Word of Mouth to Word of AI

6 years back then, on the year 2020, a conversation with me and my friend,

My friend: Hi macha, I want to buy a car, tell me which one.

Me: Buy this one, Brand A, the petrol variant. For you, this is the car.

My friend: Ok macha, super, I bought that.

Fast forward to 2026, the same conversation:

My friend: Hi macha, I am replacing my car, any suggestions?

Me: Buy this one, an EV, that is the right one for you.

My friend: AI said to buy that too, machi. I bought it.

Same friend, same question, same me giving the answer. But somewhere between those two conversations, my role quietly changed from source to second opinion.

Three shifts arrived while nobody sent a memo

The eyes are leaving the web, and so did the humans. In my first post I wrote about AI agents nudging without eyes. It turns out they are not a minority visitor anymore. Automated traffic overtook human activity on the open web for the first time in a decade in 2025, and by this year automated traffic accounts for more than 53% of all web traffic, human activity down to 47%. We are, technically, the minority now.

UX is splitting into UX and AX. Design has a new discipline this year, formalised in the 2026 Design in Tech Report: agent experience, AX, sitting alongside user experience, UX. A design commentary unpacking the split put the difference in exactly these words: “The design problem shifts from how do I help someone do this, to how do I help someone know whether it was done well.” Two different users, two different questions, one product.

The purchase path grew a middleman. User to platform has become user to agent to platform. I have written about the commerce rails being laid for this, merchant fees, checkout protocols, catalogue syndication. The structural layer exists now. What I had not fully clocked until my friend’s phone call is that the same shift applies to advice, not just checkout.

A rougher way to slice it (a thought experiment, not a finding)

Here is my own working sort, in four buckets, not an established framework.

D2C, low stakes, within a set budget. The agent simply acts. Tell it “reorder my usual detergent, keep it under 500 rupees” and there is no room, and no real need, for word of mouth. This is the one bucket I feel fairly confident about.

Products with touch, feel, or emotional weight. A car for some people. A wedding planner for others. Half of India’s car buyers already spend over 10 hours researching before deciding, and word of mouth remains, by a wide margin, the most cited reason for the final choice. Does that trust shift to word of AI, the way it did for my friend, or does it hold precisely because the decision carries feeling a machine cannot share in?

Luxury. A 2026 McKinsey piece on AI in luxury retail found that shoppers already trust an AI’s pick nearly as much as a friend’s, and trust it more than an influencer with an obvious reason to sell. At the same time, luxury brands are reinvesting heavily in human, tactile experience exactly where AI is taking over the operational layer behind the scenes. Put those two together and here is my guess: AI wins the shortlist even in luxury, but the close stays human, because exclusivity needs a witness. Word of mouth may not disappear here so much as move to a new question: whose concierge do you trust more, the human one or the AI one?

B2B referral networks. This is the deep one. A Semrush survey of over 600 US business professionals found buyers trust their peers’ opinions over anything a vendor publishes. A separate industry report found the opposite trend gaining ground at the same time: more than half of B2B buyers say their trust in AI generated vendor shortlists has grown in the past year. Both of those cannot stay fully true forever. Whether the referral network breaks, bends, or moves to a new surface is a question worth its own post, not an answer I can give in passing here.

Where is the middle line

This is the question that will not leave me alone. If agent experience research studies the software, and user research studies the human, who is studying the handshake between them, the moment my friend’s question left him and reached an AI, and the moment its answer became his answer?

Is the “respondent” in a research study now the human, the agent, or the exchange between them?

When word of mouth becomes word of AI, does trust transfer with it, or does something get lost in the handoff that nobody has named yet?

And the category questions I am sitting with longest: does exclusivity in luxury survive an agent that treats every customer’s recommendation the same way? And does a B2B referral network, built over years of dinners and shared alumni ties, hold against a shortlist an AI can generate in ten seconds?

I write about AI, behavioural economics, and user research. Mostly questions, occasionally answers. If you are thinking about any of this, I would like to compare notes.