Photo by Quilia on Unsplash

Most messaging problems aren’t actually messaging problems

What product marketers get wrong when messaging doesn’t land, and what to check instead.

In my last post, I wrote about joining Box as its first dedicated AI Product Marketing Manager and spending my first few weeks on a listening tour instead of jumping straight into messaging. One conversation from that period ended up changing my approachaltogether, and it left me with a specific lesson I still use to check my own work today.

Where we started

Early on, our messaging for Box AI sounded a lot like most AI products at the time: ask questions of your content, get instant summaries. It was clean, it was true, and it made sense for the moment it was written in, when the job was proving Box AI could actually do the thing at all. But for some reason, it wasn’t landing.

One question kept coming up

The moment that made this obvious to me came out of a very specific, very technical question, one I started hearing over and over from enterprise customers, especially those in regulated industries or heavily regulated regions like EMEA: how does it actually work?

Now, let me give you some context: Box AI is model-agnostic, which is a great differentiator, but it’s also exactly why answering this question can get…complicated! What this means is that a Box user might use OpenAI models to ask questions of their content in Box Hubs, leverage Gemini Flash models to extract data from content and feed it into a workflow, and then use a custom-built agent running on Claude to do financial analysis, all inside the same platform. So the question “how does it work” didn’t have one answer. It depended on which model was doing which job! But starting off the answer with “it depends” doesn’t work.

What’s more is that this question came up often enough, and mattered enough, that I started to see it as a pattern rather than a one-off concern, and this question needed a clear specific answer before the conversation could move forward.

That’s when it really landed for me: it’s not that customers didn’t trust Box, rather we just hadn’t been answering the actual question they were asking.

So, what’s a PMM to do?!

Where we landed

Our messaging moved from describing what Box AI could do to explaining how it could be trusted to do it. Where we used to lead with “ask questions of your content, get summaries,” we now lead with something closer to this:

“Fueled by customizable agents, Box AI delivers instant insights, extracts context to power intelligent workflows, and adapts to your business needs, all within a secure, governed, and permissions-aware AI platform.”

What’s different with this new version is that it still talks about capability, agents, insights, workflows, but it doesn’t stop there. It closes on the thing enterprise buyers actually needed to hear: secure, governed, permissions-aware.

Another manifestation of this new messaging was the AI Architecture flow diagram I mentioned at the end of my last post. Once we realized customers needed to see exactly under the hood, step by step, we simply went ahead and showed them.

That same shift showed up in other places too. I wrote a longer explainer on the Box blog going deeper into how Secure RAG actually works, for the readers who wanted the full technical story rather than the summary version. And I sat down with our CTO, Ben Kus, for an AI Explainer episode, because some of this is just easier to hear explained out loud than to read off a slide. Different formats, same underlying shift: say less about what the product could do, and more about how it could be trusted to do it.

[embed]

Product marketing isn’t just about finding the right words. Sometimes it’s about figuring out which truth deserves to be told first.

What I’d do differently now

If you’re building go-to-market for an enterprise AI product, here’s the checklist this experience left me with:

• Ask what would make someone say no, not just what would make them say yes. Our buyers weren’t asking for more features. They were asking for the specific piece of information that would let them say yes to something they already wanted.
• Get specific enough to prove it. “Secure and governed” is a claim. “Here’s what’s going on under the hood” is proof. If your messaging can’t survive someone asking “how, exactly,” it’s not finished yet.
• Watch for a pattern hiding inside a global rollout. A single global messaging framework can quietly bury a concern that only shows up for certain buyers.
• When the answer is technical, a picture truly speaks a thousand words. A diagram answered a question that paragraphs of reassurance couldn’t. If you keep hearing the same technical question, that’s usually a sign you need an artifact, not another sentence.

None of this required a new product capability. It required figuring out what was actually standing between a customer and turning the thing on, and then answering exactly that.

This is Chapter 2 of Field Notes from an AI Product Marketer, a series of reflections on building enterprise AI go-to-market while the category was still taking shape. Next time, I’ll get into why so many AI customer stories fall apart under scrutiny, and it usually has nothing to do with whether the AI actually worked! Stay tuned!