The second audience

When someone asks an AI assistant to find a flight, book a hotel, shortlist a supplier or choose a product, the assistant does the browsing. It reads pages, compares claims, checks availability and returns an answer. In agentic workflows it goes further: it completes the booking, places the order, fills the form.

The human is still the customer. But the visit, the comparison and often the decision now pass through software acting on their behalf.

This is the shift behind AEO and GEO, answer engine optimisation and generative engine optimisation. The names will not last. The underlying change will. Discovery is moving from ranked lists a human scans to synthesised answers a model composes.

The behavioural evidence is already unambiguous. The Pew Research Center installed browsing trackers on the devices of 900 consenting United States adults and observed 68,879 real Google queries during March 2025. Where an AI summary appeared, users clicked a traditional result in 8% of visits. Where one did not, they clicked in 15%. Links inside the summary itself were clicked in 1% of visits, and sessions ended after 26% of pages carrying a summary against 16% without.

Half the clicks, and more sessions that end where the answer was given.

The instinct is to treat this as a marketing problem: a new channel, a new set of optimisation tactics, a successor to SEO. That framing is too small.

If agents come to handle a meaningful share of historically human processes, booking travel, buying products, comparing services, renewing contracts, then every digital property has two audiences:

The human, who still values brand, trust, aesthetics and feel. Experience still converts, still retains, still commands a premium.

The agent, which values none of that. It values legibility, structure, consistency and verifiability. It needs to parse what you sell, what it costs, what the terms are, whether stock exists, and whether your claims hold up against other sources.

It does not reward beautiful ambiguity. It penalises it.

The design brief becomes dual purpose: the same content, the same catalogue, the same policies, expressed in a way that persuades a human and computes for a machine. Not two websites. One source of truth with two renderings.

Early evidence points to a consistent pattern in what generative and agentic systems favour.

None of this replaces the human experience. It sits underneath it.

The uncomfortable part for experience-led organisations is that the layer they invested least in, the structured, boring, factual plumbing, is the layer the new audience reads first.

I have spent the last several years building an open source project on that fourth point. StackQL exposes cloud infrastructure as data a machine can query rather than an interface a person has to operate, so that agents and people work against the same source of truth, inside boundaries someone has set deliberately. Configuration as data, and policy-bounded execution as a first-class concern rather than an afterthought.

The problem it solves and the problem in this essay are the same problem in different clothes. Getting it wrong in infrastructure costs you an outage. Getting it wrong in the customer estate costs you a customer you never knew was looking.

Digital natives have seen this movie before, from the other side.

Incumbents once dismissed the web as a channel. Digital natives understood it was a different operating logic, rebuilt around it and took the market. The advantage was never the technology. It was being organised for the medium while competitors bolted it on.

The same asymmetry is forming again.

A generation of agent-native businesses is being built for machine customers first: catalogue as API, policies as data, content as answers. For them, the human interface is one rendering among several. For most digital natives, the machine interface is an afterthought.

The risk is not that agents ignore you next quarter. It is that agent-mediated demand grows a few points a year, agents form durable preferences for the sources they can trust and transact with, and by the time it shows up in the numbers, the selection patterns are set.

This is not a task for the SEO team.

The questions it raises cut across strategy, structure and capability.

The experience advantage that defined the last era is not obsolete. Humans remain the customer, and the businesses that serve them well will keep earning loyalty and margin. But a second audience now sits between you and a growing share of demand, and it judges you on different terms.

The organisations that lead will treat that audience as a design constraint from the start, the way digital natives once treated the browser.