There is no single word, punctuation mark, or sentence structure that proves someone used AI. Humans were using em dashes, rhetorical questions, and words like exactly long before ChatGPT existed. But there are writing patterns that have become strongly associated with AI-generated content, especially when several of them show up together. Researchers are even finding measurable changes in language since the arrival of ChatGPT, including increased use of words like delve, exactly, and intricate in some kinds of writing.
So this week, let's look at a few of the patterns you may be noticing, and more importantly, what you can tell your LLM to do instead.
1. “It’s not X. It’s Y.”
Once you notice this one, you may start seeing it everywhere.
It’s not about working harder. It’s about working smarter.
It’s not a technology problem. It’s a people problem.
AI isn’t replacing your judgment. It’s amplifying it.
There’s nothing grammatically wrong with this construction. It creates an easy contrast and can be effective when used intentionally. The problem is repetition.
So instead of: “It’s not about replacing the human. It’s about giving humans better tools.”
Try: “Better tools can give people more time and information for decisions that still require human judgment.”
Tell your LLM:
Avoid “It’s not X, it’s Y,” “This isn’t X, it’s Y,” and similar negative-to-positive contrast constructions. State the main point directly instead.And be specific about the variations. Otherwise, your AI may helpfully give you: “This isn’t merely about X. Rather, it’s about Y.”
Same outfit. Different hat!
2. The dramatic fragment
You know this one too:
The truth?
The result?
Here’s the kicker.
The best part?
And the biggest mistake?
These little fragments create rhythm, which is probably part of why they work so well in social content. But when every idea needs a dramatic setup, the writing starts feeling manufactured.
So instead of:
“The result?
Your executive gets hours back every week.”
Just say: “That can give your executive hours back every week.”
Tell your LLM:
Avoid one-line rhetorical fragments such as “The truth?”, “The result?”, “The key?”, “The best part?”, and “Here’s the kicker.” Use natural transitions between ideas instead.3. Everything magically comes in threes
AI seems to love a tidy trio:
Faster, smarter, and more efficient.
Clarity, confidence, and control.
Streamline your workflow, save time, and unlock your potential.
Again, there is nothing wrong with the rule of three. Humans have used it as a rhetorical device for centuries. But when every paragraph contains three perfectly balanced benefits, adjectives, or verbs, your writing can start sounding like marketing copy generated from a template.
Tell your LLM:
Don't automatically organize ideas into groups of three. Use only as many examples, adjectives, or points as the content naturally requires. Vary sentence rhythm and structure.That last sentence matters. You don't want the model replacing every group of three with a group of two and creating a brand-new formula.
4. The AI vocabulary starter pack
Some words and phrases have become so common in AI-assisted writing that people have started associating them with ChatGPT:
Delve. Unlock. Leverage. Seamless. Game-changer.
Navigate the landscape. In today’s fast-paced world.
Research has found significant increases in some AI-associated vocabulary following the release of ChatGPT. But the presence of any one word is not evidence that AI wrote something, and your own vocabulary matters more than any internet list.
So instead of: “Leverage AI to unlock a seamless workflow.”
You could simply write: “Use AI to take a few repetitive steps off your plate.”
Tell your LLM:
Prefer simple, specific language over corporate or promotional vocabulary. Avoid words such as “delve,” “unlock,” “leverage,” “seamless,” “game-changer,” “supercharge,” and “navigate the landscape” unless they are genuinely appropriate to the writer's normal vocabulary.5. The overly polished LinkedIn formula
You’ve probably seen some version of this:
I learned an important lesson this week.
And it changed how I think about leadership.
Here’s what happened:
[Story]
The lesson?
Leadership isn’t about having all the answers. It’s about asking the right questions.
Agree?
None of those sentences is inherently bad. Put them all together, though, and you get a structure that feels less like someone sharing a thought and more like someone filling in a LinkedIn template. A good prompt can explicitly prevent that.
Tell your LLM:
Avoid formulaic social-media structures, engagement-bait endings such as “Agree?” or “Thoughts?”, unnecessary rhetorical questions, artificial suspense, and inspirational conclusions that aren't supported by the actual story. Preserve the natural structure of the original idea.This is also where specific details help.
Compare: “I learned an important lesson about leadership today.” with “My executive moved our 2:00 meeting this morning because she noticed the team needed another hour to solve the problem themselves.”
The second sentence gives the model something real to work with!
So... should we ban all of these?
No. Not really!
Please don't leave this newsletter afraid to use an em dash or accidentally write three adjectives in a row (unless your executive explicitly tells you not to!).
These are patterns to be aware of, not rules to blindly follow. And they’re certainly not proof that something was written by AI. If an em dash is naturally part of your writing style, use it. If your executive loves a good three-part list, keep it. And sometimes “It’s not X. It’s Y.” really is the best way to make a point. The goal is to keep those stylistic choices in your hands instead of letting AI make them for you by default.
And that’s where prompting matters.
A prompt like: “Make this sound more professional.” gives the model very little information about what your version of “professional” actually sounds like.
So give it constraints. Try this on your next draft (or better yet, add it as an evergreen instruction in your LLM so it can consistently follow your writing preferences.)
Draft this in a natural, conversational professional voice. Preserve my meaning and don't add facts or opinions I haven't provided. Avoid:
“It’s not X, it’s Y” and similar contrast formulas
dramatic fragments like “The truth?” or “The result?”
automatically grouping ideas into threes
generic AI/corporate vocabulary
unnecessary rhetorical questions
engagement-bait endings
overly polished inspirational conclusions
repetitive sentence structures
Prefer specific details and straightforward language. Vary sentence length and structure naturally. If I provide writing samples, prioritize the patterns in those samples over generic assumptions about what “professional” writing should sound like.Then edit the result.
Show it what you actually sound like.
Give your LLM a few original emails, messages, or posts you wrote yourself. If you're drafting for an executive, use examples they wrote themselves.
You can even ask the LLM to compare its draft with the final edited version and identify what you changed. Keep doing that, and those edits can become your own DO / AVOID list.
So the next time you catch yourself saying: “Ugh. This sounds like ChatGPT.”, don't just rewrite it. Figure out why it sounds wrong, turn that observation into an instruction, and add it to your prompt for next time. That's how your AI gets better at sounding less like everyone else and more like the person it's supposed to be helping.
AI Updates of the Week
OpenAI introduced Dots, persistent AI agents that can keep working on assigned projects in the background. A Dot has its own cloud computer, can connect to supported apps, carries context across ChatGPT, Slack, and Teams, and can proactively surface work. OpenAI says users can set rules governing which actions a Dot can take independently versus which require approval. The initial rollout is for eligible Pro and Business Premium users, with Enterprise access controlled by admins.
Microsoft announced a major Copilot redesign with Home, Code, and Autopilot. Home brings Chat and Cowork together, while Cowork is designed for work you delegate end-to-end. Microsoft also previewed Today, a personalized command center across email, calendars, Teams conversations, meetings, and tasks. Microsoft says it can identify what needs attention, prepare drafts, propose schedule changes, and surface forgotten follow-ups with the relevant context. Today is expected to enter a private preview in October.
Google added new Take notes for me settings and Quick notes for Google Meet in late September. It’s part of Google’s broader push toward more agentic work inside Workspace, with Gemini increasingly able to coordinate work across Gmail, Drive, Docs, Slides, and Chat.
We’d still love to hear from you!
If you haven’t taken our 2-minute survey yet, there’s still time. Tell us how you’re using AI, where it’s falling short, and what you want to learn next. Your responses help us understand what EAs actually need from AI.
See you next week!
And please keep sharing what you’re learning with us. The more we hear from EAs doing this work every day, the more useful these conversations become for everyone.
The Base Team
