How AI agents choose which retailers to recommend and buy from
AI buying agents do not browse like humans. They filter, score and eliminate retailers in seconds. Here is the decision logic — and where most UK retailers fail.
When a customer asks an AI agent for the best trail shoes under £130, delivered by Friday, the agent does not scroll Instagram or compare hero banners. It runs a pipeline. Within seconds it scans dozens of retailers, eliminates most of them, and recommends — or buys from — the few that pass every check. Understanding that pipeline is the difference between capturing agentic demand and silently losing it to a competitor.
Pass 1: Discovery — can the agent find you?
Discovery is not SEO in the traditional sense. Agents pull from product feeds, marketplace listings, structured data on PDPs, and partner catalogues. If your product is not in a parseable format — or your GTIN is missing — you are not in the candidate set. Period.
- Indexed product feeds (Google Merchant Center, Amazon, marketplaces)
- Schema.org Product markup with offer, availability, and review
- Open catalog APIs or agent-readable product endpoints
- Consistent naming, categorisation, and attribute coverage
Pass 2: Trust — can the agent verify your claims?
Agents cross-check signals. If your feed says in stock but your PDP says pre-order, you fail. If your price in the feed is £129 but checkout shows £149, you fail. If reviews exist only as unstructured HTML with no aggregate rating, you are less trusted than a competitor with structured review data.
Pass 3: Transaction — can the agent complete the buy?
Even trusted retailers lose if checkout requires steps an agent cannot perform: CAPTCHA, mandatory account creation, payment flows without agent support, or delivery options that cannot be selected programmatically. Agentic checkout protocols are emerging to solve this — but most retailers are not connected yet.
The scoring model (simplified)
| Factor | Weight | Typical UK retailer gap |
|---|---|---|
| Data completeness | High | Missing GTINs, null availability |
| Data freshness | High | Feeds updated weekly or monthly |
| Cross-surface consistency | High | Feed ≠ PDP ≠ marketplace |
| Delivery confidence | Medium | Vague or missing cut-off times |
| Checkout automation | Medium–High | No agentic/API path |
| Brand equity | Low–Medium | Matters less than data for agents |
How to improve your agent score
- Run a live agent-read test on your top 50 SKUs — what does the JSON look like?
- Fix GTIN and availability gaps before any other investment.
- Synchronise price and stock across feed, PDP, and marketplaces in near-real-time.
- Add structured review data and clear return/delivery policies in machine-readable form.
- Map your checkout against emerging agentic payment standards and identify blockers.
Frequently asked questions
- How do AI shopping agents pick a retailer?
- Agents typically run three filters: can I find and parse the product (discovery), can I trust the price, stock, reviews and delivery data (verification), and can I complete the purchase programmatically (transaction). Retailers that fail any critical check are excluded.
- Do AI agents prefer big brands?
- Not necessarily. Agents weight data quality, availability confidence, delivery certainty, and checkout feasibility over brand recognition. A mid-market retailer with clean feeds often beats a major brand with stale or incomplete data.
- What is the biggest reason retailers lose agent recommendations?
- Inconsistent or missing structured data — especially availability, GTIN, and price mismatches between feeds and live storefronts. Agents treat inconsistency as untrustworthy and move on.
Find out what agents see on your estate.
Fixed-scope Agentic Commerce Readiness Audit. £1,950. Ten working days. Scored report, prioritised roadmap, 60-minute debrief.
Book the 30-minute call ↗