The short version: the fastest way to get case studies out of a small business is to stop asking clients to write them and start mining the emails, messages and calls you already have. I feed old client threads into AI, pull out the specifics, and turn a project I did eight months ago into a published case study in about ninety minutes. The AI doesn't write the proof, it just stops you losing it in your inbox.
Why most small businesses never publish a case study
I ask business owners about case studies constantly and I get the same answer every time. "I keep meaning to." Then I ask when they last asked a client to sit down for one and the honest answer is usually never, or once, eighteen months ago, and the client never replied.
Here's the bit nobody says out loud: clients hate being asked for case studies. It feels like homework to them. You're asking a busy person to carve out thirty minutes, remember numbers from months ago, and write flattering things about you unprompted. Most won't. Not because they're not happy with your work, but because it's effort with zero benefit to them.
So the case study sits on the to-do list for a year, and meanwhile you're sending prospects a website with three testimonials from 2022 and nothing that shows actual results.
The client I did this for in January
I'll call her Sarah because she'd rather I did. She runs a small bookkeeping practice, four staff, based near Manchester. We'd worked together for about five months on getting her visible on LinkedIn and cleaning up her inbound lead process. Good results, nothing spectacular, and definitely nothing she'd volunteered to write up for me.
What I did have was 63 emails between us, two Zoom call transcripts from Otter, and a WhatsApp thread where she'd told me, unprompted, that she'd closed a £4,200 retainer client directly from a LinkedIn post I'd helped her write. That single line, buried in a message about something else entirely, was the whole case study.
I pulled all of that into one document, dropped it into Claude, and asked it to pull out every concrete number, date, and outcome mentioned across the thread. Fifteen minutes later I had a rough timeline: what her lead flow looked like in August, what it looked like in December, and three direct quotes she'd typed to me herself, in her own words, that read far better than anything she'd have written if I'd asked her to sit and compose a testimonial from scratch.
The exact process, step by step
- Step 1: Pull every thread with that client into one file. Emails, WhatsApp exports, call transcripts, Slack messages, whatever exists. Don't tidy it, just collect it. This took me about ten minutes for Sarah, mostly copy and paste from Gmail.
- Step 2: Ask AI to extract facts only. My exact prompt was something close to: "Read this thread and list every number, date, result, or direct quote the client gave about outcomes. Do not summarise, do not add anything, quote them exactly." This matters more than anything else in the whole process.
- Step 3: Build a timeline, not a story yet. Before? After? What changed and when? I ask for this in a simple table: date, what happened, what the client said about it.
- Step 4: Check every single number against the source. Every one. I go back to the original email and confirm the £4,200 figure appears, in that exact form, from Sarah, not paraphrased by the AI into something rounder or more impressive.
- Step 5: Write the narrative around the facts, not the other way round. I use AI for a first draft of the connecting sentences, the bits that link one real quote to the next, but every claim in the finished piece has to trace back to something the client said or a number that happened.
- Step 6: Send it back to the client for one yes or no. Not "can you write me a case study" but "here's a draft built from our own conversations, can I publish this." Sarah said yes within four minutes. That's the whole ask now, not thirty minutes of her time, one minute of her attention.
- Step 7: Turn it into more than one format. A 600 word page, three LinkedIn posts pulling out single lines, and one slide for my proposal deck. Same source material, three uses.
Start to finish, that's roughly ninety minutes once you've done it a few times. I've now done this for eleven clients since the start of 2026 and the slowest one still took under three hours, because the client's emails were a mess and I had to dig.
Where AI helps and where I don't trust it
AI is brilliant at reading 60 emails in ten seconds and pulling out the one line that matters. It's brilliant at drafting connective sentences between facts. It is not brilliant, and should never be trusted, to invent or round up a number because the paragraph reads better that way.
Here's the uncomfortable part most people writing about AI and case studies won't say plainly: it is extremely easy to ask an AI model to "make this case study more impressive" and have it quietly turn a 15% increase into "nearly doubled" or invent a percentage that was never in the source material at all. I've caught it doing this on drafts for my own clients, more than once. If a prospective client, or worse a journalist, ever checks a number against the original source and it doesn't match, you don't just lose that case study, you lose the trust of everyone who ever reads anything you publish again. So the checking step isn't optional admin. It's the difference between proof and marketing fiction.
This is the same discipline the biggest brands use, they just have whole teams doing it. When you look at how a company like Airbnb built its brand around real host and guest stories rather than invented ones, or how Away built a following on specific, checkable customer experiences, the pattern is the same one I'm describing for a five-person bookkeeping firm. Specific, verifiable, slightly unpolished beats smooth and vague every time.
The proof problem nobody wants to talk about
Most of what businesses call a "case study" on their website is a testimonial with a title on it. A testimonial is "Lilach was brilliant to work with." A case study is "before working together, Sarah's inbound leads were three a month from cold outreach, after five months of consistent LinkedIn posting they were eleven a month with two closing at £4,200 and £2,800 respectively." One is an opinion. The other is checkable, and checkable is what moves a prospect from maybe to yes.
This ties into something I keep coming back to from Alex Hormozi's work on offers, which is that proof beats persuasion every single time, and most small businesses spend their energy on the persuasion and skip the proof because proof requires digging through old emails, which is boring. I've written before about the business lessons I've taken from Alex Hormozi, and this is the one I use weekly: don't tell them you're good, show them the number.
There's also a patience element to this that's easy to skip past. Sarah's case study only existed because I keep every client thread rather than deleting them after a project ends, a habit closer to the long game thinking in Napoleon Hill's work than anything to do with AI at all. The tool didn't create the raw material, it just meant I stopped losing an afternoon to finding it. I've written more on that patience-over-speed mindset in my piece on business lessons from Napoleon Hill, and it applies directly here: the client work you did quietly six months ago is worth more to you today than the client work you're chasing right now, if you bother to go back and find it.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
A simple template you can copy today
Once the facts are pulled and checked, I write to this structure every time:
- The starting point, in numbers if you have them: leads per month, hours spent, revenue, response rate.
- What changed, three or four concrete actions, not "we improved their marketing."
- The result, again in numbers, plus one direct quote pulled from the real conversation.
- One sentence in the client's own words about what surprised them. This is almost always the most quoted line when the case study gets shared.
If you want to go a step further, turn the numbers into something a visitor can play with rather than just read. I've written about how interactive calculators keep visitors on a page and convert better than static text, and a simple before/after slider or ROI calculator built from your real client numbers does the same job a case study does, just with the reader entering their own figures instead of reading someone else's.
What this costs you in time
Ninety minutes to three hours per case study, once the emails are collected. Compare that to the honest alternative, which is asking a client to write one from scratch, waiting six weeks for a reply that never comes, chasing twice, feeling awkward about chasing a third time, and giving up. I know which one produces more published case studies, because I've done both and only one of them has an eleven-out-of-eleven success rate this year.
The bit that surprised me most doing this repeatedly is how much better the quotes sound when they were never meant to be a testimonial. Sarah's line about the £4,200 retainer wasn't written to impress anyone, she was just telling me a bit of news in passing. That's exactly why it works. Nobody writes marketing copy by accident, but people say true things by accident constantly, in emails you already have sitting in your inbox right now.
Free resource: The Case Study Writing Template.
Related reading: Turning Old Client Emails Into Case Studies With AI (My Exact Process).
Frequently asked questions
Can I really build a case study without a new interview with the client?
Yes, in most cases you can, provided you already have several months of email, message or call history with them. Old threads almost always contain the numbers and outcomes you need, you just have to go looking rather than starting from a blank page and asking the client to remember everything themselves.
How do I stop AI exaggerating the results in a case study?
Ask it to extract facts only, never to summarise or improve the wording of a number, and then check every figure against the original source before you publish anything. Treat any number the AI has rounded, softened, or made sound more impressive than the source text as wrong until you've confirmed it word for word.
What if the client won't give me permission to publish their name?
Anonymise it. Use "a bookkeeping firm in the North West" or "a client I'll call Sarah" and keep every number exact. Readers trust specific, checkable numbers from an anonymous client far more than vague, glowing praise from a named one.
How many case studies does a small business need?
Three to five strong, specific ones outperform twenty vague testimonials. Focus the process above on your three best client relationships first, get those published, and only then move on to the rest of your client list.
Related reading: Using AI to Chase Unpaid Invoices Without Sounding Like a Debt Collector and AI Tool Costs Explained for Business Owners in 2026.
