How to Use AI in Your Design Workflow Without Sending Proprietary Design Data to a Third-Party Model

How to use AI in your design workflow without sending proprietary design data to a third-party model comes down to three decisions: what you put in, which tool it goes into, and what your company has already agreed to. Most designers skip all three and paste first. That is how an unreleased flow ends up in a chat window nobody approved.

You do not have to stop using AI. You need a short rule you can apply in ten seconds.

Start by sorting what you would paste

Not everything you work on carries the same risk. Sort it into three buckets before you open any tool.

Safe to share. Public material and your own thinking. Published competitor screens, your own rough notes, generic questions like "how do I structure a research plan for a checkout redesign". Nothing here identifies a client, a product or an unreleased feature.

Share only after you strip it. Real work that becomes safe once the identifying detail is gone. Interview notes with names and companies removed, a flow described in plain words instead of screenshots, a metric turned into "a double-digit drop". If you would be fine seeing the text on a slide in a public talk, it is fine here.

Keep out entirely. Unreleased product screens, roadmap detail, customer data, contract terms, anything under NDA, and any file your security team has told you not to move. No amount of cleanup makes these safe to paste into a tool you do not control.

Check what the tool does with what you send

The same model can be used in very different ways, and the difference matters more than the model name.

• Consumer accounts and business accounts often have different terms. Some plans let the provider use your inputs to improve their models unless you opt out. Business or enterprise plans commonly commit not to. Read the current terms for the plan you are on, because they change.
• Look for a retention setting. Some tools let you turn off chat history or limit how long inputs are stored.
• Ask whether your company already has an approved tool. If it does, use that one. If it does not, ask before you pick your own. A two-line message to your manager or security contact is enough.
• Consider a tool that runs locally. Models you run on your own machine keep the input on your device. They are usually less capable, so use them for the sensitive-but-simple jobs, such as summarising your own notes.

Turn the rule into a habit

A short checklist beats a long policy you will never reopen.

1. Name the job first. "Synthesise these six interviews" is a different risk from "critique this unreleased screen".
2. Pick the bucket. Safe, strip first, or keep out.
3. Strip before you paste. Replace names with roles, products with categories, numbers with ranges.
4. Use the approved tool. If there is not one, ask.
5. Keep a note of what you shared. One line in your project notes is enough if anyone asks later.

What about screenshots and design files?

Images are the easiest way to leak something by accident, because a screenshot carries far more than the part you meant to show. Crop to the one element you want feedback on, remove real names and data from the frame, and treat anything that shows an unreleased feature as keep out. If you want help with layout or hierarchy, describing the screen in words often works as well as uploading it.

The point of all this

The goal is not to avoid AI. It is to be the designer a team can trust with it. Teams that see you handle sensitive material carefully tend to give you more room to experiment, not less. Pair that with the habit of judging what the tool gives back, and AI becomes a normal part of how you work rather than a risk you keep having to explain.