Beware the useful stick

Back in the day I worked on the buying journey for a manufacturer of roof windows. Their problem, as the client explained it, was straightforward. They made a genuinely clever product: expensive, well engineered, remote controlled, able to bring light into a house through a high or awkward roof. And builders kept telling customers they didn't need it. Get a cheaper version, the tradesman would say. And a long stick so you can open it by hand. Not because the stick was better. Because it was easier to sell and nobody at the trade counter where the builder was shopping had a reason to make the case for anything else.

When we actually researched how people made the decision, it turned out almost nobody was buying a window. They were buying light and the feeling light gave a room. By the time they got anywhere near a builder, they had often wanted it for years: leafing through interior magazines, watching Grand Designs, quietly building a picture of what wanted their own home to feel like.

The brand's job wasn't to beat the stick on price. It was to make sure the customer arrived at the conversation with their builder already knowing what to ask for. That way, a tradesman trying to be helpful by selling a good value stick was no longer the only influence in the room.

I’ve been thinking about that stick again recently. The customer journey has a new trade counter with a new ‘helpful tradesman’ standing behind it called something like Claude or CoPilot.

None of this is really about roof windows, or even about AI. Marketers have known for well over a decade that the ‘journey funnel’ doesn't describe how people actually make decisions. McKinsey formally retired it in 2009 when its Consumer Decision Journey research replaced the neat narrowing shape with something closer to a loop: an initial set of options, an active evaluation stage where people add brands as often as they remove them, a purchase, and a post-purchase experience that either shortens or reopens the next decision.

The digital journey has been messy, non-linear and different for all of us for quite a while.

What's changed isn't the nature of the journey. It's who (or what) else is now involved in it.

The education stage of the journey is frequently the bit brands underplay. Probably because it's slower to build and takes time to work. Financial services brands have often been the exception. A mortgage or an ISA has enough lifetime value that a bank can justify years of patient, unglamorous education content because the investment can eventually be repaid many times over. Most categories never had that luxury. So education often got left to whoever happened to be standing at the counter when the decision was actually being made: a tradesman with a stick to sell, or a shop assistant with their own agenda.

What's new is that a system now sits in that same spot, doing part of the active evaluation on the customer's behalf. It has no reason to prefer the stick the way a tradesman on commission might, but it has no reason to know about the remote control either. Ask it a generic question, does this window open, and it gives a generic answer, a stick works fine. The risk isn't that the system favours the easy option. It's that the easy option is often the only one it's been given anything specific to say about.

That lack of a built-in preference works both ways. G2's research into software buying found that half of B2B buyers now start their research with an AI chatbot rather than a search engine, up from under a third the year before. More strikingly, more than two thirds of buyers in the same research said they'd chosen a different vendor from the one they'd originally planned to, based on what the chatbot told them. A third bought from a brand they'd never heard of before.

That's either the best opening a smaller, genuinely expert brand has had in years. Or the modern version of getting sold the stick. It depends on whether the system can find enough reasons to recommend you.

It's tempting to merely treat this as a new version of search engine optimisation (generative engine optimisation) and go looking for the black box tricks. The research suggests that's not all that we need to think about.

The Princeton study that first explored GEO, tested exactly that assumption. It found that familiar SEO signals such as keyword density had little effect on whether a page was used in an AI-generated answer. What made a much bigger difference was what the researchers called epistemic authority: statistics, sourced claims and quotations from people who actually know what they're talking about.

The system rewards the same evidence a sceptical or information seeking customer would already be looking for. 

It’s an expertise challenge, not a technical one.

It also changes what that expertise needs to look like. Generative engines don't necessarily answer a broad question in one go. They can break it into a series of narrower questions and look for a source that answers each one. Researchers call this ‘query fan-out’.

The practical effect is interesting. A page written about "pet nutrition" in general can lose out to something that answers a much more specific question: for Tom who lives near Bournemouth and wants to know what he should feed a four-year-old Schnauzer called Tony who gets a bad tummy too often and can be choosy with his food?

Nobody used to write much content at that level of detail because there was no obvious way to reach that exact person. Now there is a system actively looking for whoever has something useful to say to Tom and Tony .

Content built around real combinations of needs, rather than broad categories will have the advantage. And that only really comes from a brand that has spent time understanding the specifics of who its category is actually for and what kind of needs they are meeting across the customer lifecyle. The instinct underneath all of this is the same one that mattered at the trade counter back in the day. Teach the person, or the system, standing between you and the customer something worth repeating. Then give the customer somewhere to keep learning more when they arrive.

What has changed is the economics.

A mortgage or an ISA could once justify years of education because the lifetime value was there to pay for it. Content that informs an AI's answer and gives the person it sends onward something genuinely useful doesn't need the same lifetime value to make sense.

More categories than ever can probably now afford to do what only the biggest, most patient brands used to bother with: actually teach people what they're buying, rather than hoping to be the easiest thing on the shelf.

Educate the AI. Then let the AI educate them.