Customer review quality steers the agent basket

If an AI assistant has already trimmed the shortlist before a shopper opens your product page, what exactly is it reading? Adobe's festive forecast puts UK online spend at a record £28.52 billion between 1 November and 31 December 2026, up 6.2% year on year, while retail traffic arriving from generative AI platforms is projected to rise 214% in November and December, up from 111% in September(Retail Technology Innovation Hub). The same holiday read puts US online spending on Black Friday at $12.9 billion, up 9.2% year on year(Webiators). As of 10 November 2026, the more useful question is not how much of that traffic converts, but which product attributes survive the shortlist.

Agent referrals climb faster than checkout trust

The two numbers move in opposite directions. Adobe's US read for the same window expects retail traffic from consumer AI assistants to grow 130% across the golden quarter, with Thanksgiving up 159%, Black Friday up 95% and Cyber Monday up 88%, while paid search grows 4% and social 17%(CX Network). Yet the same reporting cites YouGov's 2026 study: 40% of British shoppers are comfortable letting AI find the best price, 7% would let it add items to a basket, and 5% would let it complete a purchase. Discovery has been delegated; commitment has not.

That gap defines the work. Traffic that arrives pre-qualified still lands on a product page, and the page has to confirm what the assistant implied: size, material, delivery date, return window, stock status. Every missing attribute becomes a reason to leave, because the shopper already has a shortlist and can compare in one more prompt.

Discounts are deeper and less believed

Adobe expects UK discounts to climb from 12% in early November to 18% just before Black Friday, peaking at 25% from the start of Cyber Weekend, with online prices down an average of 2.5% year on year and units sold in the top price quartile up more than 13%(Retail Technology Innovation Hub). Deeper cuts are meeting a shopper who has been tracking prices since October: 34% of likely Black Friday and Cyber Monday shoppers expect to use AI shopping tools, up from 16% in 2025, and one in three is most likely to abandon at the shipping-options step(Webiators).

Price alone therefore stops being a differentiator at the moment the assistant can compare it in one turn. What the assistant cannot synthesise as easily is evidence that the product performs: how many reviews exist, how recent they are, how specific the complaints are, and whether the same defect appears across variants. That is why review depth, not discount depth, decides which listings remain in the answer.

What review depth changes on the shelf

Fewer SKUs, more evidence per SKU

Customer review volume is the cheapest signal an assistant can summarise, and it is also the easiest to fake badly. A catalogue built for search rewarded coverage: many variants, many keywords, thin pages. A catalogue read by agents rewards density: fewer listings, each carrying structured attributes, a stable price history and a review base large enough to be summarised. Retailers running both a marketplace feed and their own storefront should expect the same product to perform differently in each, because the agent sees two different sets of attributes for what shoppers consider one item.

Negative review density as a filter

Aggregate star ratings travel well and tell an assistant almost nothing. What changes a recommendation is the concentration of a specific complaint: sizing runs small, the battery degrades after six months, the described colour differs from the delivered one. Grouping reviews by defect and tracking the share of reviews mentioning each defect gives a number that can be monitored weekly, compared across variants, and used to decide whether to fix the product or fix the page.

A four-step check before the peak

The work is unglamorous and mostly internal: confirm what the assistant is reading, make the page agree with it, then watch whether the answer changes. It runs on a weekly rhythm rather than a campaign calendar, and it lives with the catalogue and customer review teams rather than with media buying. Done properly, it turns a vague worry about AI traffic into a short list of fields that can be checked on a Monday morning.

1. Sample the questions: write the twenty prompts a real shopper would use, across three assistants.
2. Record the answers: keep the raw response, the timestamp and the cited sources for each prompt.
3. Fix the gaps: repair attribute fields and consolidate review themes before adding traffic or discounting deeper.
4. Re-run and compare: repeat the same prompts weekly and log which products enter or leave the answer.

None of this guarantees a recommendation. It does make the listing legible to a system that is summarising rather than browsing, and it turns a vague worry about AI traffic into a set of fields that can be checked on a Monday morning.

FAQ

• Why does AI referral traffic matter if so few shoppers let an agent pay?

  Because discovery is where the shortlist is set. Adobe projects UK AI referral traffic up 214% this season while paid search grows far more slowly, and a shopper who reads an answer often returns later by typing the brand name, so click-based attribution understates the effect.

• Do deeper discounts still work in an agent-mediated journey?

  They work as a filter, not as a persuasion device. Discounts pull a listing into consideration, but the assistant compares prices in one turn, so the discount has to be verifiable against the product's own recent price rather than against an inflated reference price.

• What is the minimum review base worth managing?

  Enough reviews to separate a defect pattern from noise, and enough recency that the summary reflects the current product. A large archive with no reviews in the last 90 days describes a product that may no longer be shipping.

• How should a team measure success in the first month?

  By attribute coverage, review recency and answer stability rather than by revenue. Those three move first, and they are the inputs a team can actually control before the peak week.

Sources

• Retail Technology Innovation Hub, "Brits set for record £28.52 billion festive online spend amid surge in AI driven shopping" (5 November 2026)
• CX Network, "How will agentic commerce shape Black Friday 2026?" (November 2026)
• Webiators, "10 Ecommerce Black Friday Marketing Tips to Boost Sales in 2026" (November 2026)