Ever since the launch of Atlas, OpenAI's new browser, one question has probably been nagging at plenty of decision-makers in the sector: what impact will the agentic AI wave have on retail media?
Put differently, how far do AI assistants threaten this young but thriving market (Warc is forecasting $176 billion in revenue worldwide this year)? A market that will have to deal with what may be an unsolvable equation: can you keep selling ads to brands when those ads are increasingly seen, to no effect, by agents rather than humans?
I would encourage the sceptics to test Atlas, as I did. To see how you can hand ChatGPT's browser the job of filling your basket and skip every ad placement dotted along the purchase journey. It is very slow for now, but it may be the future of e-commerce.
Use of these tools is still marginal, but their spread would undoubtedly hit retailers' ad revenue, which today rests mostly on formats sold on-site: whether the flagship product, the sponsored product, which depends on searches performed within their site, or on-site display formats.
Because if product discovery moves upstream, into universal AI assistants, the ad budgets will follow - all the more so as most of these assistants have made no secret of their ambition to build contextual advertising offers. Direct impact, then, on all of retailers' on-site revenue (sponsored product, display, video), on which they margin at least 70%.
Impact too on their audience extension revenue: if a majority of retailers join OpenAI's Checkout programme, transactional data will no longer be the retailers' preserve. They will share it with the assistant, which can in turn monetise it with brands wanting to target those shoppers in classic media environments.
We will of course have to see how much consumers trust AI assistants. But they have no trouble buying directly inside Instagram or TikTok, so I see little reason they would not do the same in ChatGPT or Perplexity. Perplexity read that correctly, having integrated PayPal and Venmo to handle the transactional side.
Retailers would also - and this may be more insidious - be cut off from their ability to know their consumers' purchase intent, since the decision phase would no longer play out on their properties but inside the assistants. In those conditions it becomes impossible to sell "category intender" segments, as is widely done today, on and off-site.
The one bastion spared by the agentic wave would be - at least until the robots invade - in-store. A playing field grocery retailers have been exploring for years, notably through Mediaperformances, and one their non-food peers want to crack in 2026. The objective: diversify ad revenue beyond formats that depend on e-commerce and secure monetisation for the long term.
How far the agentic wave threatens retail media
Sponsored Product | On-site display | Audience extension | In-store | |
Threat | Huge | High | Moderate | None |
Why | The search phase moves upstream, into AI assistants, and ad budgets follow. Falling human traffic drags down available inventory. | The business will suffer from lower human traffic but will likely be less hit than SP. Formats can plausibly keep a mid-funnel role. | 1. Assistants, which share transaction data with the retailer, become potential competitors on this business 2. Retailers capture fewer and fewer intent signals and see their value proposition weakened | At least until AI robots doing our shopping become a reality |
There is urgency for retailers. Data published recently by Klaviyo showed that 56% of US consumers planned to use AI-based shopping assistants during Black Friday this year.
An Accenture study, which asked generative AI users in early 2025 about their preferred discovery sources, showed AI assistants (18% of responses) close behind the store (19%). Far ahead of retailer websites, which came seventh with only 10%.
So how do you prepare for the tidal wave? Retailers will have two options.
Option 1: build dykes
Some, like Amazon, can build dykes against the agentic wave, effectively blocking AI assistants from crawling their site, as The Information revealed.
That guarantees its products will not surface in external AI-powered shopping interfaces. Put differently, if a shopper wants to buy products sold on Amazon, they have to go to Amazon.
The risk is obviously losing sales by cutting yourself off from that business channel. But it is a controlled risk when you are Amazon, hold 40% of e-commerce market share and have become the default starting point for a shopping journey.
Amazon, which has also opened a legal fight with Perplexity to stop its Comet browser agents from shopping on its site, is one of the few retailers with the critical mass to take such an aggressive, closed approach.
The only other retailer that could afford it would be a hyper-niche site that has built a very strong direct relationship with its community and would rather sacrifice the audience and sales assistants bring in to refocus on that.
A bit like a niche media outlet abandoning SEO to concentrate on a direct relationship with its readers through a newsletter.
Option 2: ride the agentic wave
Walmart, hardly a minnow, made a different choice from Amazon: ride the wave. You have probably seen the partnership announced with OpenAI, which will let the retail giant's customers shop directly inside ChatGPT through Instant Checkout.
I think plenty will make the same pragmatic choice. Protect the growth of the flagship (retail), even if it means sacrificing a little of the agile speedboat that is retail media. While taking care, of course, to adjust the latter's economic model to the new paradigm.
Concretely how? One retailer sales house chief, met on the sidelines of the Retail Media Innovators event, told me he did not rule out that his role would eventually be "that of an agency helping brands gain visibility inside LLMs".
Here again, look at the media side. At Prisma Media and Reworld Media in particular, which, sensing the weakening of their on-site ad revenue, launched GEO offers to help brands optimise their visibility inside LLMs.
Retailers will certainly have a similar role to play, as I noted in an article detailing how Back Market and ManoMano are thinking about it. Their product knowledge, combined with their grasp of consumer expectations, can help. Whether that means optimising how LLMs perceive a brand - helping a challenger work on consideration around its differentiating criteria, for instance - or making sure their own stores are well perceived locally by LLMs.
We are getting close here to a local GEO logic of the kind developed by technologies such as Minddex.AI, whose platform lets retailers spot the geographic areas where their share of voice in LLMs sits below the national average, and act accordingly.
That obviously means staying alert to the new features and ecosystem partnerships ChatGPT, Perplexity and the rest will launch. It also means respecting a few fundamentals:
1. Welcome the LLM bots crawling the internet with open arms (OpenAI's goes by the sweet name of OAI-SearchBot)
2. Optimise your shopping feeds so assistants pick them up, and look at deploying MCP servers and A2A (agent-to-agent) protocols to smooth communication between LLM agents and e-commerce site agents.
The point is to make your catalogue compatible with Generative Engine Optimization and legible to AI, in order to capture discovery upstream. To structure product data and metadata so as to improve your chances of being mentioned in generative answers.
And, while you are at it, to turn that into a partnership and monetisation lever by offering brands ways to optimise their share of voice in universal assistants' feeds.
That is probably where retail media will come in. It is early days, but you can imagine freemium offers developing: the more you pay, the more the retailer ensures your product gets visibility inside LLMs versus its competitors'. With one ridgeline to walk: respect for the user experience and consumer expectations - you still have to surface what they are looking for.
A retail media tech chief told me he saw the rise of LLMs as a good growth lever for audience extension, with retailers launching offers that let brands work on their consideration with LLMs.
Concretely, that means using the retailer's insights on a product category or a consumer type to feed a full-funnel media plan that shapes how LLMs perceive a brand. LLMs which, it is worth remembering, take both paid media and organic into account when indexing. For now at least.
In a blog post, Michael Greene, head of platform strategy at Criteo, points out that "retailers adapted to the rise of D2C and price comparison sites, and they will do the same with the agentic era".
The threat is, it is true, also an opportunity to reinvent e-commerce discovery. Platform-native shopping agents can turn static product grids into personalised conversational experiences, solving choice overload and strengthening engagement.
Early tests show strong momentum: Amazon (again) is testing integrated offers with Rufus. So why not build these native AI assistants with a monetisation roadmap in mind?
Design them so that once usage is well established, you can integrate sponsored content and shoppable formats. Test placements, though, so they improve the purchase journey rather than disrupt it. Avoid "chatbot fatigue" by focusing on usefulness, speed and accuracy of answers. E-commerce basics, no?
Without necessarily going to war with the LLMs, you can also expect plenty of retailers to strengthen their loyalty programmes to guard against disintermediation and commoditisation.
The aim is to put their value proposition - what makes the difference in the consumer's eyes - back at the heart of such a programme. There are of course the tangible elements (price, availability, delivery speed) that AI assistants will have no trouble assessing, but there is also, above all, the intangible: everything that pushes a consumer towards one brand rather than another.
That can include exclusive experiences, personalised benefits or member-only access, creating emotional affinity and differentiation that go beyond the purely rational criteria of the algorithms.
