I've spent the past year interviewing more than 60 team leads about how their teams actually use the AI inside the tools they already pay for. The pattern was hard to miss: most of those features go untouched. Not because they're badly built. Because nobody could tell the user why they were there.
This matters more now than it used to, because AI changed what "done" even means. The same feature can give one user a finished result and leave another staring at something half-useful — same model, same button, different outcome.
When the value is that uneven, the thing that closes the gap isn't the feature. It's your story. It's whether the user understands what this is for, why it's here, and what you're actually positioning it to do. When that story is clear, even the user who didn't get the full value the first time keeps an open mind. They stay with you — because they identify themselves in the problem you're trying to solve.
Without that story, every new AI button is just one more thing on the screen — it blurs what you are instead of sharpening it. And in a crowded market, sameness doesn't keep you in the race; it removes you from it.
One of the more useful books I've read on this is April Dunford's Obviously Awesome. Her five components of positioning only got more relevant once AI entered the picture — and answering all five should be a mandatory exercise before any AI feature ships. Here's what I run against my own roadmap.
- Competitive alternatives — what would your user reach for if this didn't exist? Rarely a competitor. Usually the manual workaround they trust, or just opening Claude or GPT in another tab. Name it. If your feature can't clearly beat that, stop here.
- Unique attributes — what do you have that the alternative doesn't? "It uses AI" doesn't count; the alternative uses AI too. The real attributes are what the model alone can't give: your data, your context, the workflow it lives inside, the action it takes where work happens. If that list is empty, you're shipping a worse ChatGPT with your logo on it.
- Value — turn each attribute into something the user feels. Native access to their data means an answer about their business, not a generic one. Living in the workflow means one less tab, one less copy-paste. Finish the sentence "which means you can…" for every attribute — if you can't, it doesn't belong in the pitch.
- Target — who, specifically, is this for? AI tempts you to say "everyone," because the model can help anyone. That's the trap. Name the one role who'd be upset if you removed this tomorrow, and build for them. What's unmistakably for someone beats what's built for all of them at once.
- Market category — what does the user think this is before they click? Call it "an AI chatbot" and it inherits every disappointed assumption about chatbots. Frame it by the job — "drafts your release notes from your commits" — and it gets measured against that job, where it can win.
Soon "powered by AI" will sound as dated as "now available on mobile" did a decade ago — first a selling point, then just the baseline. The products that survive this hype won't be the ones that shouted the what. They'll be the ones built on the real value.