The short version: One good client result, mined, gives you three weeks of content, not one social post. I use AI to do the reshaping, never the thinking, because the moment you let a model invent the insight instead of your own work, the content goes flat and readers can smell it. This only works if you’re honest about the numbers and ruthless about cutting anything that sounds like every other AI post on your feed.
The client win I’m still getting content from four months later
Back in September a commercial cleaning company I work with in Reading rebuilt their Google Business Profile and review reply process using AI drafting tools. Nothing fancy. They went from replying to about 1 in 5 reviews to replying to every single one within 48 hours, using AI to draft the first pass and a human to edit it before it went live. Three months later their local pack rankings moved from page two to the top three results for “commercial cleaning Reading” and enquiries from Google Business Profile went from roughly 4 a month to 15.
That is one result. One case study’s worth of material, on paper. Except I didn’t write one post about it. I got four blog posts, six LinkedIn posts, two email newsletters, a short video script, and one lead magnet out of it, spread across five weeks, and I’m still not finished mining it.
Here’s the bit most people writing about “AI content repurposing” skip: they treat every input as equally valuable, so they take one mediocre client update and try to inflate it into ten pieces of content, and it shows. Thin source material plus AI just gives you confident-sounding nothing. The Reading case worked because there was a real number in it. If your result is vague, “the client was really happy with the results,” you don’t have content, you have a testimonial, and no amount of prompting fixes that.
The actual process, step by step
This is the sequence I use, and I stick to it in this order because doing it out of order is exactly how you end up with AI slop.
- Step 1, write the raw facts yourself, no AI involved. Numbers, dates, what changed, what the client said in their own words. For Reading this took me 20 minutes with the client on a call. This document is the only thing that matters, and it’s the one part I never hand to AI.
- Step 2, ask AI to find the angles, not write the content. I feed the raw facts into Claude or ChatGPT and ask “give me eight different content angles from this, aimed at different audiences: business owners who ignore reviews, agencies who manage GBP for clients, people who think AI writing sounds fake.” This step alone usually gives me more angles than I’ll ever use.
- Step 3, pick the strongest three or four angles and draft each in a completely different format. Not the same post reworded four times. One becomes a blog post with the full number breakdown. One becomes a LinkedIn post about the uncomfortable bit (more on that below). One becomes a short “3 things I’d do differently” email to my list. One becomes a 60 second video script for a reel, similar to how brands studying an Instagram marketing strategy turn one proof point into a dozen different visual formats rather than repeating the same caption.
- Step 4, edit out anything that sounds like AI wrote it. This is the step people skip and it’s the one that decides whether the content works. I look for three tells specifically: sentences that start with “In today’s fast-paced world,” any use of the word “use” or “unlock,” and paragraphs where every sentence is roughly the same length. If I find any of those, I rewrite by hand.
- Step 5, turn the strongest number into something interactive. The 4-to-15 enquiries jump became the basis of a simple lead magnet, a short calculator that lets a visitor plug in their current review reply rate and see a rough estimate of what fixing it could be worth in enquiries. I’ve written before about why interactive calculators convert so much better than static content, and this is the exact reason: a number a reader can apply to their own business beats a case study they can only read about.
The uncomfortable part nobody tells you about repurposing
Most of your client’s story is not yours to publish. The Reading client was fine with the enquiry numbers going public, but wouldn’t let me name them, wouldn’t let me share the exact review text, and asked me to round the ranking position rather than state it precisely. That meant every single piece of content had to be built around anonymised or rounded numbers, and AI is bad at knowing when a number has been softened for a reason. Twice, when I asked ChatGPT to expand a draft, it “helpfully” invented a specific business name and a made-up review quote to fill a gap I’d deliberately left vague. If you’re not checking every draft against the original facts document from step one, you will publish something your client never agreed to, or worse, something that isn’t true.
The other uncomfortable truth is that this only works if you have wins worth mining. A lot of small business owners try to skip straight to step three, generating polished-sounding content from a client relationship that never produced a real number. You end up with the content marketing equivalent of Joe Pulizzi’s warning about brands chasing volume over substance, something I wrote about in more detail in the business lessons from Joe Pulizzi, where the point that stuck with me most is that consistency without a real story behind it just trains your audience to ignore you faster.
Why one result beats ten generic posts
I used to write two generic “5 tips for small business marketing” posts a week. They got maybe 40 views each and no replies. The month I switched to mining one real client result, my LinkedIn engagement on those posts tripled, and two people booked calls directly off the back of the review reply post because they recognised their own business in the 1-in-5 stat. Specificity is the whole game. Nobody shares “AI can help your business grow.” People share “this cleaning company went from 4 to 15 enquiries a month by replying to every Google review within 48 hours.”
This is the same reason brand strategy breakdowns work as content in the first place. When people study how GoPro built its marketing strategy around customer-generated footage, or how Figma grew through its community and templates, or how Mailchimp turned its own product usage into content, the pattern is always the same: real, specific, checkable proof, reshaped into different formats for different audiences. None of those brands wrote one case study and moved on. They mined it for years.
What this looks like on a normal week for me now
Every time I finish a piece of client work that produced a measurable change, I add one line to a running document: the number, the timeframe, the client’s own words if I have permission to use them. Once a month I sit down with that document and AI and do the five-step process above on whichever result is strongest. That’s it. It sounds almost too simple to be a system, but the system was never the AI part, it was building the habit of capturing the real number the moment it happens, because six weeks later nobody remembers whether it was 12 enquiries or 15.
The AI does the boring reshaping work, the angle-finding, the first drafts, the format conversion. I do the fact-checking, the voice, and the decision about what’s true enough to publish. Split it any other way round and the content stops working, no matter how good the model gets.
Frequently asked questions
How much client detail can I use without permission?
Assume nothing is fair game until you’ve asked. Even a happy client can be uncomfortable seeing specific numbers, review text, or their name in public. I always send the finished draft back to the client before publishing anything with their details in it, even loosely anonymised, because what feels fine to me as the writer isn’t always fine to them as the business owner.
Isn’t this just AI content spinning with extra steps?
No, and the difference is where the AI sits in the process. Spinning is asking AI to generate the substance. What I’m describing is asking AI to reshape substance you already gathered by hand from a real client conversation. If you skip step one, the raw facts document, you’re just spinning, and readers notice within a sentence or two.
How many pieces of content can one client result realistically produce?
From a strong result with a clear before-and-after number, I get between four and eight distinct pieces of content across formats, blog, LinkedIn, email, video script, and sometimes a calculator or lead magnet. Weaker results with vague outcomes might only stretch to one or two before you’re repeating yourself.
Do I need expensive AI tools to do this?
No. I use ChatGPT and Claude on their standard paid tiers, both around 20 US dollars a month, and that covers everything in this process. The cost isn’t the tool, it’s the time spent gathering real facts from clients in the first place, which is the step you cannot shortcut with a better subscription.
Related reading: What Your AI Meeting Notetaker Is Doing With Your Client Calls and Why Every AI-Written Newsletter Sounds the Same (And What I Changed In Mine).
This builds on my main AI marketing guide, my main guide on the topic.
