Example · Case study
Verity's product pages convinced people, but delivery, returns, and warranty resolved three different ways. Agents bailed to support. We made every promise one citable value across all surfaces.
Verity Goods is a premium DTC beauty brand — strong reviews, strong imagery, persuasive copy. The trust signals were all there, just spread across five surfaces.
In early 2026, support tickets from agent sessions ran 38% higher than from web sessions. Shopping agents couldn't connect delivery, returns, warranty, and ingredients — and pushed buyers to support or to competitors.
We tested 60 real buyer questions against ChatGPT Shopping, Perplexity, and Claude. Three breaks covered 80% of the failures:
'Clinical-grade' was a phrase; the actual list sat in an image tab. Agents quoted the claim but not the evidence, which read as unverified.
The PDP said 3–5 days, the policy page said 5–7, the cart said 2–4 express. Agents showed the most cautious one; a competitor promising two days won the recommendation.
'Money-back guarantee' was a banner; the conditions were a 4,000-word PDF. Agents cited half the promise, and the gap became a support ticket.
"Our pages were strong — but only for humans. Now they're strong for agents too, without any loss for customers."
Four workstreams over 6 weeks. The goal: one version of each rule that people and agents both read.
We listed every trust claim and where it lived — 122 statements across PDP, policy, FAQ, support, and reviews — and grouped them by what the buyer asks before purchase.
Tactic
Crawl + manual tag. Output: a claim sheet with source and a consistency status per rule.
Delivery, returns, and warranty each got one canonical value, injected into every surface and exposed as structured policy an agent reads directly instead of inferring from prose.
Tactic
Single source per rule, rendered into PDP, cart, policy, and FAQ, plus JSON-LD. A CI drift check blocks any surface that disagrees.
{ "@context": "https://schema.org", "@type": "Product", "name": "Daily Repair Serum", "offers": { "@type": "Offer", "priceCurrency": "EUR", // one canonical delivery value, injected everywhere "shippingDetails": { "@type": "OfferShippingDetails", "deliveryTime": { "@type": "ShippingDeliveryTime", "transitTime": { "@type": "QuantitativeValue", "minValue": 2, "maxValue": 4, "unitCode": "DAY" } } }, "hasMerchantReturnPolicy": { "@type": "MerchantReturnPolicy", "merchantReturnDays": 30 } }}The condition now travels with the headline. Where there was a 30-day window, the agent sees '30-day' next to 'money-back' — not three clicks away in a PDF.
Tactic
Structured warranty and return fields, not prose. The condition is data, so it can't get separated from the claim.
A weekly check asks the shopping agents the real buyer questions and records what they say about the brand — and where a competitor answers and Verity didn't.
Tactic
Answer-coverage check against ChatGPT and Perplexity, with a drill-down on the specific claims that fail.
Three headline numbers — and the main one breaks down.
−42% policy escalations
The 42% broke down into four levers we measured separately:
The win wasn't a rewritten policy page. It's that one rule has one value, everywhere — so the next product launches consistent, and a CI drift check plus a weekly coverage test keep it that way without anyone maintaining a checklist.
Example · Case study
Example · Case study
Want the same on your shop?
Check my shop