We stripped the word "AI" from our marketing. Here is why.

Last week I went through 17 files in our codebase and removed the word "AI" from 180 places. Product names. Page titles. Meta descriptions. Schema markup. Status pills in the customer dashboard. The system prompts that the model itself sees.

We sell a Claude-powered reply system for service businesses. The thing literally is AI. We just stopped calling it that.

This is the story of why, what changed, and what I think the lesson is. If you sell something AI-powered and your conversions feel sticky, the rest of this post might be worth your time.

The moment

A clinic owner in the East Midlands, on a 15-minute scope call I did a few weeks ago. She runs a single-site aesthetics practice, does her own front desk between treatments, gets most of her enquiries through Instagram DMs after 7pm. Exactly the customer this thing is built for.

About 8 minutes in, after I had walked her through what the product actually does, she leaned forward and asked the question that has been rattling in my head since:

The question itself was the answer. The word "AI" had set off an alarm in her head. Not because she objected to the technology, she did not even know what was under the hood. She objected to the framing. The word made her picture a chatbot that would embarrass her in front of a patient asking about lip filler.

I said: "It is software that reads the DM and writes a reply in your voice, with your real prices. Patients cannot tell."

She booked Founding tier the next day. Lite is £49 a month, full system from £1,500 for context.

That conversation happened maybe four times across different calls before I clocked the pattern. UK clinic owners over 40, doing their own admin, do not get excited about AI. They get nervous about it.

The market context

2026 is a year of AI fatigue, depending on the audience. Two trends are running at the same time, and most product-marketing copy is on the wrong side of one of them.

Trend one: developers, ML engineers, agency owners, early adopters, consumer-app users (especially ChatGPT users) are more excited about AI than ever. For these audiences "AI-powered" is still a feature, not a tax. The novelty has dulled but the energy is real.

Trend two: small-business owners outside tech, particularly in service industries, have had two years of bad ChatGPT-wrapper products pitched at them. They have seen badly-built chatbots embarrass clinics. They have heard horror stories about AI quoting prices that did not exist. They have an immune response.

If your buyer is in trend one, lead with AI. The shorthand works for you. If your buyer is in trend two, the same shorthand works against you. The same product needs different framing.

Our buyer is firmly trend two. The marketing was on the wrong side.

What we actually changed

Concrete diffs, because abstract case studies are worth nothing.

Product names.

Hero copy.

Dashboard UI strings (the customer sees these every day).

The system prompts themselves. The LLM used to be told "you are an AI assistant for [Business Name]." Now it is told "you are an assistant for [Business Name]." Same instruction set otherwise. The model no longer refers to itself as AI in its replies. Which means the customer, even if they pasted the AI's reply into ChatGPT to interrogate it, would not see the word.

180 occurrences total. 17 files. About 90 minutes of focused work plus the time to write this post.

What we kept

This is the part I want to be careful about, because the post would read as anti-AI if I left it implicit.

The product is exactly the same. Same Claude Sonnet model. Same Anthropic API. Same prompt-engineering decisions. Same conversation logging, same dashboard, same Cal.com integration, same suppression list. Not one line of model-facing code changed.

We still link to Anthropic on the technology page. We still mention that the underlying model is from a major LLM provider. Customers who want to know what is under the hood can find out in about 30 seconds. We just do not lead with it.

This is not anti-AI. This is anti-AI-as-the-headline.

The lesson, written cleanly

This is the same lesson copywriters have been teaching for 80 years. Don't sell the drill, sell the hole. AI is the drill. Replied-DMs-while-you-sleep, more bookings, fewer-missed-enquiries-at-9pm-on-a-Tuesday is the hole.

The buyer does not care about the implementation. The buyer cares about whether her clinic still has a booking when she wakes up.

If your buyer is excited about AI (selling to developers, ML engineers, agency owners, ChatGPT power-users), leading with AI is the right call. The shorthand works for you.

If your buyer is anxious about AI (most service businesses, most non-tech industries, particularly anyone who has ever had a bad chatbot interaction), the same shorthand becomes a tax. You are paying for the word "AI" with your conversion rate.

The cheap experiment to run on your own site

Try this for 20 minutes:

1. Open your homepage in a private browser.
2. Count the number of times the word "AI" appears.
3. Count the number of times a clear customer outcome appears (more bookings, less admin, faster replies, recovered revenue, fewer no-shows, whatever your product's outcome is).
4. If the ratio is wrong (more AI than outcome), you might be paying the AI tax. Try inverting it for two weeks and see what happens.

You do not have to commit to anything. You can roll it back in an hour. The cost of the experiment is one batch of HTML diffs.

Honest caveats

A few things this post is not.

It is not a controlled A/B test. I did not split traffic 50/50 and run a power-calculated test. I just rewrote the site over a weekend and watched the qualitative inbound shift. If you want to be rigorous, run an actual experiment. The framework above is enough to be worth testing, not enough to be a final answer.

It is not universal advice. There are markets and audiences where leading with AI is exactly right. If you sell developer tools, infrastructure for AI engineering teams, or any product where the buyer wants to know "is this real AI", drop "AI" from your marketing at your peril.

It is not anti-Claude or anti-Anthropic. We are still Claude users. We still pay Anthropic's API bill every month. Removing "AI" from the marketing has nothing to do with how much we like the underlying technology, which is a lot.

One concrete recommendation if you ship this week

Rename your product. Or do not rename it, but rewrite the headline above the fold. Take the words "AI", "AI-powered", "powered by AI" out of the H1 and the meta-description. Replace them with the strongest outcome you can credibly promise.

If your homepage right now leads with "AI-powered reply system for clinics", try "Reply system for clinics. Books appointments while you sleep." Both describe the same thing. Only one tells the buyer what they actually want to know.

Ship it. Watch what happens. Roll it back if you don't like it.