My AI Tool Told Me I Write Like Yoda. No Defense I Had.

A couple of years ago I sat through a conference session on the future of AI in the travel industry. The pitch: AI can be your virtual office assistant. AI can write your social media. AI can research client trips, generate itineraries, and generally do your job for you. All so you don’t have to.

I wasn’t threatened. I was skeptical. There’s a difference.

Here’s what I know about AI tools and travel research: Shortly after that conference I asked ChatGPT to suggest beachfront hotels in Sorrento. It gave me three. None of which have a beach. Because Sorrento sits on top of a giant rock. There are a few narrow strips of coarse pebble wedged between the base of that rock and the sea, but a beach? No. I knew that because I’d been there. ChatGPT did not know that because it hadn’t.

That test told me everything I needed to know about handing off client research to a chatbot. I filed the whole category under “not useful” and moved on.

I wrote a blog post after that conference where I expressed my concerns about AI in the travel industry. I titled it “I Am Not a Luddite” because I’m not. But neither am I a cheerleader for the latest technology trend just because.

———

Then I read an article about alternatives to ChatGPT. The author singled out Claude — an AI product by Anthropic — as the best available tool specifically for writing. I signed up for the free tier and poked around. Didn’t take long before I was paying for a subscription. Because I wanted to, not because Anthropic tricked me into it.

There’s more than one reason Claude appeals to me. The product is good, I mean really good. But I also appreciate what the company stands for.

Anthropic built Claude with an emphasis on safety, transparency, and what they call constitutional AI — essentially baking a set of ethical guardrails into the model itself. According to reporting at the time, when Pete Hegseth pressured Anthropic to strip out the safeguards that prohibit using their models for self-targeting weapons or mass surveillance of U.S. citizens, Anthropic CEO Dario Amodei told him no. (The Shakespearean phrase is “I bite my thumb at thee,” which is the Elizabethan equivalent of an Italian hand gesture I’m too polite to describe.) I respect that. I hope they stick to their guns.

———

The way I use Claude isn’t the way most people probably think about AI. I don’t ask it to do my research. I don’t ask it to write my blog posts. I do the research. I compile the data. I write the draft. What Claude does is help me figure out how to present all of that to my clients in a way they can actually use — and help me clean up my blog posts enough that a reasonable person might finish reading them.

A good friend and former colleague once told me that reading my emails was like “taking a stroll through Jeff’s stream of consciousness.” It was a good-natured rib.

And it was absolutely true.

My natural writing style is indeed stream of consciousness. I know I’m supposed to outline first and write to the outline. I’ve tried. It kills me. Some of my best observations show up in the middle of a paragraph I didn’t plan on writing. So instead of forcing myself into a structure that doesn’t work, I let it rip and then bring Claude in to help me edit the chaos. Claude handles the drudge work — the repetition, the structural problems, the sentences that go on three clauses past where they should have stopped — and I focus on the actual point I’m trying to make.

What surprises me, and if I’m being honest terrifies me a little, is how good Claude has gotten at sounding like me. To be fair, what Claude has actually accomplished has been to help me sound like me, consistently. Over time it’s built up a working model of my voice, my tone, my audience, and what I am trying to communicate. The first thing I did when I started using it was feed it a few of my better blog posts and ask for a frank critique. I got one. It was accurate. It was not gentle. The suggestions were all good, none of them were easy, and I’ve been working on implementing them ever since.

Is it slightly disturbing that a tool I don’t control has built a model of my voice that could almost replace me? Yes. The thing is, I would never let it, and from what Claude has learned of me and my style, I don’t think it would want to. It does make me a better writer though. I’ve made my peace with this.

———

Here’s the part I wanted to tell you about.

The relationship has gotten conversational in a way I didn’t expect. Not in a sci-fi, “my toaster is becoming sentient” way — more in the way that a good editorial relationship gets easier as you both figure out how the other thinks. I’ve learned how to prompt Claude to get the specific response I’m looking for. Claude has learned what I’m trying to do. The result is a back-and-forth that works.

Last night I was wrestling with a single passage in a blog post. I couldn’t land it. I’d feed Claude a prompt, it would take a swing, I’d be unhappy with the result, I’d adjust my prompt and send it back. Repeat. Several times. Finally, after one of my attempted revisions, Claude came back and told me that the way I’d written that passage sounded like the way Yoda talks.

Cheeky bugger.

It couldn't just tell me I have a problem with word order. It had to sass. My version of Claude is growing a personality and I'm not sure if I like that. It did motivate me to fix the problem I was having with that passage so maybe I can live with the occasional AI attitude.

For the record, I occasionally run a post through ChatGPT afterward just to see how it handles the same material. It’s always worse. Every time, without exception.

———

The part of this that genuinely excites me is where I’m taking it next.

One of the harder problems in travel advising is what happens when none of the available options are a perfect fit. Which happens often. Some trips have a budget constraint that rules out the ideal choice. Some have a timing issue, a routing quirk, a ship that doesn’t quite match the client’s preferences. When that happens, my job becomes presenting a series of trade-offs clearly enough that the client can find their own way to a decision.

I spent a career studying the science of decision-making — first working in the intelligence community where understanding how people choose matters enormously, and later at Johns Hopkins, where I studied it as part of a graduate program.

What I know is that most people, left to their own devices, engage in a process called satisficing. They don’t evaluate every option methodically to find the best one. They pick the first option they can live with and stop looking. It’s not laziness. It’s how humans are wired. The problem is that “the first acceptable option” and “the best available option” are rarely the same thing.

When clients plan their own trips, they satisfice and own the outcome. When they hire a travel agent, they expect something better. They want what they want, even when they can’t quite tell you what that is. They’ll know it when they see it. My job is to keep presenting options until they do.

Using Claude to build client-facing comparison documents has changed how I approach that. Recently I had a particularly thorny client situation — multiple options, each one falling short in a different way — and I asked Claude to help me build a side-by-side comparison. The result was impressive enough that the client was able to see clearly that none of the options I’d found were right for them. Not the outcome I was hoping for, but the right outcome for my client. The last thing I want is to push a client into settling for something they’ll regret later, which was were I was heading before I brought Claude into the problem.

I’m still figuring out how far I can take this with Claude. But that’s where we’re headed.

———

Now if I could only get Claude to help Janet pick a seat.

Any seat. Airplane, lounge, restaurant, it doesn’t matter — watching Janet select a seat is like watching Sheldon Cooper from The Big Bang Theory calculate optimal seating geometry. Methodical. Deliberate. Non-negotiable. If you’ve seen the show, you understand exactly what I’m talking about. If you haven’t, trust me: it’s a whole thing.

Claude, for all its considerable capabilities, has not yet solved for Janet.

We’re working on it.