AI is brilliant at the data. It still misses the person behind it.
Customers rarely say what they mean. They buy for status, out of fear, or because a page simply felt right. Those signals live in context, and context is where today's AI is weakest. Here's how I split the work between people and machines.
AI tools organise research, summarise interviews and draft options faster than any person. They still struggle with subtle human context and with biases that never appear in the data. Design and positioning decisions need a person who reads that context.
Who does what
Two threads, braided where the work is shared. Neither one carries the project alone.
Where AI falls short
Subtle context
A customer says the price is fine and then never buys. A person hears the hesitation. A model reads the words.
Learn the termBiases it can't see
People overweight recent events and fear losses more than they value gains. Models mimic some biases in aggregate but for different reasons.1
Learn the termDesign isn't written down
Taste, rhythm and what feels trustworthy in your market are learned by looking, not by reading a rulebook.
Learn the termStakes and empathy
A clinic, a law firm and a roofer all need trust, and each earns it differently. That judgment is human.
Learn the termWhere AI is better than us
Organising data
Hundreds of reviews, survey answers or support tickets, sorted and tagged in minutes.
Learn the termSummarising research
Interview transcripts turned into themes you can check against the recordings.
Learn the termChecking at scale
Every page tested for missing headings, broken schema or a fact that contradicts another page.
Learn the termFast variations
Ten headline drafts to react to, so the person's time goes into choosing.
Learn the termLet Ai carry the boxes. Keep a person in charge of what goes in them.
How I use this on projects
- I run customer interviews myself, then use AI to transcribe and tag them.
- I read the tags against the recordings, looking for what people didn't say.
- Positioning and page goals are decided by people, in a workshop.
- AI drafts variations. I edit, cut and choose.
- AI checks the finished site for structure and consistency before launch.
Keep reading
All insights
Where to next
Bring people back into the picture
Tell me about your customers. I'll show you where your site speaks to the data and misses the person.
Sources
- Idan Horowitz and Ori Plonsky, LLM Agents Display Human Biases but Exhibit Distinct Learning Patterns, arXiv:2503.10248, March 2025. arxiv.org