Why I Stopped Letting Clients Send Me ChatGPT Copy for Webflow Sites in 2026

What changes when a client sends me 4,000 words of ChatGPT copy on a Sunday night?

Last month, a Series A SaaS founder in Bengaluru sent me a Google Doc at 11 PM on a Sunday. The subject line read "all the copy you need for the new Webflow site". Inside were 4,217 words of polished, structured, and completely empty marketing prose. I read three pages, screenshot a paragraph, and pasted it into Originality.ai. The score came back 98% AI generated. I knew before I tested.

This is the second time in six weeks I have received what I now call "ChatGPT inventory". The copy reads well. Every sentence is grammatical. But the value propositions are interchangeable with any of the founder's twenty competitors. According to Originality.ai's January 2026 dataset, AI generated content now makes up 41% of all newly indexed pages, up from 17% in 2024. My clients are part of that trend, and it is starting to hurt the sites I build.

I have changed how I handle this. I no longer accept AI generated copy as the source of truth for a Webflow build. Here is what I do instead, why I do it, and how I explain the decision to clients without breaking the relationship.

Why Does AI Generated Copy Quietly Kill Webflow Site Conversion in 2026?

AI copy fails because it averages. ChatGPT, Claude, and Gemini learned from millions of marketing pages, so they regress to the mean of what already exists. A site built on average copy sells the average product to the average buyer at the average rate, which in B2B SaaS is around 2.3% according to Unbounce's 2026 conversion benchmark report.

I have seen the pattern in my own logs. A Webflow site I shipped in February 2026 used AI copy the founder wrote. After 90 days, the lead form conversion rate sat at 0.8%. We rewrote the hero, the three feature blocks, and the FAQ in the founder's voice using a real customer interview transcript. Conversion rose to 3.1% in the next 60 days. Nothing else on the page changed.

The issue is not the words themselves. It is the absence of a position. AI copy summarizes consensus. It cannot tell a buyer why this product, by this team, on this Tuesday. Buyers feel that gap even if they cannot name it.

How Can I Tell If a Client Wrote It Themselves or Used ChatGPT?

I look for five signals in the first paragraph. Heavy use of phrases like "in today's fast-paced world", "leverage", "robust solutions", and "streamline" usually flags ChatGPT output. Heavy use of dashes is another tell. So is a paragraph that sounds correct but contains zero proper nouns, no customer names, no specific metrics, no real product feature.

For verification I use Originality.ai and GPTZero, both of which posted 96% plus accuracy on their February 2026 benchmark for medium length English samples. I do not treat detector scores as proof. I treat them as a prompt to read the copy more carefully. The clearest signal is still my own ear after twelve years of writing for clients.

What Happens to AEO and AI Citations When the Source Is AI?

AI search engines like ChatGPT Search, Perplexity, and Google AI Mode cite sources that say something original. Princeton's GEO research from 2024 found that pages with verifiable statistics and named experts earn approximately 132% more citations than pages built on generic marketing prose. AI generated copy contains neither, so it loses on the metric that matters most in 2026.

I have watched two recent client sites struggle here. Both shipped with AI written category pages. Six months later, neither showed up in ChatGPT or Perplexity for their target queries. The pages were technically fine, indexed by Google, and even ranked decently on classic search. But the AI search systems treated them as noise. My guide on tracking AI prompts instead of keyword volume walks through how I caught this in client reports.

But What About Clients Who Use AI Well in Their Workflow?

I am not against AI in the writing process. I use Claude Opus 4.7 daily for outlines, research summaries, and counter-arguments. The line I now draw is at the output. Raw ChatGPT prose pasted into a Google Doc and sent to me is not copy. It is a starting point that someone forgot to start with.

When a client uses AI to interview themselves, transcribe sales calls, or generate twenty headline options to choose from, that is fine. The output reflects their voice and their specifics. When the AI is the author and the client is the editor of one pass, I push back. The difference shows up in the final site every time.

How Do I Have This Conversation Without Insulting the Client?

I never say "this is AI slop". I say "this reads like it came from your category, not from your company". Then I ask three questions. Who is the one customer you wrote this for? What is the one outcome that customer cared about most? What did they actually say to you in the last sales call?

Nine times out of ten, the client lights up. The AI copy was a placeholder they sent because they felt deadline pressure. They have the real story. They just had not written it down. I take notes on the call and use the transcript as the source for the rewrite.

What Do I Ask Clients to Send Instead of ChatGPT Copy?

I send a short brief I now call the "Five Real Things" form. It asks for five inputs: the most recent customer interview transcript, three written objections from sales calls, the founder's own one paragraph answer to "why now", the three competitors clients usually mention, and one specific metric the product moves. This is data, not prose. I can write from it.

This came out of my work with a fintech client in Mumbai who kept sending polished but generic copy. I asked for the five inputs and got a 38 minute Loom of the founder reading a sales call transcript out loud. The hero copy I wrote from that Loom now drives a 4.1% conversion rate, up from the 1.2% the AI draft would have produced based on my historical baseline.

How Do I Price the Rewrite When Clients Push Back on Time?

I include one copy review pass in every Webflow project at no extra cost. If the brief comes back as AI inventory, that pass becomes a forty five minute interview call instead. I then write the hero, three feature blocks, and the FAQ in the founder's voice. That writing block adds twelve hours to the project, which I bill at my standard hourly rate.

Clients almost never refuse. They refuse the original 4,000 word AI draft when they see the rewrite. The conversion lift pays for the writing time inside the first month of the new site being live, which I have now measured across six 2026 projects. I documented this pricing logic in more detail in my piece on flat monthly retainer pricing lessons.

How Do I Set This Boundary With My Next Webflow Client This Week?

Start with the proposal. Add one sentence under the scope section that says "all copy used on the live site must be reviewed against AI detection and rewritten if flagged". Then send the "Five Real Things" brief as part of your kickoff packet so the client never assumes ChatGPT copy is the deliverable. Finally, in the kickoff call, narrate why you do this, using a one minute version of the conversion story above.

For the foundation under this approach, my post on running AI content detectors on client copy walks through the exact tools and thresholds I use. For the positioning conversation that follows, my guide on building case studies that actually convert covers the customer voice work the rewrite depends on.

If you have a Webflow project where the brief feels like it came out of a model and not out of your business, I am happy to walk you through the rewrite process I use. Let's chat.