Ground rules
To help you use AI tools in design safely, we expanded on the ground rules we have as a company. Please read them carefully:
Enter only public data into services such as ChatGPT, nothing that an NDA protects:
- Don’t enter any client-specific information (e.g., brand or service names).
- Don’t enter anything that can be considered intellectual property, either ours or the client’s.
- Don’t enter a unique selling proposition or feature of the service.
- Don’t enter raw research data. Make sure you are not sharing any data that is under GDPR.
- Don’t enter characteristics of a target group or some customer insights that are unique to our client.
- Don’t enter workshop inputs/findings or anything that was part of some other activity we did with the client.
- If you are working for a client that already has strict data protection rules, consult with your PM first.
It is mandatory to fact-check all results generated by AI tools:
- ChatGPT is learning from the text data provided by users and the online results it was trained on, so it is prone to misinformation and outdated data.
This is not meant to discourage you from using AI. On the contrary, the goal is to keep up with technological progress and explore different ways of adding AI to our design process. You can use the mock data if it helps make the output more actionable (e.g. using a mock brand name).
Data protection
Before using a new AI tool, check its data policy (this is a good principle for all new tools).
ChatGPT
ChatGPT uses inputs from its users for model training. Because of that, it is possible that someone can access the information you put into it. There’s a mandatory step on your end to use ChatGPT as safely as possible. To protect your data, please make a Privacy Request. OpenAI will stop using your input for model training once you make this request.
You can also turn off Chat History & Training in your account settings, but this will prevent you from seeing your past conversations in the sidebar.
Check the latest information on ChatGPT data controls here.
Notion AI
When using NotionAI, your data is protected under Notion's standard data protection practices. Notion AI will not use your data to train models. Check the latest information on NotionAI here.
Prompt writing tips
Ask to assume the role of an expert:
Act as a UX/UI designer / Product Strategist / UX writer and [insert prompt].
Provide the context and some examples (if possible, without breaking the NDA):
The project is a [brief project description]. Its primary goals are to [list goals]. The target audience includes [target demographic].
Give a tone of voice and length requirement:
Write a one-sentence interaction description for [specific interaction] in active voice, using a clear and formal tone.
Ask for answers to be in a specific format, like a table or bullet points:
Generate a UX best practice checklist of a [specific feature] on a [desktop] in a [table format]. Include a column with sources.
Ask for more than one result and then compare (3–5 results are usually enough):
We recently came up with a [user problem]. Suggest 3 possible solutions and list the pros and cons for each.
Ask for prompt formulas:
Suggest 5 prompt formulas that a UX/UI designer could use when working on a [e.g. new feature request].
Always double-check the information an AI tool provides. Don’t accept the first result; use the regenerate response function, refine your initial prompt, and save your successful prompts for future reference!
Don’t put AI-generated outputs into the files that clients can access unless you’ve reviewed (and amended) them. Don’t mislead clients into thinking that some output was yours if you used some AI tool to generate it.
Analysing a brief or feature request
Here are some prompts you can use in your next task:
Write a concise problem statement based on the text below: [insert text]
What's the best way to design a solution to this brief as a UX/UI designer: [insert text]
As a UX/UI designer, what questions could I ask a product owner / developer / stakeholders to get better insight into design requirements for [feature], based on the text below: [insert text]
Create action points for UX/UI designer based on this feature request: [insert text]
Define key [business] metrics for a [feature]