“AI can do what you do, right?*” That’s been the question we’ve been getting as writers and content designers — whether literally or implicitly — since LLMs started joining the conversation.

In some ways it’s a good question. AI has gone from laughably quaint to actually pretty good in record time. Here at Slack, even the most talented writers and designers are using it to draft and refine UI copy, help center articles, and language systems. At first glance, it’s almost indistinguishable from what we’d write (especially for those of us who love an em dash).

But as any word nerd will tell you, “almost” can make the difference between meeting user and business goals, and seeing user trust plummet after years of careful investment. AI products can be an incredibly valuable tool in our toolbox for moving fast and solving problems in novel ways, but only when expertly applied.

It all comes down to trust — and content designers are uniquely equipped to solve AI’s trust and communication problems.

🛠️ ✏️ Content designers are the trust architects of AI products

As AI takes on more of the work of generating language, the question is no longer whether machines can sound human enough. It’s whether those conversations earn trust, accurately set expectations, and reflect our human values. That gap is exactly where content design lives. With our deep expertise in applied linguistics, humanities, language systems, and behavioral economics, we are equipped to be the connective tissue between technical capability and the human experience.

With all this in mind, we sat down as a Slack content design team to outline exactly how we envision our discipline and our day-to-day changing as we both use and design AI tools.

☀️🌻 Our core tenets

These are the guiding principles that we apply to our work.

• Trust is the product — a single broken trust moment can undo everything Slack is trying to accomplish.
• Human judgment is irreplaceable — content designers must provide the ethical accountability and contextual reasoning that keeps AI behavior honest.
• Language is infrastructure — AI agents use words and conversations to function, so language expertise is necessary to prevent AI from failing.

🖊️ 💻 How our work has changed

Like most creators in tech, we’re increasingly deploying AI to handle UI copy for user flows, system audits, and documentation. This has empowered us (and our cross-functional partners) to move faster when creating content, but it also means fewer strings that we lovingly handcraft (and therefore, more mistakes).

The buck stops with us when it comes to craft and quality. Just like human-written and -designed UX, it goes to crit; it gets reviewed. Everyone benefits from extra eyes — even LLMs.

We’re working to protect the quality bar for Slack, both as users and producers of AI experiences by:

• ✍️ Embedding our values and Slackiness into the tools that our product, design, and engineering teams use to build features for Slack
• 🏆 Aiming to set the industry standard for accurate, ethical, and human-centered outputs in our native AI offerings like Slackbot

What does that look like in practice?

• We developed evaluative heuristics, methods, and tools to catch trust failures before they ship.
• We measure, improve, and take responsibility for AI output quality and consistency standards.
• We created agent identity and language patterns that are usable by humans and LLMs alike.
• We educate users about how to use AI in Slack in the flow of work, both in the product and in our help documentation.
• We continue to actively collaborate across every part of the product development lifecycle, from ideation to creation to iteration.

But what does that actually look like? We’ll tell you!

👻 Handling “Hallulus”: Building trust when AI hallucinates

Shortly after we launched, our clients flagged that Slackbot appeared to be “leaking” private information. The good news: there weren’t any leaks — just hallucinated information.

The bad news is, even the perception of a leak does a number on user trust. Together with design and engineering, we released a quick labeling fix to address concerns. But we knew that the UX treatment needed to be even clearer, and that this “fix” was really an opportunity to build even deeper trust with users.

As a next step, we partnered with our UX research team to put alternative designs in front of users. They loved our approach. The new design direction we championed:

• Proactively tries to get the correct information automatically instead of asking the user to do more work.
• Avoids misleading labels and obscure redactions.
• Accounts for more types of hallucinations.
• Gives users a way to click through and learn more about why LLMs hallucinate.
• Makes Slackbot feel more transparent.

It’s a project that continues to change along with the technology, but it demonstrates how transparency is a trust-builder — and a value driver.

🤖 Skills to pay the bills: Building skills for AI consistency

In the early days when people were using AI to code new features for Slack, their UX copy was missing the mark on our Slack style guide (we get it, not everyone can capture our charm and wit). We needed a way for agents like Slackbot to follow the same instructions and guidelines every time. So here’s how we created a skill that reviewed content for voice, tone, usage, and terminology.

1. Define and document
   An AI skill is only as good as its instructions, so we started with our Slack style guide and usage guidelines. This rigorous pre-work to define and document what it means to be “Slacky” was a necessary first step.
2. Build a skill
   We converted our resources into formats that work for AI tools, because we wanted the process and output to be consistent every time. Slackbot makes building skills easy, but we also converted our usage guides to markdown files that coding agents like Claude and Cursor can use.
3. Evaluate with a human lens
   We tested the output for accuracy and usability (a content designer’s bread and butter). When we found errors, we clarified the skill’s inputs. We refined the output to be easy to understand. We “ate our own dogfood” and used the skill to edit our own content. After all, everyone needs an editor.
4. Let it loose!
   Once we were confident that the skill was working, we shared it with everyone at Slack. Our teammates immediately started using it to write error messages, button text, onboarding flows, you name it! This resulted in better first drafts of UI copy, huge time savings, and a valuable tool for reviewing content against our Slack style.

But the work isn’t done. We continue to tinker with these skills to make sure the output is consistently high quality.

💅 Don’t be rude: Aligning on the etiquette of working together with AI

Beyond how we communicate with users, we also find ourselves thinking about how we communicate with each other as we build products. AI is changing how we communicate at work. Not just what we produce, but how we relate. Think about:

• 🙂 The tone-softened message to smooth things over with a colleague.
• 🙏 The thank-you that sounds warm but generic.
• ✅ The performance feedback that’s accurate, yet clearly not written by a human.

More and more, AI is becoming part of these moments. Sometimes they feel fine, but sometimes, they feel off (or even icky). Most of us haven’t had a chance to actually talk about it.

At a recent Slack Design onsite, we ran an AI Communication Etiquette Yuck or Yum activity: 10 real scenarios, small group debates, and a lot of strong opinions. We synthesized what came out of it into a spectrum showing where we landed as a team on each scenario and six shared values we think apply well beyond our team:

• Be aware of the power dynamic. Use AI in ways that don’t exacerbate existing power disparities.
• Disclose AI use and give credit. Be transparent; purposeful obfuscation of AI use is a problem.
• Intent matters. Consider the impact on both you and the recipient.
• Keep things high quality. Content should meet human standards and respect people’s time.
• Protect emotional impact. AI can support sincerity, but should never replace it.
• Promote equity. AI should be a tool that improves equity in the workplace, not undermines it.

Everyone’s relationship with AI is still evolving. At Slack, we’re united by a commitment to authenticity, good intent, and ownership.

✈️ 🧭 Where do we go from here?

Like everyone else, we’re figuring this all out as we go. But we believe we are about to enter a golden age of content design. The questions around quality, trust, and judgment are only growing. The language experts who seize the opportunity will have the chance to shape the future. Will we embrace our humanity and our ethics, or will we bow down to our robot overlords?

*No, AI can’t do what we do… yet.