How AI Is (and Isn’t) About to Transform UX Research & Product Discovery
“AI is replacing repetitive work, not the need for people doing customer research.”
That’s the headline I shared “ages ago” by AI timelines - shortly after Lenny Rachitsky’s popular thread on AI’s effects on Product Management ignited discussions everywhere.
My LinkedIn version of this post struck a chord, but it only skimmed the surface.
There was more to the story…
Quick recap: AI will not replace people doing researchers (yet)
Data bias & traceability remain unsolved.
Organizations still crave direct customer empathy.
Successful teams need a human to judge when AI output is “good enough.”
AI large‑language models can summarize interview notes in seconds, but they still hallucinate facts or miss subtle emotional cues.
Until provenance, privacy, and judgment improve, companies will keep humans in the loop.
Here’s a breakdown of the tasks I think AI can/will do and where humans are still key
A fast case study
Last quarter, a client I was working with tested a new onboarding flow with 12 customers.
Instead of spending two days manually coding transcripts in Excel, they uploaded the sessions to an AI platform.
In under eight minutes it clustered pain points, flagged surprising sentiment spikes, and even suggested insights with decent phrasing for an executive summary.
The team reinvested those saved hours refining follow‑up questions, making sure the insights were highly evidenced and deciding which aspects of what was observed were most worth building.
In total, they actually clocked more time for deep thinking, really thinking hard about the findings and interpreting them - instead of feeling rushed to get just anything sent to the inboxes of the stakeholders and C-suite.
Oh we’re not handing things over to AI just yet…
The case study above is not meant to say that you can just drop results in an AI tools and let it handle the rest.
There was a lot of human involvement in there.
It’s important to know what AI still can’t do, and probably won’t be able to for some time -
What AI still can’t replicate
Authentic rapport. Building trust so customers reveal motivations beyond surface complaints. This is more or less important in different scenarios, but think about how important it is for your product/service contexts and customer base.
Contextual improvisation. Zooming in on a throwaway comment and discovering a hidden use‑case. I haven’t found an AI that can guide me to the right spaces to dig into deeper - in analysis or AI moderation.
Organizational storytelling. Reading the room, tailoring insights to stakeholders’ biases, and sparking action. I use AI all the time to fine-tune my own insight statements, but the final call is mine - I know the client/team member, I understand whether a really shiny statement will be a spark for action, or a dud.
These pieces are honestly more like art forms than science, and are not even close to being solved by algorithms.
Your emerging role: AI pilot & guardian
I’m starting to think of myself less as a “person doing customer research tasks” and more as an airline captain:
Pre‑flight. Feed AI the right corpus - clean data, clear objectives, bias warnings.
Autopilot. Let models crunch transcripts, sentiment, and usage logs while I monitor dashboards.
Mid‑flight checks. Sample outputs, validate anomalies, recalibrate prompts.
Landing & debrief. Translate findings into narratives stakeholders understand and believe.
“AI is the junior or mid-level co‑pilot; you are still responsible for a safe landing.”
Why I believe 95 % of companies will keep talking to customers themselves
Even with perfect automation, we can’t ignore that empathy compounds.
If you’re familiar with the founding story of everyone’s fave design tool, Figma, they spent a remarkable amount of time and energy with the earliest users while their product was still in a kind of forever-closed beta.
They weren’t just trying to learn by deep involvement. They were developing empathy.
You aren’t always your customer. But you can embed yourself in their world enough to feel what they do.
This isn’t the case if we use AI for every part of the research process.
And those customers won’t end up having empathy for us - and letting us off the hook when we make a less-than-awesome product change (that we might need to backtrack).
Product managers, designers and researchers who observe real‑world usage patterns design better solutions and cultivate mission‑driven teams. Boards and investors know this, which is why “voice‑of‑customer” OKRs keep trending upward.