---
"title": "Who controls discovery when AI answers first?"
"date": "2026-10-06T07:14:40-04:00"
"url": "https://www.criteo.com/blog/who-controls-discovery-when-ai-answers-first/"
"description": "AI is reshaping the shopper journey. So what can brands and retailers do to control, measure, and connect sales across AI-first commerce?"
"image": "https://www.criteo.com/wp-content/uploads/2026/10/BLOG-Discovery-will-separate-from-checkout-_-Header-Image-v1.png"
"author": "Rob Taylor"
"category": "Commerce AI"
---

When ChatGPT hit the mainstream in late 2022, very few of us could have predicted what would come next.




AI adoption since that time has grown exponentially, but it’s the world of commerce where these assistants are really mixing things up.




In our [Commerce & AI Trend Report](https://www4.criteo.com/en/2026_commerceai_trends/) from earlier this year, we found that 47% of shoppers use AI to research and compare products, while 39% use it to discover new products and recommendations. But this doesn’t mean they’ve abandoned everything else.




Among shoppers who regularly use AI to shop, 96% still use other channels along the way, including search, brand sites, and retailer sites. The journey is fragmenting rather than consolidating.




Traffic data points the same way. Ecommerce visits to retailer sites have continued to grow – [6.8% year over year according to a recent Similarweb study](https://finance.yahoo.com/technology/ai/articles/similarweb-state-ecommerce-2026-report-120000041.html) – throughout the rise of AI assistants.




And this fragmentation creates an interesting challenge for commerce, because the place where a shopper **decides what to buy** may no longer be the place where they **eventually buy it**.




For retailers and brands, that raises some commercially important questions:




- Who influences the purchase decision?
- Who captures the commerce signals?
- Who is best suited to connect those signals to the sale?




## Who owns the conversation?




Picture a shopper looking for a new coffee machine. We’ll call him Brian.




Brian has limited counter space, a specific budget he wants to stick to, and absolutely zero interest in spending half an hour cleaning the thing after his morning espresso.




Luckily for Brian, he has options.




He could browse a retailer’s site, compare products using a search engine, watch a few video reviews, or ask an AI assistant to narrow things down for him. The end result is that, by the time he reaches a product page, much of the deliberation may already be done.




AI hasn’t replaced the existing shopping journey. Instead, it’s added another place where a shopper’s needs are interpreted and choices are narrowed.




Where this step happens matters.




[Inside a retailer-owned assistant](https://www.criteo.com/blog/the-next-frontier-of-retail-media-is-already-here-inside-ai-powered-conversational-search/), discovery can stay connected to the retailer’s checkout and reporting. On an external AI platform, like ChatGPT or Claude, the retailer or brand doesn’t control the discovery experience – even if Brian does eventually click through to its site.




It’s tempting to put retail media on one side and performance media on the other. But these are discovery environments, not media categories. Both retail media and performance media can play a role in external AI discovery, depending on the advertiser, destination, and commercial relationship.




What matters is what each player controls, what it can monetize, and what it can measure.




## When the retailer owns the discovery layer




For retailers with their own AI shopping assistant, sponsored recommendations can be an extension of their existing Criteo Sponsored Products integration, not a net-new build.




Let’s say Brian shares his requirements with a retailer’s own assistant: limited counter space, a firm budget, minimal cleaning. The assistant turns that into structured signals, like keywords and filters, and the retailer (or its LLM partner) can pass them through that integration. Criteo returns relevant products, and a Sponsored Product can appear within the assistant. Brian can then continue to the retailer’s checkout.




For retailers, this capability creates another monetizable surface.




For brands, it creates a new way to reach someone who is actively explaining what it is they want to buy.




There’s a measurement advantage here, too. An AI assistant owned by the retailer means they can use existing retail media workflows to separate AI-assisted commerce performance from traditional search-influenced sales reporting.




So, while the way a shopper discovers a product might change, the retailer can still retain that all-important connection to checkout that makes closed-loop retail media so valuable. The same holds for brands that own the conversion. When the sale completes on the brand’s own site, the brand owns that transaction, so what it can measure depends on the tracking it has in place.




## What changes when discovery happens elsewhere




Now let’s imagine Brian takes his question to an external AI platform like ChatGPT.




There’s an important distinction to make here. Advertising in ChatGPT follows its own activation flow, and [OpenAI states that](https://openai.com/index/our-approach-to-advertising-and-expanding-access/) ads remain separate from ChatGPT’s answers and do not influence them.




In the case of [Criteo’s partnership with ChatGPT Ads](https://www.criteo.com/news/press-releases/2026/08/criteo-accelerates-chatgpt-ads-integration-with-europe-expansion/), our input is really about providing a commerce layer. Brands and retailers can advertise on ChatGPT using budgets, campaign inputs, and eligible product feeds they provide, and they can select destination URLs with relevant tracking parameters.




So Criteo provides a lot of the commerce context, but ChatGPT controls the ad decisioning, including which ad gets rendered and where, when, and to whom it’s shown.




The larger difference becomes clearer after the click:




- Send Brian to a brand’s own site, and visits and downstream sales can be measured where Criteo event tracking is available through integrations like OneTag or server-to-server connections.
- Send Brian to a retailer, and the brand’s conversion visibility may be unavailable or modeled depending on the brand-retailer relationship, agreed data sharing, retailer-side Criteo integration, or an enabled measurement partner.




This is all in-the-weeds implementation stuff – but it’s also in some ways a budget decision.




These factors determine what a brand can realistically expect to learn from its spend on AI platforms. Can it assess sales performance? Or does the activation need to be judged through the consideration and engagement that’s actively being observed?




External AI discovery is an emerging source of both consideration and demand, but its measurement, optimization, and place in cross-channel budget allocation are still developing.




## Getting credit isn’t the same as creating value




Discovery via external AI platforms doesn’t necessarily move checkout away from either the retailer or brand. It’s more a question of whether the AI exposure or discovery signal can be connected to the eventual sale.




To pin that down, we need to be very clear about what measurement is supposed to prove. Right now, we can do that in three core ways:




1. **Delivery** tells us what happened: impressions, clicks, spend, and product-level engagement.
2. **Performance** tells us which outcomes can be attributed: visits, conversions, attributed revenue, and return on ad spend, where the destination, integrations, and available data permit.
3. **Incrementality** asks what additional value the exposure generated. Answering that requires a controlled test or another appropriate causal methodology.




An attribution report tells us which activation receives credit for a sale. It doesn’t tell us whether the shopper would have bought anyway.




It’s also worth being careful about cross-channel comparisons. ROAS from an AI platform and ROAS from standard display aren’t measuring the same thing, so putting them side by side tells you more about the maturity of each measurement method than the value each channel created. At this early stage, it’s more useful to focus on leading indicators. Site engagement, the share of visitors who are new, and new-to-brand purchase rates all point to whether AI-assisted discovery is bringing in demand that wasn’t already there before.




So, where the signals don’t connect, the most valuable work is in locking down data sharing, integrations, and partnerships. Closed-loop retail media measurement remains vitally important, but extending the right connections into new discovery surfaces moves right alongside it.




## Reconnecting discovery with commerce




There’s still plenty that retailers and brands can do to reach shoppers like Brian without pretending there aren’t still lots of open questions around AI-assisted commerce:




1. **Get the product information right.** Product details, offers, inventory, and landing pages are part of the discovery infrastructure.
2. **Write for prompts, not keywords.** Shoppers describe a situation to an assistant rather than typing a target phrase, so creatives built for keyword matching actually work against how these systems judge relevance. Criteo Prompt Smart Ads, currently in beta, generates creative variations and context hints shaped around precisely that behavior.
3. **Know what you can promise (and what you can’t)**. Establish who controls the surface and which signals can reach the destination’s measurement system before committing the budget. For brand-to-retailer campaigns, confirm the data-sharing agreement and any measurement partner that needs to be enabled.
4. **Keep the reporting distinct.** Retailer-owned assistant reporting and external AI platform reporting shouldn’t be combined when the conditions behind them are different.
5. **Plan how you’ll test additional value.** Use incrementality testing where the available data and test design make it possible.




There’s a lot there, but, importantly, none of this belongs in the post-campaign debrief for the first time. Nobody benefits from discovering the limits of a measurement promise after the budget has been spent.




Retailers need to protect the value of their first-party shopping experiences while making products discoverable elsewhere. Brands need to understand whether an activation is actually buying measurable performance, qualified consideration, or earlier influence on demand.




It’s possible discovery and checkout may continue to separate. Ensuring they stay connected takes the right data, agreements, and measurement for each environment.




Before your next AI media investment, make sure someone can explain how that connection will work – and what you’ll actually be able to prove when the sale happens.