You're Now Designing for Two Audiences. One of Them Can't Be Persuaded.
My background is in conversion rate optimisation. I've spent a long time thinking about why people don't buy, diagnosing the specific friction points that sit between intent and action, and building counter-arguments to the objections visitors carry with them when they arrive at a page.
The discipline has a rigour to it that I think gets under appreciated. It isn't button colour testing. At its core it's applied psychology. It’s structured attempt to understand how humans make decisions under uncertainty and to design experiences that work with that process rather than against it. The O/CO framework I use with clients maps objections to counter-objections and places those counter-arguments at the precise moment in the journey when the objection is most likely to surface. The diagnostic work that precedes it is the same detective process regardless of the brand or the category. What's stopping the visitor? What would need to be true for them to proceed?
The answer is always in the data and then applying a conversion methodology to it. And it worked a treat. And that may be about to change.
Robert Cialdini noted the psychological levers that dictate this methodology. Social proof, because humans look to others for guidance when they're uncertain. Authority, because credibility signals reduce the cognitive load of evaluation. Scarcity and urgency, because humans procrastinate and need a reason to act now rather than later. Commitment and consistency, because people behave in ways that align with prior decisions and self-image. Liking, because we buy from people and brands we feel a connection with.
These aren't tricks. They're a map of how human cognition actually works when it's navigating uncertainty about a purchase. Every design decision in effective conversion optimisation is traceable back to one of these mechanisms. A well-placed review carousel is social proof deployed at the moment of maximum doubt. A countdown timer is scarcity and urgency made visible. A founder story is liking and authority compressed into a few hundred words. Visual hierarchy exists because human attention is selective and has to be guided. Progressive disclosure exists because humans have a limited working memory and shouldn't be asked to process everything at once.
The entire architecture was built for a specific visitor. A human with emotions, limited attention, social instincts, and a cognitive system that takes shortcuts when it can.
That visitor hasn't gone anywhere. But a second visitor type arrived this year. And it operates on completely different terms.
What the second visitor is doing on your site
AI shopping agents don't browse in any sense that maps to human browsing. They don't arrive at your homepage and form an impression. They don't scroll through a PDP and feel reassured by the photography. They don't experience the weight of a countdown timer or the pull of a review carousel.
What they do is query structured data and resolve product identity from machine-readable signals. An agent handling a customer's shopping query is trying to answer a specific question: is this brand a verifiable, credible entity that I can confidently recommend or transact with on behalf of my user? The resolution process draws on sources the agent trusts independently. The includes GS1 registry data, Organisation schema on the brand's domain, independent review platforms with established authority, editorial mentions in credible external sources (I’m looking at you popular Reddit posts). It cross-references these signals against each other and builds a confidence level for the recommendation.
None of the persuasion architecture touches any of that process.
A countdown timer is a psychological lever. The agent doesn't have psychology. It reads the inventory availability field in your product schema and makes a factual assessment of whether the item can be purchased. A review carousel is a human-facing design element. The agent reads the structured data underneath it if that data exists and is properly marked up, and then goes to Trustpilot and Google Reviews to check whether the brand-level sentiment there is consistent and substantial. Visual authority such as premium photography, clean layout, the signal quality of a well-designed site is invisible to the agent. The data infrastructure underneath the design is everything it cares about.
The agent isn't deciding whether to trust you in the way a human may trust you. It's checking whether you're verifiable. Those are different problems requiring different solutions. Think Captain Kirk and Doctor Spock.
Two stacks, almost no overlap
I want to be precise about the gap here because I think the instinct when people first hear this argument is to reach for reassurance. Surely good UX helps agents too? Surely a well-structured site benefits both audiences?
In some narrow technical senses, yes. A site with good accessibility markup is easier for an agent to parse. A site without layout shifts is less likely to cause agent interaction failures (ever wonder why Cumulative Layout Shift is so important n Web Core Vitals). But these are edge cases, not the main point. The core skills of conversion optimisation includes understanding human psychology, sequencing information for a cognitive system with emotions and social instincts, placing persuasion signals at the moments of maximum friction which have no meaningful application to an audience that lacks those characteristics.
The human stack is what CRO has always been. Diagnosis of friction through session recordings, heatmaps, user testing, and the Golden Question - ‘what almost stopped you from buying today?’. Hypothesis development grounded in the specific objections real visitors are carrying. Counter-objections placed at the precise moment of objection. Testing with real traffic to validate. Iteration. The DiPS framework the O/CO methodology championed by Conversion Rate Experts. Cialdini's principles applied to specific conversion problems for specific audiences applied through the customer journey.
None of this changes. None of it becomes less important. The human visitor is still visiting and still needs all of it.
The agent stack is different in kind, not just degree. Structured product data with GTIN coverage verified at GS1. Organisation schema establishing the brand as a verifiable legal entity with consistent identity across independent platforms. sameAs links that hold up when followed. Brand-level review depth on Trustpilot and Google Reviews, actively managed and growing. Feed health in Merchant Center. Policy schema including MerchantReturnPolicy and OfferShippingDetails that tells the agent your checkout is reliable enough to recommend. Person schema connecting the humans behind the brand to the entity itself.
None of these are persuasion techniques. They don't move a visitor from doubt to action. They're verification infrastructure. They answer a different question for a different audience using a different set of skills and tools.
The techniques don't overlap. The commercial objective is the same. Getting a customer to choose your brand over upper competition. The path to that objective is completely different depending on which visitor type you're optimising for.
Whether they converge, I genuinely don't know
There's a version of this argument where the two stacks eventually merge. A brand with a coherent entity infrastructure is, by definition, a brand that presents clearly and consistently. The discipline of getting structured data right forces a rigour about product information that probably improves the human experience too. Better product descriptions, more accurate sizing information, cleaner policy language. And as agents become more capable of interpreting natural language and visual content, the gap between what the human sees and what the agent reads may narrow.
There's an equally plausible version where they stay separate for a long time. The skills required for each stack are genuinely different. Conversion rate optimisation is rooted in behavioural psychology. Entity infrastructure work is rooted in data governance and structured data standards. The tools are different. The success metrics are different. The people who are good at one are not automatically good at the other. And organisations that try to assign agent readiness work to the CRO team, or try to assign conversion optimisation responsibility to whoever manages the product feed, are going to find that neither gets done well.
I'm still working out what the right answer is here, and I'd be sceptical of anyone who claims certainty about how it resolves. This includes the big AI tech companies. What I'm more confident about is that the mistake to avoid is assuming the two stacks will naturally converge without deliberate effort, or that optimising one is a substitute for building the other.
They're probably separate for longer than most people expect. The organisations that handle it best will be the ones that build both deliberately and resist the temptation to assume coverage from one side.
What this means for anyone coming from CRO
If your background is in conversion rate optimisation the way mine is, this moment is intellectually interesting and professionally uncomfortable in roughly equal measure.
The interesting part is that the diagnostic discipline transfers. CRO is fundamentally about understanding how a decision-maker evaluates a choice and removing the obstacles between evaluation and commitment. The decision-maker has changed. The evaluation process has changed. The obstacles are different. But the underlying question of ‘what is stopping this visitor from choosing us’ is the same. The answer for an agent involves verifiability and data coherence rather than psychology and friction, but the habit of asking the question precisely before reaching for solutions is exactly the right starting point.
The uncomfortable part is that most of the tactical expertise doesn't transfer. The years spent understanding which persuasion signals work at which moments in the journey, the intuition for where friction lives in a checkout flow, the feel for when a hypothesis is strong enough to test. None of this applies to an audience that can't be persuaded. Starting from scratch on the agent stack means accepting that competence in one doesn't confer competence in the other. That's a harder position than most practitioners want to sit with.
The brands I'm most confident about for the next few years are the ones where someone has decided to take both seriously. Not one person doing both. Two distinct workstreams with different owners, different success metrics, and different toolkits, run in parallel. The human visitor needs great CRO. The agent needs a clean entity. These are separate jobs. Both matter. Neither substitutes for the other.
The infrastructure for agent commerce is live now. ACP is processing real transactions. Universal Cart is rolling out this summer. The brands treating agent readiness as something to think about later are already behind the ones that started building six months ago. The persuasion work, the A/B testing, the friction reduction on the human journey. Keep doing all of it. It still compounds. The human visitor is still there.
Just stop assuming it's also solving the other problem. It isn't.