How an ad enters the conversation

On 9 October, THG Commerce announced that it had integrated Criteo Sponsored Products into AI Shopping Assistant, a conversational product-discovery service built using Google Cloud Gemini Enterprise. THG Commerce calls itself the first European commerce-technology provider to move this advertising format into conversational search; an independent trade publication corroborated the launch, but not the first claim or the metrics. [1 · THG Commerce · integration announcement with Criteo, 9 October 2026] [2 · Prolific North · independent trade report, 9 October 2026]

The assistant extracts keywords and filters from a conversation, while Criteo relevance models match them to eligible advertised products. If a product clears the threshold, it appears clearly labelled beside organic recommendations. The companies did not disclose whether sponsorship changes the order of organic results or how the auction operates. [1 · THG Commerce · integration announcement with Criteo, 9 October 2026] [2 · Prolific North · independent trade report, 9 October 2026]

THG Commerce offers the assistant as standalone software as a service (SaaS) that can connect to an existing storefront without replacing the commerce platform. THG Commerce and Google Cloud previously launched AI Shopping Assistant on 17 June and reported an early deployment; those earlier figures concern the assistant as a whole, not the effect of Criteo advertising. [1 · THG Commerce · integration announcement with Criteo, 9 October 2026] [3 · Google Cloud · AI Shopping Assistant launch, 17 June 2026]

Large multipliers without a reproducible denominator

THG Commerce says sponsored products achieved click-through rate (CTR) more than five times the average and conversion more than twice the average. The introduction frames CTR against a standard site average, but the conversion baseline and the main results sentence are not precisely defined. It is unclear whether conversion is measured per click, conversation, session or impression. [1 · THG Commerce · integration announcement with Criteo, 9 October 2026] [2 · Prolific North · independent trade report, 9 October 2026]

The test period, number of conversations and shoppers, absolute CTR and conversion, traffic allocation, control group, confidence intervals and attribution rules are not published. The evidence therefore cannot separate the ad format from the higher prior intent of people who already use the assistant, seasonality, product mix or placement. [1 · THG Commerce · integration announcement with Criteo, 9 October 2026] [2 · Prolific North · independent trade report, 9 October 2026]

Even a real increase in CTR means only more clicks per impression. It is not incremental sales, return on ad spend (ROAS), gross profit, repeat purchase or customer lifetime value (LTV). THG Commerce and Criteo did not disclose media spend, advertiser revenue, the retailer’s share, organic cannibalisation or the effect on returns. [1 · THG Commerce · integration announcement with Criteo, 9 October 2026] [2 · Prolific North · independent trade report, 9 October 2026]

From monetising intent to a conflict of trust

The product logic is understandable: conversation captures a need before the shopper has formed an exact query and creates new advertising inventory. Descriptive evidence from 31 million users of a major Chinese travel platform found that embedded AI generally complemented search and was used disproportionately for exploratory tasks. Evidence from another platform does not show that THG Commerce advertising improves choice or sales. [4 · arXiv · Shopping with a Platform AI Assistant, March 2026]

The channel also creates a structural conflict: one interface both advises the shopper and earns money when a sponsored product is chosen. In two preregistered experiments with 2,012 participants, conversational AI steered choices toward randomly designated sponsored books much more often than traditional search; labelling alone did not significantly reduce the effect. This is a laboratory risk, not an audit of THG Commerce. [5 · arXiv · Commercial Persuasion in AI-Mediated Conversations, April 2026]

A minimum verifiable safeguard is an unchanged organic benchmark, a clear explanation of paid status and auditing of sponsored-answer share. Performance needs a randomised persistent holdout among comparable conversations: the same assistant, but advertised products cannot enter the result. That would support incremental gross-profit measurement alongside assistant abandonment, complaints, returns and repeat purchase. [1 · THG Commerce · integration announcement with Criteo, 9 October 2026] [2 · Prolific North · independent trade report, 9 October 2026] [5 · arXiv · Commercial Persuasion in AI-Mediated Conversations, April 2026]

Expert commentary

For a retailer, this is more than another ad slot. Conversation surfaces purchase purpose and constraints earlier than an ordinary search query, so a suitable recommendation can shorten the path to a product. Yet that advisory proximity also makes commercial influence less obvious: users may treat the answer as neutral help even when part of the result is sold to an advertiser. [1 · THG Commerce · integration announcement with Criteo, 9 October 2026] [2 · Prolific North · independent trade report, 9 October 2026] [5 · arXiv · Commercial Persuasion in AI-Mediated Conversations, April 2026]

The reported multipliers are plausible for a high-intent audience, but they do not yet demonstrate causal value. People who open the assistant and continue a conversation may already be closer to purchasing than the average site visitor. Comparison with a site-wide average magnifies the impression; a fairer benchmark is similar conversations using the same organic assistant with advertising randomly disabled. [1 · THG Commerce · integration announcement with Criteo, 9 October 2026] [2 · Prolific North · independent trade report, 9 October 2026]

Economics should be calculated at incremental gross-profit level. The numerator is sales and ad revenue that would disappear without the integration; the denominator includes media cost, Criteo’s share, implementation, operations and possible cannibalisation of organic items. Without that calculation, high CTR can be evidence of engagement but not of a return. [1 · THG Commerce · integration announcement with Criteo, 9 October 2026] [2 · Prolific North · independent trade report, 9 October 2026]

Clear labelling is necessary but insufficient for trust. Experiments with AI intermediaries suggest commercial steering can persist even with a “Sponsored” label, while the platform role changes how an agent evaluates a paid option. A practical test should examine not only label visibility but stability of organic ranking, explainability of recommendations and the user’s ability to request an ad-free alternative. [5 · arXiv · Commercial Persuasion in AI-Mediated Conversations, April 2026] [6 · arXiv · Whom Do AI Agents Work For?, September 2026]

If the model scales, platforms with strong catalogues, abundant conversations and sustained advertiser demand gain an advantage. Brands receive access to the moment when preferences form; smaller sellers face the risk that paid visibility becomes a condition of presence. Watch the organic-product share, concentration of impressions and customer-acquisition cost by category. [1 · THG Commerce · integration announcement with Criteo, 9 October 2026] [2 · Prolific North · independent trade report, 9 October 2026]

Over the next 6–12 months, convincing evidence would be a publicly defined experiment: absolute CTR and conversion, sample size and period, a persistent control, incremental profit, repeat purchase, returns and complaints. Until then, the integration is a meaningful new retail-media format, but its benefit to customers and retailer economics remains a hypothesis. [1 · THG Commerce · integration announcement with Criteo, 9 October 2026] [2 · Prolific North · independent trade report, 9 October 2026] [5 · arXiv · Commercial Persuasion in AI-Mediated Conversations, April 2026] [6 · arXiv · Whom Do AI Agents Work For?, September 2026]

Sources

  1. THG Commerce · integration announcement with Criteo, 9 October 2026 — Primary source for the launch, placement mechanics, labelling, early metrics and SaaS delivery. The metrics are company-reported; methods and absolute values were not published.
  2. Prolific North · independent trade report, 9 October 2026 — Corroborates the date, integration, adjacent sponsored and organic recommendations, and reported early metrics; it does not independently validate the results.
  3. Google Cloud · AI Shopping Assistant launch, 17 June 2026 — Primary context for the assistant’s initial launch and earlier Myprotein results. Those figures concern the assistant overall, not the new advertising integration.
  4. arXiv · Shopping with a Platform AI Assistant, March 2026 — Descriptive study of 31 million Ctrip users: chat complemented search and was used disproportionately for exploratory tasks. It neither studies THG or Criteo nor identifies an advertising effect.
  5. arXiv · Commercial Persuasion in AI-Mediated Conversations, April 2026 — Two preregistered experiments, N=2,012: randomly designated sponsored products were chosen more often through conversational AI than traditional search. This provides experimental risk context, not an audit of THG.
  6. arXiv · Whom Do AI Agents Work For?, September 2026 — Controlled model experiments show that responses to sponsorship depend on the AI agent’s assigned role; the preprint does not study THG’s product or replace evidence from real shoppers.