A freelancer can use AI to write a case study by feeding it organised project notes and checked numbers, asking for a draft in a fixed structure, and verifying every claim against evidence before the client sees it. Send the draft with a short approval email, publish only what the client has approved in writing, and keep that email on file.
The free generator pages that dominate search results hand you a template or a paragraph from a few form fields. None of them can tell you whether you may publish, whether your numbers are true, or how to get a busy client to reply. Those are the freelancer's real problems. The example below follows a freelance web designer who rebuilt a childcare nursery's enquiry form and wants to publish the result. All figures are illustrative.
Ask for portfolio rights at kick-off, not at the end
The cheapest approval is the one you ask for before the work starts, when the client is pleased to have hired you. Check what your contract says about portfolio use and confidentiality. If it is silent, add a short clause to the proposal or the kick-off email, and have a lawyer look at the wording you adopt for your own contracts. Here is a plain starting point.
I would like to describe this project in my portfolio once it is finished. Before anything is published I will send you the draft, with every figure and quotation marked, and nothing goes live without your written approval. You can ask me to remove your name, logo or any figure, or to decline entirely.
The clause does two jobs. It tells the client that approval is theirs, which makes a yes more likely. And it gives you something to point at later if the conversation drifts. For a nursery, add one more line of your own: no photographs of children, and no detail that could identify a child or a family, unless the nursery confirms in writing that it holds the parents' consent for that use. If you want a way to spot awkward clauses before you sign, checking a client contract for red flags with AI covers portfolio and confidentiality terms.
The inputs sheet: notes and numbers before any prompt
An assistant writes a better case study from a tidy sheet than from your memory of the project, and you will check the draft faster against it. Keep one sheet per project from day one and add to it as you work. At the end it takes ten minutes to finish, not an afternoon to reconstruct. If your raw material is interview recordings rather than project notes, writing case studies from customer interviews with AI covers that route.
| Field | What goes in it | For the nursery (illustrative) |
|---|---|---|
| Client in one line | Size and type, no names yet | A 55-place childcare nursery |
| The problem | In the client's own words, with the date of the email or call | "Parents start our form and give up." (email, 3 April) |
| What you did | Three to five bullets, past tense, no adjectives | Cut the form from 14 fields to 5; moved the show-round line above the button; rewrote the page in plain language |
| Scope and dates | Start, finish, what was not included | Five weeks, April to May; no new photography |
| Results | Each number with its source, window and baseline | Form enquiries: 36 (June to August last year), 78 (June to August this year); form completion 41% before, 68% after |
| Quotes | Verbatim from emails or calls, with the date | One line from the owner's email of 14 July |
| Other factors | Anything else that could explain the results | A flyer drop in July |
| Approvals needed | Name, logo, each number, each quote, any images | All five |
The "other factors" row is the one freelancers skip and the one that protects them. Nursery enquiries tend to be seasonal, bunching before a new intake, so a comparison with the previous spring could flatter the redesign. The designer compares the same three months a year apart, and records the flyer drop so that it is acknowledged rather than discovered by a sceptical prospect.
Before you paste anything into an assistant, check the setting. Put client material only into a plan where training is off: a business plan, which for ChatGPT Business and Claude Team does not train on your content by default, or a personal account where you have switched the training setting off. Both of those business plans need two seats, so a solo freelancer is usually on a personal plan. Leave the client's name out until approval arrives and write "the nursery" in the sheet.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowGetting numbers from systems you do not control
Most freelancers cannot see the client's analytics, form tool or booking system, so the results column depends on a request that is easy to forget. Ask at kick-off for a read-only export or a dated report at the end, and put a reminder in your own diary for roughly 60 to 90 days after launch, when there is enough data to compare. The request is small if you make it specific.
Could you send me a CSV export of form submissions by date for June to August this year and last year? Counts only, please: no names or contact details. A screenshot of the form tool's completion report for the four weeks before and after 1 June would help too.
Keep the file you receive, unaltered, with its date in the file name, and re-add the totals yourself before you quote them: a figure you computed from the export is one you can defend, and one you copied from an email is not. If the client cannot or will not share data, ask them to confirm each number in writing instead, and say in the ledger that the figure is client-reported. A client-reported figure is still publishable if the client approves it, but the case study should say where it came from.
From sheet to first draft in a fixed structure
Give the assistant the whole sheet and a structure that leaves little room for padding. A fixed structure also makes your case studies comparable with each other, which helps a prospect reading three of them.
Write a case study of 450 to 600 words from the notes below, using only facts in the notes.
Structure: a headline with the main result in it; a snapshot box (client type, what I did, timeline); Challenge (80 to 120 words); What I did (3 to 4 short paragraphs); Results (state each figure with its window and baseline, and mention any other factor); Client quote (use only the quotations I have supplied, verbatim; if none, write [quote pending]).
Rules: no adjectives of degree (dramatic, huge, soared, transformed). Do not add tools, numbers, names or quotes that are not in the notes. If something needed is missing, list it under MISSING at the end instead of guessing.
NOTES: [paste the inputs sheet]The line about missing information is the most useful one. Without it an assistant fills gaps with plausible details, and a plausible detail in a case study is a false claim with your name on it. With it, the draft ends in a short list of questions you can answer from your records.
What the first draft got wrong, and the edited version
Expect the first draft to be tidy and a little too pleased with itself. On the nursery job it made four mistakes, and each is typical.
- Inflated language. "Enquiries soared by over 100%." The arithmetic was fine, since 36 to 78 is a 117% rise, but "soared" is a claim about feeling, and the rise had a second cause.
- Borrowed credit. "Thanks to the new form, the nursery now receives more than double the enquiries." The flyer drop in July had vanished from the draft.
- A tool nobody used. It said the form was "built with a leading form platform" when the nursery's existing form tool had been kept.
- A quote nobody said. Despite the instruction, a closing line about parents "loving how easy it is" appeared. Quotes are the one place where an assistant's smoothness is dangerous, so every quotation gets traced to the email it came from.
The fix is simple but needs a person. Here is the results paragraph before and after editing.
| Assistant's draft | Freelancer's edit |
|---|---|
| Thanks to the new form, enquiries soared by over 100%, and parents love how easy it is. | Between June and August the form brought 78 enquiries, against 36 in the same months a year earlier. Completion rose from 41% to 68%, measured from the form tool's reports for the four weeks before and after launch. The nursery also ran a flyer drop in July, so this page does not credit the form with the whole rise. |
The edited version is longer and less exciting, and it is the one a careful prospect can check for themselves. It states the window, the baseline and the source, and it concedes the second cause, so a sceptical reader has nothing to catch you out on.
The claims ledger: every number with its source and caveat
Before the client sees the draft, list every factual claim in it and attach evidence. This is the step that separates a case study from a brochure, and it is also what makes the client's approval fast, because they can check a table in two minutes instead of reading prose for hidden problems.
| Claim in the draft | Evidence | Window | Caveat | Client OK |
|---|---|---|---|---|
| Form enquiries rose from 36 to 78 | Form tool export of submissions (CSV, saved) | June to August, a year apart | Flyer drop in July; no sole credit claimed | Pending |
| Completion rose from 41% to 68% | Form tool report: views and submissions | Four weeks before and after launch | Counts only people who reached the form page | Pending |
| Form cut from 14 fields to 5 | Before and after screenshots | At launch | None | Pending |
| Five-week project | Project log and invoice dates | April to May | None | Pending |
| 55-place nursery | Nursery's own wording on its site | Current | Client confirms the figure | Pending |
| Owner's quotation | Her email of 14 July, copied exactly | Not applicable | Client approves the wording | Pending |
Four rules keep the ledger honest. Every number points to a file you can open, not to memory. Every comparison names its window and baseline. Anything that could explain the result besides your work is written down. And ratings, awards or inspection outcomes that the client mentions are quoted exactly as the issuer worded them, with the date, and the client confirms them, because you cannot check those yourself.
Then use the assistant as a hostile reader of its own draft, in a fresh chat with only the draft and the sheet.
Act as a sceptical editor. List every factual claim in this case study, one per line. For each, mark it SUPPORTED (quote the line in the notes that supports it), UNSUPPORTED (nothing in the notes) or INFERENCE (a conclusion the notes do not state). Flag any quotation not found word for word in the notes. Do not rewrite anything.On the nursery draft the editor pass returned 14 claims: 10 supported, 2 inferences ("parents prefer the shorter form" and "the redesign caused the increase") and 2 unsupported (the tool name and the closing quotation). It caught the tool name, the closing quotation and the borrowed credit that reading had found, plus one inference that reading had missed. The fourth error, inflated language, is a style fault rather than a claim, so only a person catches it. If the pass finds more than you did, trust it and look again. If it finds fewer, do not trust it, because a model reviewing its own work can be too generous. The ledger and your own reading are what count, and the assistant is a second pair of eyes. If you are wondering whether to tell clients you use AI in this process, should freelancers tell clients they use AI? sets out the case for saying so.
The approval email, and what to do when the answer is slow
Make the client's job as small as possible. Send one email with the draft and the ledger attached, three questions that can be answered with a word each, and a date. An assistant can tidy your wording, but write the three questions yourself because they decide what you can publish.
Subject: Your case study draft: three quick questions by [date]
Hi [first name],
Thanks again for a good project. As mentioned when we started, I'd like to describe it in my portfolio. The draft is attached (one page), together with a one-page list showing where each figure comes from, so you can check it quickly.
Could you reply with a word or two on each of these?
1. Name: may I use the nursery's name and logo, or would you prefer it anonymised ("a 55-place nursery")?
2. Figures: are you happy with the numbers as written? Mark any you would rather round or leave out.
3. Quotation: is the wording of the quote at the end right, or would you like to change it?
I won't publish anything until you have replied. If it's easier, call me for ten minutes. Many thanks,
[your name]Note the sentence "I won't publish anything until you have replied". It removes pressure, and it keeps you honest if the reply never comes. Send a reminder on day five that restates the three questions, and a second on day ten that offers the anonymised version. Silence is not approval, however polite your first email was. If nothing arrives after the second reminder, keep the draft private and move on.
When the reply comes, it will usually be one of three outcomes. Approved as written: save the email thread as your record and freeze that version, noting the date. Approved with changes: apply them exactly, resend the changed version, and wait for a fresh yes to the final text. Declined or anonymised: publish the anonymised version only if the client agreed in writing, with figures rounded or generalised as they asked. In all three cases the approved text is the only text you reuse, and any later edit, even a typo in a number, goes back to the client. For wording on testimonials in particular, where the legal line sits when AI helps write testimonials is worth a read before you ask a client to put their name to a quotation.
The version the nursery approved
This is the final text after edits and approval, shown as an illustration. The figures are invented for this example, and the quotation is left as a placeholder because a real one has to come from the client's own words.
A shorter enquiry form for a 55-place nursery: 36 form enquiries became 78
Snapshot: a 55-place childcare nursery. Form redesign and page rewrite. Five weeks, April to May.
Challenge: the nursery's enquiry form had 14 fields and asked for a child's date of birth before a parent's name. Only 41% of the people who opened it sent it.
What I did: I cut the form to five fields (parent's name, email or phone, the child's age band, the preferred start month and one open question), moved the line about show-round availability above the button, and rewrote the page in plain language using the nursery's own words from parents' emails.
Results: between June and August the form brought 78 enquiries, against 36 in the same months a year earlier. Completion rose from 41% to 68%, measured from the form tool's reports for the four weeks either side of launch. The nursery also ran a flyer drop in July, so this page does not credit the form with the whole rise.
[Quotation from the owner's email of 14 July, approved word for word]
It runs to about 180 words, well under the 400 to 700 you might publish, because this is the compact version for a proposal. The published page adds a paragraph on how the five fields were chosen and a screenshot of the old and new forms, with no child or parent visible.
Publishing, repurposing and keeping it true
Put the date on the page and a "figures last checked" line. Diary a reminder for twelve months ahead to re-check the numbers or take the page down, and remove it promptly if the client withdraws consent. Link it from your proposals and the relevant service pages: a case study that sits on a portfolio page nobody visits does nothing for you.
Repurpose only from the approved text. An assistant is good at this, because the facts are already settled and the job is reshaping.
From the approved case study below, write: (1) a LinkedIn post of 120 words with a one-line hook and no hashtags; (2) a two-sentence proposal paragraph for a prospect with a similar problem; (3) a three-bullet summary for a one-page PDF.
Use only the facts and figures in the approved text. Keep the sentence that mentions the flyer drop. Do not add claims.The instruction to keep the flyer sentence matters: shortened versions are where caveats get dropped. Turning one project into a steady flow of posts is a bigger topic, covered in turning expertise into posts with AI. Whatever you repurpose, run the same check: every number in the short version appears in the approved long version.
The same routine for a copywriter, a bookkeeper and a photographer
The steps stay the same; the evidence and the risks change with the trade. Three variations show where.
A freelance copywriter usually has softer numbers: email open rates, click-throughs, a change in enquiries after a rewrite. The ledger should name the email platform's report and the send dates, and the caveat column matters more, because a list that grew or a seasonal peak can swamp the effect of better wording. A music school whose enrolment emails were rewritten might see replies rise from 9 to 17 a month, but if the school also changed its fees that term, the case study says so.
A freelance bookkeeper faces confidentiality first. Almost nothing about a client's finances should appear, even with permission, so the case study is anonymised from the start ("a four-person kitchen-fitting firm") and the numbers describe the process, not the money: month-end close down from nine working days to four, say, with the firm's agreement to that figure. Ask the client to approve the wording, and never paste ledgers or bank data into an assistant to make the draft richer.
A freelance photographer has proof that is also personal data. The images are the case study, so the ledger tracks releases: who is in each photograph, whether they agreed, and what use they agreed to. For a care agency's shoot, that means staff releases, and no service users in frame. The draft text can be written by an assistant, but the permissions record is yours.
The time it takes is modest once the habit exists. Allow about ten minutes to finish the inputs sheet, twenty to draft and edit, forty for the ledger and the sceptical-editor pass, ten for the email, and thirty to publish and repurpose: roughly two hours of your time, plus the wait for the client. Without an assistant, the drafting and repurposing alone would add a few hours. The software cost is an individual plan at around $20 a month (check the plan page, because list prices change), and the bigger cost to guard against is a published claim you cannot defend. A routine where permission comes first, numbers come from files and the client approves in writing is what makes the finished case study safe to use in a proposal.
Questions freelancers ask about AI case studies and approval
Can I publish a case study without approval if I leave the client's name out?
Not automatically. If the sector, size or numbers make the client recognisable, they can still object, and your contract's confidentiality clause may cover the work itself. Check the contract, and if it is silent, ask. An anonymised version that the client has agreed to is usually the best compromise between privacy and proof.
What if the client never replies?
Treat silence as a no. Send reminders on day five and day ten, offer a shorter anonymised version, then keep the draft private. Never publish on a 'no reply means yes' basis, even if your email said so, because the approval you can show later is the only one that protects you.
Which AI tool should a freelancer use for case studies?
Any major assistant can draft from structured notes. The choice that matters is the data setting: put client material only into a plan where training is off, either a business plan or a personal account with the training setting switched off. Keep one project folder per client instead of pasting from memory.
How long should a freelance case study be?
Between 400 and 700 words with a short snapshot box covers most freelance projects: enough for the challenge, what you did and the results with their context, and short enough to read in three minutes. Write a second, longer version only for large projects where a prospect will want the detail.
Further reads
- AI Automation for Freelancers: Lead Capture to Onboarding in One Flow — Build the portfolio-rights question into your onboarding flow.
- ChatGPT Prompts for Freelancers: Scope Creep, Chasers and Updates — Prompt wording for client emails, chasers and updates.
- Interior Design Portfolio: Write Case Studies From Project Photos With AI — A sector version that builds case studies from project photos.
- Can a Personal Trainer Edit Testimonials With AI? What Is Allowed — What is acceptable when editing a client's words.
- Using AI in Copywriting: Protect Each Client's Voice Step by Step — Protect each client's voice when you draft on their behalf.
- How Freelancers Use AI for Invoices, Proposals and Chasers — Use the same chaser habit for approvals and unpaid invoices.
- Sign Company Marketing: Case Studies and B2B Emails With AI — A job sheet, case-study prompt, consent email, local page test and three outreach emails a sign company can use this month.
- AI Marketing Tools for Small Businesses: A Shopfitter's Case Study Pages — A step-by-step method for shopfitters: project paperwork and photos in, a checked one-page case study out, with client sign-off before it goes live.
- Upwork AI Policy: Can Freelancers Use ChatGPT on Client Work? — What Upwork's terms actually say about generative AI, where client contracts override them, and disclosure wording to use before you start.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: the ChatGPT Business and Claude pricing pages for training defaults, seat minimums and individual plan prices, read in November 2026.