Product Discovery Is Changing. Most Brands Aren't Ready

Product pages were built for humans. Hero images, lifestyle shots, trust badges, emotional copy. Information buried in videos and carousels — because that's how humans like to consume. Text was never the focus.

For agents, it's the opposite. They can't watch your product video. They can't parse your lifestyle imagery. They need structured text, explicit specs, machine-readable data.

Most PDPs don't have that.

Here's the other problem. E-commerce was designed to make you scroll, browse, impulse buy. Agents don't do that. They go straight to what matches the criteria — price, specs, reviews, availability. No wandering. No "oh, that looks nice."

The data is already showing cracks. Cyber Week 2025 marked a shift away from impulse buying toward mission-based shopping. Retail site traffic is down 21% year-over-year — not because people are buying less, but because they're browsing less. And the industry is nervous. Kearney's modeling suggests EBIT erosion of up to 500 basis points as agents drive price transparency and smaller, more targeted orders.

The era of "scroll and discover" is ending. The brands that don't adapt will become invisible.

How Agents Actually Buy

When an agent shops on behalf of a human, it's not browsing. It's executing.

It pulls structured information — not paragraphs. It compares across 50 options simultaneously — not one page at a time. It filters on explicit criteria: price, specs, compatibility, availability. No distractions. No impulse decisions. No interest in your brand story.

The agent has one job: find the best match for the human's requirements. Everything else is noise.

Companies are scared. As one retail analyst put it: when you tell your shopping agent "I need detergent," it's going to find the best price, the highest algorithm ranking, the best reviews. It cuts out the messy, human element of discovery. The marketing dollars spent on inspiration shopping? Nearly worthless when the purchase decision is outsourced to a cost-minimizing algorithm.

What's Broken Today

Most PDPs bury critical information in unstructured copy. Specs hidden in paragraphs. Compatibility mentioned in passing. Dimensions in a tab nobody clicks.

Humans tolerate this. They scroll, scan, piece it together. Agents don't. If the data isn't explicit and parseable, it doesn't exist.

Schema markup helps — in theory. But most implementations are an afterthought. Incomplete. Inconsistent. Outdated. JSON-LD exists on the page, but the actual product attributes don't match the structured data.

The result: agents skip your products entirely. Or worse — they make bad recommendations because they're working with garbage data. Either way, you lose.

The Fix

PDPs need to serve two layers now.

The machine layer — structured, explicit, parseable. Clean schema markup that actually reflects the product. Specs as data, not prose. Compatibility, dimensions, use cases — all in formats an agent can consume without guessing.

The human layer — persuasive, visual, trust-building. This still matters. The human makes the final call (for now). But it's no longer the first filter. It's confirmation after the agent has already shortlisted.

Think of it like SEO was for Google. You optimized for the algorithm and the human reader. Now you're optimizing for agents and the human buyer.

Doing This at Scale

Manual updates won't cut it. You can't have someone hand-edit schema markup for 10,000 SKUs.

Build a single source of truth. All product attributes — specs, compatibility, use cases, availability — in one structured database. The PDP pulls from this. The schema markup generates from this. The product feeds export from this. One source, many outputs.

Templatize the agent-readable layer. PDP templates that auto-pull structured attributes into consistent formats. Human-written copy handles persuasion. Auto-filled specs handle accuracy. Separation of concerns.

Use AI to fill gaps. Most catalogs have inconsistent attribute coverage. Use LLMs to bulk-extract missing specs from existing descriptions. Review in batches, not one by one.

Test with your own agent. Run an AI agent against your PDPs. Can it extract what it needs? Compare against competitors. Whose data is easier to parse? This becomes a new QA layer.

Treat feeds as first-class citizens. Product feeds — Google Shopping, marketplaces, API endpoints — shouldn't be an export from messy PDPs. If the feed is clean, the PDP can inherit from it. Flip the model.

The Mental Shift

PDPs today are designed as pages. Layout, above-the-fold, visual hierarchy.

Tomorrow, PDPs are data objects that render as pages for humans and expose as structured data for agents. The page is just one view of the underlying product data.

Whoever builds this infrastructure first has a moat. Agents will preference sources that are easy to work with. If your competitor's data is cleaner, their products get recommended. Yours don't.

The first satisfying experience wins. For humans, it used to be about visuals and copy. For agents, it's about data quality. The brands that figure this out will own the next era of product discovery.

The ones that don't? They'll wonder why traffic dried up.

If you're thinking about this — how product discovery changes when agents become the first filter — I'd love to hear what you're seeing. DM me or drop a comment.