I think of myself as an AI realist (a label stolen from my very smart friend, Hannah Miet Levine). Which is why I find the extreme posturing from both sides of the anti/pro AI debate so tedious. 

No, I don’t think we should be using ChatGPT to write everything for us. 

Yes, I can see how it’s useful for people who aren’t writers.

No, I don’t think we should blindly trust Sam Altman and the people running the big AI companies.

Yes, I want to stay up-to-date on what these tools can and can’t do, so I can use them accordingly and explain to clients when they make sense vs. when they’re just overcomplicating things.

And most importantly, I want to be transparent about how I use these tools and how I don’t. 

(A quick caveat: I follow client instructions and contracts on how and when to use AI to the letter. Some clients encourage it, and some prefer I don’t use it at all. Both are totally fine with me.)

AI tools are like microwaves: useful for some things, terrible for others

Sometimes you need chicken nugget content – standardized, quick, straightforward. Nothing too adventurous, but reliably tasty. We’re talking short product descriptions for a brochure, or the outline of a paragraph I’m stuck on.

That’s when I have no problem using the microwave, AKA a generative AI tool like Claude or ChatGPT. 

For example, I write the first product description, then input instructions for Claude to write the next 10 in the same style based on the source material. 

But I don’t use it to write my drafts or strategy docs. That would be like trying to make chicken piccata in the microwave. For that, I’m cooking from scratch on the stovetop.

I'm not ashamed that I have a microwave and sometimes use it. That doesn't mean I use it for everything.

The professionals who are thriving in the AI era understand this distinction. They're not trying to use ChatGPT to write their brand's signature thought leadership piece any more than they'd microwave a Thanksgiving turkey.

How I use AI tools, with examples

I generally turn to a handful of tools on a regular basis: 

• ClaudePro (the paid version doesn’t send my inputs to the training model, though I’m still cautious about what I enter) for editing, outlining, and brainstorming

• Perplexity for research and fact-checking (with a big caveat that I fact-check it myself)

• Writesonic for SEO research (though Perplexity can also do this and I’m always down to streamline)

• ChatGPT for planning and random questions (like finding kid-friendly restaurants in San Diego or suggesting recipes from what’s in my fridge)

Here are some examples of exactly how I use them for my work:

Sunday planning sessions

I tell Claude what I have on deck for the week — tasks, clients, deadlines, estimated time requirements, and my schedule. We go back and forth to create a schedule that works, so I don't have to think about prioritization during the week.

Interview transcript cleanup

I input sections of interview transcripts into Claude and ask it to clean up filler words. I always double-check any quotes I end up using against the original transcript, but this helps with readability and finding the best parts to use.

I will also input the article brief or outline and ask it to suggest the best quotes to pull from the transcript. I don’t always take the suggestions, but it usually surfaces some good ones and helps jog my memory of the convo faster than rereading the whole thing.

Draft comparison

If I'm feeling stuck on editing and I have an example of the finished product I'm going for (like a LinkedIn post my client loved or an article that performed well), I'll ask Claude to compare my draft to the example and give me tips on how to reach the same level. This has required some training on my part, but it's a useful sounding board.

Research starting points

I use Perplexity to jumpstart research, much in the same way I used to rely on Google. It’s still super important to verify the sources, but I find it more intuitive than Google these days. 

A great example: I work with a client who sells to both a US and Canadian B2B audience, and we create two versions of blog articles for each audience. I input sections that are US-specific (mentioning US stats or government agencies, etc.) and ask it to fact-check it and suggest substitutions for a Canadian audience. I also specify what I prefer for primary sources and ask it to only rely on info published within the past 3 years. 

Templates

I’ve worked with Claude to make custom templates for myself, for things like proposals, content audits, and brand strategy playbooks. 

A million ideas to choose from

When I write a headline and know that it’s subpar, I will ask Claude to give me 10-15 alts on the headline. Then I either pick one and tweak it, or read the list and get inspired to write a better one myself. 

It also really helps when I can’t figure out a good transition between sections. I rarely use the actual copy it provides as suggestions, but it gives me some jumping off points.

The bottom line: I feel comfortable using these tools for the functions mentioned because I am quite aware of their limitations and never blindly trust the outputs. Which leads me to…

The veracity problem

Recently, I used Perplexity to find a stat for an article I was writing. It took me 8 minutes just to verify one statistic, discover that the initial results were all based on 11-year-old information, and then find an accurate, up-to-date number.

Fact-checking Perplexity results - Watch Video

LLMs aren't magic genies. They're searching the same internet we all have access to, and they aren't digging into every source with a critical and analytical brain.

Another more nuanced example

Perplexity gave me this sentence in response to a query:

“By 2030, projections indicate 70% of global consumers will primarily shop through social media.”

The source it cited for this was a blog post that cited an industry publication that cited an original source: DHL's 2025 e-Commerce Trends Report. 

But when I dug into the actual report to verify, I realized that sentence was misrepresenting the data. It was written as if there has been a mathematical extrapolation of data to get a "projection,” when it was actually just citing a survey result.

My rewrite for accuracy: “In a recent survey, 70% of respondents predicted that they would be primarily shopping on social media by 2030.”

I changed it to make clear that a representative but small sample of people are predicting their own future behavior. It's not as authoritative, but it’s ACCURATE.

PSA: This is why you hire journalists for your brand content. We really, really care about accuracy.

I’ve also found that most AI tools will hallucinate when asked to analyze text you give them. I once asked Claude to find repeated words or phrases in a draft I had written, and it repeatedly fabricated several different repeated words. 

You basically have to treat these tools like earnest, green interns: Expect to invest some time in teaching them, don’t expect them to get everything right, and know that ultimately YOU are responsible for the quality of the work.

How other content pros are using AI tools

Here are a few notable examples from my network. (Click the images for the full posts.)

Calvin Hennick regularly shares examples of how he uses AI, and I find it so helpful.

Sarah Greesonbach shares her specific method for pulling quotes out of transcripts for writing white papers.

And my friend and collaborator Hannah Miet Levine on how she uses AI without letting it think for her:

AI is still a terrible writer

I’ve tried, mostly for funsies, to have Claude or ChatGPT generate whole drafts. They are uniformly Not Good: formulaic, anodyne, predictable, full of made up shit, illogical, uncanny valley-esque regurgitations. 

I asked Claude to write this newsletter, giving it a bunch of LinkedIn posts I’ve written on the topic and then all my back issues for examples of my tone and style. It did not go well. (This was written 100% by me, in case that’s unclear.)

Sometimes I’ll do what Erica Schneider recommends and get a Not Good first draft, maybe 80% of the way there, that I can aggressively edit to 100%. I’m fine doing this because I’ve been a professional editor for 15+ years. So whether it’s a sloppy, typo-ridden rough draft from a stringer for a local newspaper or a bellicose, stuffy “thought leadership” speech by an exec, I know how to shape it right up. Same with AI drafts. 

But I wouldn’t recommend this method to people who aren’t trained editors, because it ends up being way more effort and work than if you wrote something yourself.

What I’m reading/watching/staring at

Because a marketer is only as good as their creative inputs

FICTION 📖 Tom Lake by Ann Patchett | Another exquisite novel by one of my favorite authors. It’s the first book I’ve read set during the pandemic that didn’t make me cringe. A juicy, hopeful story about what leads people to the right path.

NONFICTION 📖 The Mother Next Door: Medicine, Deception, and Munchausen by Proxy by Andrea Dunlop & Mike Weber | I’m only 1/3 of the way through this, but it’s really well done. If you followed the horrid tale of Gypsy Rose Lee or that one scene from The Sixth Sense stuck with you, it’s worth reading this book. Dunlop is a journalist with a personal connection to Munchausen by proxy, and Weber is a medical child abuse investigator who’s seen many of these cases. It’s really eye-opening to learn that this is not some gothic, rare disorder; many “normal” middle-class moms have been found to be harming their children.

(I also find it an absolutely vital piece of the argument in favor of robust public school and social work systems in this country — the right-wing push for “parental rights” and more public funds for home-schooling and private religious schools erodes the safety nets that catch kids who are victims of medical child abuse. It’s so difficult to detect and diagnose, and the parents who commit it hide behind privacy and parent’s rights to conceal their awful crimes. End of rant.)

LAKES 💧 Tahoe & Michigan | Last month, I got to spend a weekend in Lake Tahoe with friends and then a week in Saugatuck, Michigan with family. Swimming in Lake Michigan with my son was one of the most joyful things I’ve done all year.

Arthur and I have a li’l surprise for you!

My 43-page LinkedIn Starter Pack is now completely FREE. You can download your copy here.

It’s got walk-through videos, checklists, prompt ideas, AI prompts for creating your own style kit, and lots of examples. I basically included everything I learned from trial and error over nearly 2 years of posting consistently and growing my audience and solo business on LinkedIn.

Thanks for reading Sleight of Brand! Read past issues and subscribe here, and reply to this email to tell me what else you’d like to see in upcoming issues! 💌