Annie Tsai

For most professionals today, writing is no longer a side task. It’s how you show your thinking, influence decisions, lead teams, and shape outcomes. Yet writing remains one of the highest-friction parts of the job for many. The “brain-to-paper” (or in most cases, keyboard) struggle is very, very real filled with blank pages, endless edits, and largely inconsequential revisions that do little else but burn time.

And so many of you have turned to AI as the savior in generating content that looks and sounds enough like you to pass but causes a double-take here or there. So have we all moved beyond AI slop? The answer in the long tail is most definitely no, but let’s hope many of you are already deep in the journey. Those who use AI tools regularly to contemplate and pressure test ideas or organize information will be the first to tell you that while AI content generation tools have advanced quickly, using AI to completely write for you is most certainly not there yet.

Instead of aiming for “Can AI write for me?”, consider instead “How do I write better, faster, and more consistently with AI as a partner?”

Writing with AI Is a Leverage Play

The most competitive people heading into 2026 will be the ones that understand how to use their tools to scale their thinking into clear, fast, and high quality communication across channels, formats and styles, and teams. And they’ll be able to do it without generating the most content or hiring more people to get the job done. The trick here is not to hand over the writing, though. Instead, you need to teach our tools how you think so that it can meet you at the edge of execution.

Start Here: Train for Speed and Fidelity

Here’s what high performing AI writing assistance looks like:

1. Feed it your best writing to establish the baseline. Make sure that there’s enough variety and quantity so you can cover multiple variations of similar styles. Include different content types like memos, articles, blog posts, think pieces, sales pitches, voiceovers, speeches and presentations, product documentation – all of it. Show the system what “great” and “accurate” looks like.
2. Be explicit in your voice, as well as when your voice has variation. Add a sentence in every prompt that clarifies the tone you are targeting upfront. Examples are, “Use a curious, share, and slightly irreverent tone,” “Use simple, punchy sentences and do not use acronyms or confusing industry terminology,” or “Use a more personal tone, with longer, more verbose sentences but keep the target reader as an industry marketer.” The more consistently you reinforce your voice and writing style(s), the faster AI can reflect it accurately.
3. Give feedback in the edit. Active feedback helps all AI models improve their output targeting, so don’t just delete paragraphs you don’t like. Tell your LLM why it didn’t work. Was it too generic? Was the framing such that it sounded like an AI wrote it and not you? If the latter, write it how you would write it and include that as feedback. The feedback signals over time are what separate finely-tuned partners from poor outputs.
4. Pair it with your real workflow. Have your AI tools draft internal updates, restructure your deck outlines, rewrite your LinkedIn posts so it can learn. The fastest way to get better output is to put the system inside your actual content stream and learn with realtime editing.

And Always, Avoid These AI Tells

It’s still very easy to tell when content is AI generated. But, how do you know the tells? Oh, let me count the ways.

AI is very good at injecting subtle tells in very not-so-subtle ways:

• Using corporate cliches in ways that you never would (“cutting-edge solutions”, “dynamic synergies”)
• Saying something and nothing at the same time (“In today’s fast-paced world, innovation is key”)
• Hedging everything to the point of exhaustion (“While x is true, y is also worth noting” – and doing this four times in one output)
• “Isn’t” to “Is” framing (“The problem isn’t AI. The problem is you.”)
• Repeating everything in three’s (“We believe AI helps. AI is useful. It can be helpful.”)
• Really, really short sentences in series’ (“You tried. I tried. We all tried the ice cream. And nothing.”)
• Close with an empty call-to-action (“So, let’s embrace the future – together.”)

Now, I’ll be the first to admit that I do some of these things in my natural writing. I write a Tuesday column for print and online newspaper the Daily Journal out in the San Francisco Bay Area, and for a long time, I fought the em-dash in favor of “space-hyphen-space” formatting. But unfortunately, that was edited into oblivion in my work and now I’ve grown fond of the em-dash that was one of the first “AI tells” the writing community ganged up on. I also like the having examples in three’s as it gives me the opportunity to use an Oxford comma (now you have an idea which decade I learned to write), and I will take every chance I get to inject an Oxford comma into my writing.

The real tell is here is the repeated combination of these writing structures and a few additional formatting decisions and your content starts to scream “AI WROTE ME!”

The goal here is not to turn over your writing entirely to AI while you sip a cocktail. Instead, think of AI as the tool that can help you organize your thinking, compress your synthesizing time, and edit to satisfaction faster and better than you ever have before.

(See what I just did here?) 😉

Close: Build Your Writing Stack

In order to build the system that helps you consistently write better, you need to train it to work with you. Start with this:

1. Pick several pieces that you’re proud of. Feed them as inputs into whatever AI tool you prefer to use and ask it to summarize your tone, writing style, structure, and for each piece, ask it to describe the target audience. Work with your AI to establish writing personas if there are variants that you want to use as style outputs. Give each of the personas a name that you can reference in the future.
2. Create one prompt that you can reuse weekly for a status update, a post, report, or something else that is relatively low-stakes.
3. Test your AI on content pieces that are small but real such as an intro paragraph, a cold email, a bio summary rewrite, translating notes to insight outputs, etc.

The mistake here is in trying to replace the work of writing with AI. Instead, train your AI tools to work with you. The writers and operators who build this stack now will ship faster, more cohesive and compelling content that gets the job done.

I’d love to hear your feedback on how you’ve trained your AI tools to support you in shipping faster and better in comments – Happy New Year!

(Yes, I wrote this newsletter all by myself.)