What DMEXCO 2026 tells us about brand discovery in the age of AI

Eight trends and five moves for marketing and e-commerce leaders, from two days of sessions in Cologne.

This year, DMEXCO chose “Scaling Intelligence” as its theme. Across stages, the conversation had moved on from what AI can do. The question in Cologne was how it creates value, and what changes when AI starts doing the searching, the comparing and, soon, the buying on behalf of customers.

We attended the sessions closest to our work: brand discovery, AI search, agentic commerce, measurement and marketing organisations. This article walks through what we heard. The full report brings it together in 13 pages, with every figure attributed to the session it came from.

The starting point: wide adoption, narrow impact

The opening session put numbers on a gap that ran through the whole event. According to DMEXCO and the German digital association BVDW, 89% of companies already use AI regularly. Only 37% report a measurable impact on operating profit (EBIT), and just 6% grow their EBIT by 5% or more thanks to AI.

What sets those 6% apart is less about tools than about choices. They redesign workflows around AI instead of adding it on top. They measure its impact in the P&L. They favour focus over endless pilots, make AI a leadership priority, and aim for growth rather than efficiency alone.

For marketing, the speakers were clear that the opportunity is real, as long as teams start from customer value rather than from the tool. Producing more content faster only matters if it meets a real customer need, or frees budget to spend somewhere better.

Source: “Scaling Intelligence in Numbers”, DMEXCO / BVDW opening session, DMEXCO 2026.

Eight trends we brought back from Cologne

1. Your next customer may find you without ever typing your name

Microsoft Advertising opened its session with a striking figure: automated AI traffic is growing eight times faster than human traffic. More customer journeys now start with an AI recommendation, and a growing share of customers may never search for a brand directly at all.

The session described three eras of the web that now coexist: the human web, where people do everything themselves; the LLM web, where they ask AI first and then act; and the agentic web, where AI acts for them. Customers move between all three, sometimes within the same purchase.

To be part of those AI-led journeys, a brand has to pass three tests:

• See. Can AI access what you offer? Your catalogue, feed attributes, metadata and pricing need to be easy to read.
• Understand. Can it match your offer to a specific need? Explain who the product is for, what it does well, and the trade-offs a customer might weigh.
• Recommend. Is there enough credible evidence to put you on the shortlist? Your own claims count, but so do reviews and third-party sources.

One practical consequence: the “fan-out” queries AI runs behind a single prompt rarely match a classic keyword list. Capturing that demand means moving further up the funnel, before shoppers type the obvious search.

Source: “How the AI Economy is Rewriting Brand Discovery”, Microsoft Advertising, DMEXCO 2026.

2. AI traffic now converts better than any other source

A year ago, traffic coming from AI tools converted 38% worse than other channels. In March 2026, it converted 42% better, according to Adobe Analytics data presented by ShopAgentic. AI-referred traffic to US retail sites grew by 393% between Q1 2025 and Q1 2026, and those visitors spend more time on site and view more pages than visitors from other channels.

The quality of that traffic is the other half of the story. ShopAgentic reported that ChatGPT delivers about twice the new-customer acquisition rate of search engines. Criteo added that 80% of customers coming through ChatGPT ads are new to the brand.

Buying itself is moving more slowly. 22% of shoppers have already bought inside an AI tool, while 50% have bought after researching with one. As ShopAgentic put it, the agent is a discovery channel first: checkout follows trust, it does not lead it.

Sources: “Amazon said no. Shopify said yes. What’s your answer?”, ShopAgentic (based on Adobe Analytics, April 2026), and “The Hidden Intelligence Behind AI Assisted Commerce”, Criteo, DMEXCO 2026.

3. Your product data is your new storefront

Several sessions described the same new type of customer. ShopAgentic called it the 2027 customer: an AI agent that does not see your design, reads extremely fast, takes everything literally, compares 200 shops in parallel and has no patience.

“Stop building an agent for your customers. Start building for your customers’ agent.”

Kai-Thomas Krause, co-founder, ShopAgentic

For that customer, product attributes matter more than marketing copy. Material, dimensions, compatibility, delivery time, return policy and stock are what an agent reads. The figures cited on stage show how far most brands still have to go: product pages score just 66% on machine readability, the lowest of any page type (Adobe), 86% of AI agent queries rely exclusively on the most-cited brand sources (Yext), and 25% of US e-commerce is expected to be agentic by 2030, a $500 billion market (Bain).

On, the Swiss sportswear brand, described product data as a strategic lever rather than a back-office task: brands that provide the richest context around their products are the ones that keep control of how AI talks about them.

Sources as cited at DMEXCO 2026 by ShopAgentic, On (“AI-ready product data”) and Criteo.

4. In China, AI agents already shop at scale

eTOC Germany offered a look at what agentic commerce looks like when it is already part of daily life. In China, 39% of search is now driven by generative AI, double the level of a year earlier, and around 80% of shoppers ask AI before they buy. More than 300 million agent payments have been made, each still confirmed by a human at checkout.

It works because agents live inside the apps people already use, with payment built into the same ecosystem. Merchants are moving from rule-based automation, where they tell the system exactly what to do, to outcome-based operations, where they set the goal and let AI find the steps.

The session was just as clear about the bill: walled gardens, low-price spirals, and brands made invisible by the platforms that own the customer relationship. Appetite also differs sharply: 84% of Chinese consumers are excited about AI products and services, against 38% on average in major European markets (Ipsos).

“China may build the fastest agents. Europe has the opportunity to build the most trusted ones.”

Moonie Zhu, CEO, eTOC Germany

Source: “The Invisible Transactions: China’s Blueprint for Agentic Commerce”, eTOC Germany, DMEXCO 2026 (citing Ipsos 2025/26 and China internet statistics 2026).

5. Humans set the strategy. Agents execute the buying.

PubMatic described a structural shift in how media is bought. Around two thirds of open-internet spend is now performance-led, and it grows about twice as fast as brand spend. According to the projections presented on stage, 30% of media buying could be agentic by 2028, and 50% by 2030.

The point is not a faster assistant for yesterday’s workflows, but systems redesigned for machine-speed decisions. Humans still set the strategy, the rules and the governance; agents handle the complexity of execution.

That changes what transparency means. When autonomous systems decide, advertisers need to know why a decision was made: which signals and rules applied, and whether it can be verified.

Source: “The Agentic Era of Advertising”, PubMatic, DMEXCO 2026. Projections as presented on stage.

6. Marketing now has two audiences: people and AI agents

Gamma’s Chief Growth Officer described a marketing team split along a new line. Every marketer has two jobs, informing and persuading. With AI agents as a second audience, informing becomes a strategic job of its own.

• Team Agent produces content and data that AI can process and repackage for people, such as guides written to be read by agents. It runs on the models’ clock, which is very short: what shapes AI answers can change from one month to the next.
• Team Human inspires and influences through events, creators, storytelling and real people. It runs on the clock of human memory, which is long: a great TV ad stays with people for years.

The result at Gamma: in one year, AI search went from barely visible to its first source of new users, outside word of mouth. The full-stack marketing generalist gives way to a “barbell” team, strong on agents and strong on humans.

Source: “Scaling Marketing in the Age of AI”, Gamma, DMEXCO 2026.

7. AI makes production easy. It doesn’t make ads effective.

Kantar looked at how AI-generated advertising actually performs. Consumers are split: 51% say they like seeing AI-generated content in ads, 49% do not. The more useful finding concerns effectiveness. There is an “AI effectiveness tax” only when AI is obvious: campaigns where AI shows are twice as likely to be ineffective.

Two examples made the point. One AI-made video ad was technically impressive but scored in the bottom 25% on branding: viewers said it could have been an ad for anything. Another, built on the brand’s established codes, reached the top 1% on branding and the top 5% on enjoyment. AI does not replace a strong creative idea linked to the brand.

The Bold CMO panel, with leaders from Decathlon, BMW and Lorenz Snacks, reached the same conclusion from another angle:

• Brand trust is what makes customers believe in your value and your price.
• AI is a co-pilot, never in the driver’s seat.
• ROAS still matters, but new-to-brand customers matter more.
• Community and third-party opinions are the best product marketing there is.

Sources: “The AI effectiveness tax? What CMOs need to know about GenAI advertising”, Kantar, and “The Bold CMO” panel (Decathlon, BMW, Lorenz Snacks), DMEXCO 2026.

8. In B2B, trust goes human

The People Branding Company shared what is working in B2B marketing and sales as AI-generated content floods every feed. The common thread: people, small rooms and real numbers.

• Amplified employee content. SAP Taulia boosted its best-performing employee posts with LinkedIn Thought Leader Ads, reaching 2.3 times the engagement of traditional campaigns.
• Small, exclusive formats. Away from big roadshows, toward intimate events for decision makers, with 25 to 40 peers in a room. Langdock’s AI forum brings together 70% customers and 30% new leads.
• Business cases over case studies. Decision makers want the ROI in their own language, from cost-saving calculators to spreadsheets for CFOs.
• Building in public. One company documented a merger in 25 candid LinkedIn posts, from the announcement to a 100-day recap, reaching around 185,000 impressions. In times of AI slop, that level of transparency signals there is nothing to hide.

Source: “How to Win in B2B Marketing & Sales in 2027”, The People Branding Company, DMEXCO 2026.

Five moves to make now

Several sessions ended with concrete actions. We grouped them into five moves, mostly adapted from the ShopAgentic keynote and the Microsoft Advertising session.

1. Decide your agent policy. Open, curated or closed? Decide which agents may read, which may buy, and on what terms. Name an owner for this channel: it is neither SEO nor marketplace.
2. Stop blocking your customers’ agents. Audit robots.txt, CDN and firewall rules against real agent user agents. Allow reading, define buying separately, and track AI visits as their own channel in analytics.
3. Turn adjectives into attributes. Start with your top 50 products. Move delivery times, shipping costs and returns into your product feed, out of the footer.
4. Measure your AI share of voice. Ask real buyer questions to ChatGPT, Gemini and Perplexity. Count how often you appear and how you are described, every week, and report it like you report paid search today.
5. Get protocol-ready. Ask your platform and payment provider for their roadmap on agentic commerce protocols (UCP, ACP), and pilot on part of your range.

Europe is roughly twelve months behind the US on these shifts. As one speaker put it, that is prep time, not a reason to wait.