Comparing Apples with Oranges

Show a founder their best product page and most of the time it looks done. Sharp photography. Confident copy. Variants that select cleanly. Calm returns language. On a laptop, the brand feels organised.

Then ask what an agent gets if it has to shortlist that same SKU against three peers. Wrong Google Shopping category. Thin taxonomy. Materials and care buried in prose the machine never structured. Size and colour variants flattened into mush. Availability that has drifted from the page the human just admired.

The human experience was an apple. The machine merchandising was an orange. For years that mismatch lived in SEO or ads hygiene. Now it is a demand problem. Agents shopping for customers either absorb the friction, or they skip you.

Two products sharing one SKU

Human UX and machine merchandising are not the same job. They share a SKU and a brand name. They do not share the same definition of "good."

Human UX

Persuasion, clarity and trust on the page: layout, imagery, copy rhythm, review social proof and a journey that feels finished to a person.

Machine merchandising

Structured truth an agent can parse: category fidelity, attributes, policy clarity, availability and variant identity clean enough to shortlist and compare.

You can win the first and lose the second without anyone in the leadership team noticing, because the audit method is still "browse our site." That test is for humans. It does not test agents.

A perfect website can mask a data hygiene mess. Agents do not forgive wrong Shopping categories or empty taxonomy.

Where the gap shows up

Four failure modes keep turning up in mid-market catalogues. None of them need exotic technology to spot. All of them quietly take you out of machine comparison.

Category mis-map. Titles get optimised for click-through and drift away from the taxonomy machines use to place you in a competitive set. Humans still find the product. Agents put it next to the wrong peers, or nowhere useful at all.

Thin attributes. Colour, material, dimensions, compatibility, care, age range, wattage: whatever the category needs for comparison often lives as marketing adjectives instead of fields. The PDP reads well. The feed is sparse. Sparse feeds are invisible shelves.

Variant mush. Parent and child identity collapses. Sizes and colours share ambiguous titles. Agents cannot tell whether they are comparing like with like, so they favour the cleaner competitor.

Policy and availability opacity. Returns, shipping promises and stock truth are written for people and delayed for systems. When an agent cannot verify a promise, it invents, hedges or excludes. Invention is a brand-safety problem dressed up as discovery.

None of this looks dramatic in a board pack about site conversion. It looks dramatic when share of answer and agent shortlists become the shelf that matters.

Ownership is the real question

Plenty of brands between £5m and £20m still organise as if merchandising owns the human story, technology owns the feed pipe and growth owns "AI discovery." That split produces theatre: beautiful pages, a feed that is "live," and a budget line for mediated discovery that never names who owns agent-legible truth.

Machine merchandising is a commercial discipline. It cuts across merch, tech, CX and growth. Somebody has to reconcile human title tricks with machine category truth when those incentives pull apart. Somebody has to own completeness by category, not the comfort of a green tick that says the feed exists. And somebody has to own what happens when an agent invents materials or returns language the brand never published.

If that owner has no name, you are not funding discovery. You are funding hope.

What good looks like before you fund discovery

Before the next proposal for answer-engine work, ask for a split view of one commercial SKU: what the human sees on the left, what the agent receives on the right. Put the gaps on the table. Category. Attributes. Variants. Policies. Freshness.

Then ask for completeness rates by category for required and high-value optional attributes, not a green tick that the feed exists. Ask for the SLA on price and availability sync. Ask who notices first when an agent fabricates a claim: marketing, CX, legal, or the customer.

These are buyer questions, not practitioner checklists. They separate operators who can hold a team accountable from founders who get sold a stack.

Five questions for the ELT

1. Who owns machine merchandising? Not the website. Not "the feed." Agent-legible catalog truth across functions.
2. What is our completeness rate by category, and which categories are commercially silent to machines even when the human PDPs look strong?
3. Where do human titles and merchandising incentives diverge from Shopping taxonomy, and who reconciles the two when CTR and category fidelity conflict?
4. What is our SLA for price, availability and policy sync between the human surface and the structured surfaces agents consume?
5. When an agent invents materials, claims or returns language we never published, who detects it, who corrects the canonical facts, and who owns the customer impact?

If those answers are vague, pause the spend. Dual-audience influence starts with catalog honesty, not with another campaign narrative.

Brands that treat human UX and machine merchandising as one conversation will keep brand and product sharp while discovery rewrites itself. Brands that confuse apples with oranges will keep polishing the storefront while agents shop elsewhere.