How to Invite AI In While Preserving the Craft of Product Writing

By Candice Pires

The dominant conversation about AI and writing is focused on output: volume of copy, created faster. But working at the intersection of writing and design, a focus of mine has been instead on how we can use AI to scale a writer’s thinking within the product team, without losing the fundamentals of our craft. Those being: 

• Good UX copy requires empathy for the user and respect for their time.
• Good brand copy requires broad contextual knowledge.
• Consistency builds trust.
• Writing—and more importantly editing—informs thinking.

Recently I’ve been using an AI in-the-flow copywriter I built using Claude. It’s governed by clear editorial direction and content strategy and is helping not just the writing team, but all designers on the team to move beyond early-stage placeholder copy. The benefit? In a fast-paced environment where prototypes shift between one standup and the next, a writer’s thinking shows up in the work earlier—and stays there. 

Here’s what else inviting AI into the process has taught me:

Don’t define direction for AI tools too quickly

Writers, like strategists, can be navigators for product teams. We use our research skills to get to know the user, the brand, and the direction of the product. Often these things are in evolution precisely at the time that we are concepting and it can be dangerous to define them too early. 

Because writing with AI tools requires us to have a point of view, we need to protect time to test assumptions and make sense of everything coming in, from a brand brook to user testing results. In early-stage work, ambiguity is information. It tells you where the real design challenge lives. Through sketches and conversation, teams develop an understanding of the product’s nuances and tension points. Rush to define everything for the AI, and you skip the messy discovery that makes a product resonate.

The tool guides, the writer decides

When I shared the prompt with designers I was clear that it was to ‘guide generation of good copy.’ Not give final copy to paste into design. I encouraged them to make changes, and I remained the final editor. I provided it as a project-level prompt, but also as a markdown file for anyone working directly in code. I also let them know that I would be continually updating it.

For both the team and myself, the tool was a support but not a key originator of ideas. On a recent luxury fashion project, on the contrary, some of our best brand-led copy came from in-person visits to a retail store and listening in to conversations between shoppers and assistants.

Ask for feedback often

As with all new workflows, adoption was mixed. Some designers used the tool, others didn’t. The ones that did had mixed reviews about the copy outcomes (as did I). One told me that my job was not at risk, but kept using it. This is proof of how designers actually work: they’re not looking for finished copy, they’re looking for something good enough to design against. The gap between what the tool generates and what a writer would have written is precisely where craft lives—and if designers can feel that gap even while finding the tool useful, they understand the writer’s role even more concretely.

Be curious. Ask designers what doesn’t feel right to them. What else they might want from the prompt. Which leads me to…

Be prepared to update your thinking

A master prompt should never be static. In design, we move fast. We’re always iterating, and learning from clients and users. So it’s important to update the prompt with that thinking. And this is the beauty of a collaborative AI tool; you own the prompt and can tinker with it without disruption.

Structure clearly, get better outputs 

The process of formulating an AI copywriter prompt was not so different to formulating writing guidelines for a client writing team. As with all prompt writing, a clear structure yields best outputs. To kick things off, I took the editorial direction and content strategy I developed, translated it into a master prompt to govern what the LLM generates, and gave designers access. Then, via Claude Enterprise, I set up a ‘Project’—a space where I could invite designers to use Claude under the instruction of my master prompt.

Here’s a sample structure to get started with: 

1. Project context – Describe the role you want your AI copywriter to take on.
2. Writing functions – Identify the types of writing you want the LLM to do. For an e-comm product, it could be conversion, creative, and conversational copy. For healthcare, maybe educational and instructional copy.
3. Tone – A fully defined tone should remain a deliverable toward the end of a project, but evolving it as we go helps hone thinking.
4. Examples of good – Each time you identify copy that resonates, add it to your master prompt. Provide references of ‘More like x’ and ‘Less like y’ to help clarify what is ‘good.’
5. Input – Describe what people will be asking for and how to treat requests. Such as ‘always override character counts.’
6. Output – I like to ask for a range of options. This could be degrees of brand expression, from none (so very functional copy) to really pushing brand voice. State how you want answers formatted.
7. Rules – Words to avoid, words to use, overriding principles of good UX writing, brand guardrails.

What’s next

The new workflow is helping to better share writing direction and content strategy across the team, and the approach is being extended to new sectors, like telecom. New ways to incorporate AI tools —from identifying GEO content gaps to smoother CMS authoring— are continually emerging.

Each time, it becomes even clearer that the quality of these machine-assisted outputs is a direct reflection of the quality of the thinking that human writers and content strategists put in. Our focus is keeping that bar high by ensuring it’s our team’s human-led product writing fundamentals and judgment that dictate our AI collaboration.

About the author: 

Based in London, Candice Pires is a Writing Director at Work & Co. She enjoys making things that people connect to and that improve how we live, through building brand voice and centering audience needs. She has shipped impactful work for clients including LOEWE, Decathlon, and Pandora and is an established journalist for international media like The Guardian.