Show AI-assisted work in your portfolio without breaking client confidentiality
Short answer: You can show proof of AI-assisted work without showing client files. Ask for written permission first. If the client says no, publish an anonymous case study or a stand-in project built from your own material. In both cases, say how you used AI and how you checked the result.
Why this is hard
Clients hire assistants to handle private work: inboxes, calendars, customer lists, and draft documents. Most of that work is not yours to show. A new client still wants proof. AI adds one more question: “Did you check what the tool wrote?”
A freelancermap guide on NDA-protected work gives a clear default. Show the work only with written permission from the client, after the NDA ends, or when the NDA does not cover it. Otherwise, do not share. See the article by Natalia Campana. The guide is general advice, not legal advice. For a specific contract, ask a qualified lawyer.
Choose your route
| Situation | What you can show | Risk |
|---|---|---|
| Client gave written permission | The work that the message names. Show the client’s review draft first. | Low, if you stay inside the written limits. |
| No permission, or the client said no | An anonymous case study: the problem, your process, the result. No client name, exact figures, or internal terms. | Medium. Small details can still identify a client. |
| No client work you can use | A stand-in project with fake data that you build yourself. | Low. Label it as a sample. |
The guide also warns that a private link or a live walkthrough is still disclosure under most NDAs. Do not treat “private” as “allowed”.
Example (hypothetical)
This example is not a real client story. An assistant manages a coach’s inbox. She cannot share any email. She builds a stand-in project instead:
- She writes 10 fake client emails with made-up names.
- She uses an AI tool to draft replies and a meeting summary.
- She marks every draft change she made by hand, such as a wrong date or a tone that was too strong.
- She adds a one-page note: the prompt she used, her checklist, and what she would not send to an AI tool.
A client can read this in five minutes. It shows judgment and care. It exposes no private data.
Steps: build one proof page
Start here
- Pick one task. Choose a task that clients ask for often, such as inbox triage or meeting notes.
- Write the problem in one sentence. Example: “Meeting notes took too long and missed action items.”
- Describe your process. Say what the AI tool did and what you did by hand.
- State your checks. List what you verify: names, dates, numbers, and tone.
- Give a result you can support. Use a real count from your own records, such as “3 errors caught before sending”. If you have no count, describe the result in words and do not invent a number.
- Add a data line. Write what data you did and did not put into an AI tool.
Advanced extension
Ask a past client for a short, approved testimonial that names the quality of your work and not the private details. Agree on the exact words in writing. Keep a simple log of permissions: client, what you may show, where, and until when.
Limits and trade-offs
- Anonymous is not always safe. A niche industry or a unique result can identify a client. When in doubt, ask.
- Public posts carry the most risk. The guide notes this for sites such as LinkedIn or GitHub.
- Stand-in projects are weaker proof. They show method, not results with a real client.
- Disclosing AI use. The NIST Generative AI Profile (July 2024) suggests that organizations consider telling users when content is AI-generated. It is voluntary guidance. Check your client agreement for its own rules.
- Philippine context. Many readers work with clients in other countries. Contract terms and privacy rules can differ by country. Ask the client what applies.
Common questions
What if my contract says nothing?
Silence is not permission. Ask in writing before you show any client work. Keep the reply with your records.
Can I show the prompt I used?
Yes, if the prompt holds no client data. Replace names, figures, and internal terms with placeholders. A prompt with a real client name inside it is still client information.
Be careful. A screenshot can show file names, tabs, email addresses, and other private details. Rebuild the screen with fake data instead. The freelancermap guide notes that blurring a logo is rarely enough.
How many proof pages do I need?
Two or three strong pages are enough. Choose different task types, such as inbox work, research, and scheduling. Each page should take a reader about five minutes.
Do this week
Write the one-sentence problem and the checks list for one task. Then send one short message to a current client: “May I describe this type of work, without your name, on my profile?” Save the reply.
Sources
- Can I share NDA protected work on my portfolio? Tips and advice. freelancermap, Natalia Campana, published July 31, 2024, updated October 7, 2026. Practitioner guide, not legal advice.
- Artificial Intelligence Risk Management Framework: Generative AI Profile (NIST AI 600-1). National Institute of Standards and Technology, July 2024. Voluntary guidance.
