The UX Mistake Everyone's Making with CustomGPTs

Raise your hand if this sounds familiar: Your company is all-in on AI. Everyone’s spinning up CustomGPTs. They’re everywhere. For everything.

Need a job description written? There’s a GPT for that. Want to draft a strategy memo? Someone made one last week. Wondering if someone’s made a GPT that helps you find all the other GPTs? Probably.

To be clear: I’m not mad about it. I think it’s great that people are learning and experimenting. I actively encourage my team to create them. But let’s be real: just because you built a GPT doesn’t mean anyone’s going to use it.

This is not Field of Dreams. If you build it, they probably won’t come. (IFKYK)

And if that sounds familiar — if you’ve noticed that nobody’s using the CustomGPT you poured your heart into — you’re not alone.

But why does this happen?

UI ≠ UX (Especially in AI)

I’m sure your CustomGPT is solid. It probably answers a very specific slice of questions extremely well. But the truth is: if people don’t know it exists, or don’t know what to expect from it, it doesn’t matter how good it is.

This is classic UI vs UX territory. A nice start screen and a clever name might check the “interface” box. But UX is about behavior, and behavior is hard to change. Especially in a world where most of us are just trying to get through the day without context switching ten times before lunch.

This Isn’t New (But We Keep Forgetting)

I’ve been doing this long enough to see this pattern repeat. Every time a big new tech shift comes along (cloud, mobile, now GenAI) product teams forget what we know about human behavior. We get swept up in the newness and we start building for what’s possible, instead of what’s usable.

Here’s the good news: while technology evolves fast, people’s brains don’t.That’s our edge. That’s where good UX comes in.

So I pulled out a framework I created back when I was teaching Human-Computer Interaction at the university level. It’s called Barriers to Enterprise Tool Adoption and it maps out the common reasons people don’t use tools that could genuinely help them.

Barriers to Enterprise Tool Adoption

There are two major hurdles when rolling out any new tool internally. I call them:

🌊 The Gulf of Enticement

“I guess I’ll give it a try.”

This is the space between awareness and initial use. If people don’t know your GPT exists, or if they’re not convinced it’s worth the effort to try, they’ll never get to that first click.

🌊 The Gulf of Habit

“I’m not going to switch unless it’s way better than what I already have.”

This is the much tougher gap. Once people are in their flow — copying text from the last doc, checking with a peer, or doing it manually — it’s really hard to get them to use something new. You have to make the experience so good that it’s worth the behavior change.

In the full framework (image below), I break down seven key dimensions: Awareness, Access, Performance, Ease of Use, Insight Effectiveness, Information Quality, and Value Realization. If even one of these breaks down, adoption can stall:

A framework I created for identifying barriers to enterprise tool adoption

So You Made a CustomGPT. Here's How to Make It Stick.

✅ 1. Make sure it’s solid

Kick the tires. Hard. Ensure it gives reliable, accurate answers — without hallucinations — and that it covers the kinds of questions people are actually likely to ask. (Yes, you can see usage metrics for CustomGPTs if you're an Enterprise admin.)

📝 2. Explain what it does and what it doesn’t

Use the CustomGPT start screen wisely. Don’t just slap a name and move on. Write a short, helpful description, and include 2–3 great example prompts. (This functionality is also available in the Enterprise tier: you can customize the GPT's name, instructions and example prompts via the GPT builder interface.)

📢 3. Get the word out

Internal marketing matters. Consider:

• A dedicated Slack channel for GPT launches

• A shared list of all org-approved GPTs

• Easy instructions for how to access and share them

Pro Tip: Many people still don’t know how to find or launch a CustomGPT, even within orgs that are heavy OpenAI users.

🔗 4. Insert access into the user’s actual workflow

This is the big one: Don’t make people stop what they’re doing to go look for your GPT. Find ways to surface it at the moment they need it, inside the tools they already use.

Here'e an example: Let’s say you created a CustomGPT that writes job descriptions to comply with company rules and compliance boilerplate. Amazing! But ask yourself this: When and where do people actually write job descriptions?

Probably not by opening ChatGPT cold and typing “Write me a job description for a Senior Product Designer.” More likely, they’re:

• Copy-pasting an old JD from Google Drive

• Slapping together bullet points in a Word doc

• Asking a teammate for theirs on Slack

So instead of just saying “Hey, we made a GPT that writes job descriptions!” and dropping a link in a channel that gets buried by Thursday, embed access where people already do the work. 

Here's what I mean, leveraging the Job Description example above:

👉 Add a line of helpful microcopy or a “Need help?” link right at the top of your company’s job description template in Google Docs or Confluence. “Not sure where to start? Try our CustomGPT for job descriptions, trained on our best postings.”

Small nudge. Right place. Right time. Big difference.

This reduces cognitive load and makes using your GPT the path of least resistance, not a side quest.

One Last Thought (Hi OpenAI, if you’re reading)

Wouldn’t it be amazing if EnterpriseGPT had a way to auto-surface CustomGPTs to users based on topic? Imagine launching ChatGPT and seeing:

“Hey! I’ve been trained to answer questions about your company’s Design System, thanks to a CustomGPT created by Alex in UX.”

Just saying. It would save a lot of internal coordination and make discovery much easier. (You’re welcome Sam Altman.)

TL;DR

CustomGPTs are powerful. But if you want people to use them, you have to treat them like products:

• Build with a clear need in mind

• Design for usability and reliability

• Promote them like they matter

• Integrate them into the flow of work

Because even in the age of AI, if you build it, they still might not come.