How to Use AI to Write Client Website Copy Without It Sounding Like AI

---

There is a specific kind of AI (ChatGPT, Claude etc) copy that makes me want to close the tab immediately.

You know the one. “In today’s fast-paced world, having a strong online presence is more important than ever.” Or the headline that could belong to literally any business on the planet. Or the sentence that uses three parallel phrases separated by em dashes where one clear sentence would have done the job.

I’ve written that copy. By accident. More than once.

And every time it happened, it was because I gave AI nothing to work with and expected something good to come out. That is not how this works.

AI writes usable website copy when you give it the client’s actual language, discovery call notes, testimonials, their own emails, instead of asking it to write from a blank prompt. Google does not penalize AI-assisted copy. It penalizes copy that says nothing, which is exactly what AI produces when it has nothing specific to work from.

That’s the whole thesis of this post. So let’s get into, how to use AI to write client website copy without it sounding like AI.

Can AI Write Website Copy That Doesn’t Sound Like AI?

Yes. But the input is the entire game.

Most designers sit down with a blank prompt and type something like “write homepage copy for a wedding photographer in Charleston.” And AI does exactly what you asked. It writes generic homepage copy for a generic wedding photographer in a generic location.

It’s not wrong. It’s just not anyone’s.

The copy that doesn’t sound like AI is the copy that sounds like the specific client you are working with. It uses her words. It reflects her clients’ specific fears and desires. It references the kinds of weddings she actually shoots. It sounds like something she would say in a conversation, not something a language model would produce for any photographer anywhere.

The only way to get that is to give AI her actual language before you ask it to write anything.

Does AI-Written Copy Hurt SEO?

No. And I want to be really direct about this because there is a lot of anxiety in the design community around this question that isn’t based on how Google actually works.

Google does not penalize AI-assisted copy. It penalizes thin, unhelpful, generic content that doesn’t serve the reader. The reason so much AI copy feels like it hurts SEO is because so much AI copy is thin, unhelpful, and generic. That’s not an AI problem. That’s an input problem.

Copy that is specific, useful, and written for a real person in a real situation performs well in search regardless of how it was drafted. Copy that hedges everything, says nothing concrete, and could belong to any business in any industry performs poorly. The medium is not the issue. The content is.

What does get flagged and penalized is copy that is clearly produced at scale with no human review, stuffed with keywords in ways that don’t serve the reader, or scraped and republished without meaningful transformation. A designer who uses AI to generate a first draft and then thoroughly edits it to sound like her client is doing something Google has no problem with.

For the technical SEO setup around copy once it’s on the site, RankMath is what we recommend. It walks you through optimization in a way that’s actually approachable, and it pairs well with whatever copy process you’re using.

What Should I Feed AI Before Asking It to Write Anything?

This is the part most people skip and it’s why most AI copy sounds like AI copy.

Before I ask AI to write a single section of a client’s website, I feed it everything I have.

Discovery call notes. The specific things she said about her clients. The language she used to describe her own work. The problem she keeps seeing that she knows how to solve. The things she’s proud of that she doesn’t say loudly enough. I paste all of it.

Testimonials. Her clients’ words are the most valuable raw material in the entire project. When a past client says “she made the whole process feel effortless and I never once felt like just another job,” that phrase is doing enormous work. It tells you how to position the experience. It gives you language that is specific and real. Feed it directly to AI and ask it to incorporate the sentiment and the phrasing into the copy.

The client’s own emails. How someone writes in an email is often very close to how they actually talk. The sentence structure, the words they reach for, the things they emphasize. A few emails from your client to her own clients are worth more than an hour of prompting.

Competitor pages. Not to copy them, but to show AI what the category sounds like so it can help you differentiate. “Here are three competitor websites. Here is what my client does differently. Write copy that reflects the difference.”

A prompt I actually use: “Here are my discovery call notes, three client testimonials, and two emails this designer wrote to past clients. Using her language and specific details, write homepage copy that speaks directly to [her ideal client], leads with [the main outcome she creates], and sounds like a real person rather than a template.”

The difference in output between a blank prompt and that prompt is genuinely significant.

What Are the Tells That Copy Was AI-Written?

I want to name these specifically because knowing what to look for makes editing so much faster.

Hedging everywhere. “May help,” “can often,” “in many cases.” AI hedges because it doesn’t want to be wrong. Real copy makes confident claims.

Tricolon everything. Three parallel things separated by em dashes or commas, everywhere, in every sentence. It’s a rhythmic tic that AI reaches for constantly. One clear thing stated well is almost always better than three things stated in parallel.

“In today’s fast-paced world.” Or any variation. This phrase and its cousins are AI’s way of providing context it doesn’t actually have. Delete on sight.

Headlines that could belong to any business. “Thoughtful design for the modern creative.” “Where strategy meets story.” If you could put a competitor’s logo on it and it would still make sense, it needs to be more specific.

Em-dash pileups. AI loves an em dash. Personally so do I, but when every other sentence has one, it starts to read as machine-generated rather than human-written. Break them up. Use periods. Use commas. Vary the rhythm.

No specifics anywhere. No locations, no client types, no particular outcomes, no named anything. Generic copy is almost always AI copy that wasn’t given enough to work from.

How Do I Edit AI Copy So It Sounds Like the Client?

Here is the editing pass I run, in order.

Read it out loud first. Anything that sounds wrong will reveal itself immediately. If you stumble, the reader will stumble. Fix it.

Replace every hedge with a confident claim. “May help you attract better-fit clients” becomes “attracts better-fit clients.” If the claim isn’t true enough to state confidently, it shouldn’t be in the copy at all.

Add one specific detail per section. A location. A type of client. A named outcome. An actual number if you have one. One specific detail does more work than three paragraphs of general language.

Read it as the ideal client, not as the designer. Does this sound like something written for me, specifically? Does it describe my situation accurately enough that I feel seen? If not, what’s missing?

Give it back to the client for a voice pass. Send her the draft with this note: “This is a starting point built from what you shared in our discovery call. Read through it and anywhere it doesn’t sound like you, change it to your words. Add anything I missed. The goal is for this to sound exactly like you.”

That pass is not optional. It’s the step that turns AI-assisted copy into copy that is actually hers. And it usually takes her less than an hour because she has something concrete to react to rather than a blank page to fill.

Should the Client Know I Used AI?

Yes. And here is how I frame it.

“Part of my process now includes using AI to generate a first draft of your copy based on everything you shared in our discovery call. This means instead of starting from nothing, you’ll receive a full draft that already sounds like you and reflects your specific clients and outcomes. Your job is to edit it until it’s exactly right. This approach gets you to a finished website faster without sacrificing the voice and specificity that make your site actually convert.”

That framing positions AI as a speed tool, not a substitution for your thinking. Because that’s what it is. You are still the one who ran the discovery call, identified the positioning, decided what to lead with, and will edit the final draft until it’s right. AI wrote a first draft. You built the strategy around it.

What About Copy That Needs to Get Cited by AI Search?

This is the newer frontier and it’s worth a mention here because it’s changing how I think about copy structure.

AI search tools like ChatGPT, Perplexity, and Google’s AI Overviews tend to pull from pages that answer questions directly, near the top, in plain declarative sentences. Copy that hedges, buries the main point, or structures everything as flowing prose without clear section breaks gets passed over. Copy that answers a specific question in the first two sentences of a section gets cited.

This means the copy we write for client websites should increasingly be structured with clear questions and direct answers, especially on pages that are meant to rank. It’s a small shift in how you organize a page and it makes a real difference in whether that page gets surfaced in AI-powered results.

Our Showit SEO Course covers this in detail, including how to structure pages for both traditional search and AI search in a way that works for both simultaneously. And DesignEdit covers the evolving AI search landscape weekly, which is genuinely the best way to stay current on something that is moving this fast.

FAQs