Agentic Commerce: What companies need to know in 2026

Hardly any other term is currently used as frequently and rarely explained as clearly as Agentic Commerce.

Time for a classification: What is already possible today? What is still the future? And where should retailers and manufacturers start now? We look at current examples from the market, share our assessment from practice and answer the questions that companies are currently most concerned with.

What is agentic commerce?

AI assisted commerce

AI supports people in their purchasing decisions. It makes recommendations, answers questions and compares products - but people decide and buy themselves. Conversational commerce, product search via ChatGPT or Perplexity, AI-based style advice: all AI assisted commerce.

Key feature: the human remains in the loop. The AI assists, it does not act.

Agentic commerce

Here, the user gives their authority to act to the AI. The AI searches, compares, decides and buys - autonomously, within predefined framework conditions. In the best-case scenario, the human only receives the confirmation message: "Done."

This is the decisive difference to classic process automation: no rigid trigger, no fixed rule. Instead, complex considerations, e.g. about price, quality, delivery time and availability - across systems, in real time.

Key feature: AI has genuine decision-making autonomy. The human delegates, not just the execution, but the purchase decision itself.

Agents for Commerce

This is the inward-looking perspective: AI agents that automate processes on the sales side - in ERP, in PIM, in the store backend, in customer service. No end customer sees this directly, but it changes how commerce organizations work internally.

Status quo: What is available today - and in what form?

The McKinsey report provides a clear figure: 84% of European consumers already use AI tools in their everyday lives. 63% of them also use these tools in their shopping journey - for research, comparison and inspiration.

At the same time, only 30% would trust an AI to complete an order automatically - even if the budget was limited in advance.

That's the real tension: the technology is partly there. The acceptance is not yet there.

What works today

• Conversational commerce via ChatGPT, Perplexity, Google Gemini: product research, comparison, recommendation
• Multimodal search: upload a picture, find outfit ideas or similar products
• AI chatbots in your own store: product advice, FAQ, availability queries
• Personalized recommendations in real time based on behavioural data
• Internal agents: text generation, image optimization, returns forecast, dynamic pricing
  

What does not yet work (in Europe)

• Autonomous checkout by an AI: Technically possible to some extent, not legally permitted in the EU
• AI-controlled payment: data protection, PSD2, EU AI Act stand in the way
• Fully autonomous purchasing decisions without human approval in the B2C context

B2B vs. B2C: where is the greatest potential?

B2C: Agentic commerce begins where trust is less critical

In the consumer sector, technology is ahead of trust. The desire to make purchasing decisions oneself is deeply rooted.

This does not mean that B2C agentic commerce will not come. But it will take hold in low-involvement purchases first: Repeat orders, consumer goods, standard products without emotional attachment.

B2B: Better conditions for autonomous purchasing processes

The situation is fundamentally different in the B2B context:

• Ordering processes are often repeatable and rule-based
• Automation has been normal in ERP systems for decades
• Purchasing decisions are rational, less emotional

This makes B2B purchasing perfect for agents because the organizational trust in automation has already been established.

A concrete scenario: an AI reads the CRM, identifies 1,000 customer contacts and orders Christmas cards. Not a huge technical effort, but a real process gain.

Assessment: Agentic commerce will arrive earlier and more widely in B2B than in B2C. The technology is similar - but acceptance in the business context is already much more advanced.

What will change from a retailer's perspective?

The good news first: it's not a revolution. Nobody has to start from scratch. But there are clear shifts that you should prepare for.

The customer journey is changing - but not dramatically

Awareness, discovery, consideration: these phases are increasingly being shaped by AI tools. Customers no longer necessarily come to your website via Google - they may come via a ChatGPT recommendation, a Perplexity response or a Google snippet that pulls directly from the product feed.

This changes how you have to think about visibility. And it changes which data points suddenly become important.

Brand takes on a new dimension

How does AI perceive my brand? What signals does it draw from the web? Which products appear in its recommendations - and which don't?

In the webinar experiment, ChatGPT favored products from About You - without any specifications. The reason: About You has the "most comprehensive, structured product catalog" known to the model. Branding in the AI age starts with data quality.

Your own store remains relevant

A frequently asked question: Do we still need the store if AI tools take over discovery? The answer: Yes - but for different reasons than before. The store will become less of an entry point and more of a contact point for customers who have already made a decision. And still important for brand presentation, after-sales and customer loyalty.

Visibility and SEO: what is important now?

SEO remains relevant - but the mechanics are changing.

How AI tools obtain data today

Not all LLMs crawl the web in real time. OpenAI/ChatGPT, for example, in many cases simply queries Google - and aggregates the first organic hits. Anyone who is not visible on Google is also not visible to ChatGPT.

Other providers such as Anthropic or Perplexity actively crawl the web. But even there, being crawled does not automatically mean being taken into account. Between the collection of data and the actual inclusion in a result, there are validation steps, caching and model decisions that are difficult to comprehend from the outside.

What this means

• SEO visibility remains the basis: if you cannot be found organically, you will not be recommended by AI either
• Content quality beats quantity: Purely AI-generated texts rank worse. Human-written or at least assisted texts perform better - especially for top 5 rankings
• Question-answer structure is becoming more important: AI understands and prefers content that is structured like answers to specific questions
• Structured product data: Machines read differently than humans. Those who structure their product catalog well are cited more often
• Observe the knowledge cut-off: All language models have a training date. Keeping your content up to date is not a nice-to-have

Practical tip: Write content in such a way that it answers specific user questions. Users formulate their research in AI tools as a question, for example: "Which hiking boots are waterproof and under 150 euros?" It is precisely this question-answer mechanism that is made more relevant by LLMs.

Agents for Commerce: a brief outlook

Agents for commerce - i.e. AI agents for internal processes on the retailer side - are a topic in their own right and deserve a separate article. So here are just the key points:

• There are potential applications along the entire e-commerce value chain: Text generation, image optimization, returns forecasting, dynamic pricing, fraud detection, customer service
• Real-life example: bonprix has reduced its returns rate by 2 percentage points thanks to an AI agent for size recommendations - a measurable, immediate effect
• Systems such as Intershop, Shopify and Commerce Tools are already building agent capabilities into their platforms
• Multi-agent setups (keyword: OpenClaw/Moldbot) are powerful but complex - and require a good API infrastructure, budget and patience

Recommendations: What should you do now?

Start immediately

• Structure product data and make it machine-readable - for feeds, for AI, for search engines

• Convert content to question-answer logic: How would a customer ask the question to which this text is the answer?

• Ensure SEO visibility - it remains the basis for AI visibility

• Identify and pilot internal use cases: Returns forecasting, copywriting, customer service - start small, learn, scale

Prepare for the medium term 

• Build or upgrade API infrastructure: No access for external agents without API

• Test conversational commerce interface

• Optimize and play out product feeds for Google and OpenAI - also think about new protocols such as UCP or ACP

• Actively think about the brand in AI channels: What should the AI know about my brand - and where does it get it from?

Wait and see

• Agentic checkout for end customers: Technically interesting, not yet legally possible in Europe, user demand still unclear
• Hand over checkout control: No first mover advantage recognizable. Retain control over customer access for as long as possible

Organizational

• Create a culture of error: AI experiments need room to fail - this is not a risk, this is the prerequisite for learning
• Equip employees with tools and encourage experimentation - otherwise uncontrolled shadow AI use will result
• Communicate clearly: What may be done with AI, what not, within what framework

FAQ

Is Agentic Commerce even legally permissible in the EU context?

As things stand today: No - at least not in the form of autonomous checkout. The EU AI Act, the GDPR and PSD2 each have their own requirements for autonomous purchasing decisions. This is one of the main reasons why Perplexity's Buy-with-Pro feature is not yet available in Europe. It remains to be seen how the legal situation will develop - but no rapid changes are expected.

How does the AI decide which store to buy from?

In a test, products from one provider were preferred because the product catalog was particularly extensive, clearly structured and easy to evaluate. This is exactly what makes the difference: if product data is complete, comprehensible and up to date, the chances of being selected increase.

For retailers, this means that data quality becomes a question of visibility.

Will the wider retail community be prepared to hand over the checkout to Agentic Commerce?

It will probably be a commercial decision: How much revenue will I miss out on if I don't participate? The parallel with Amazon is not inappropriate - those who sell there have largely given up direct customer contact and are primarily logistics and price providers. Similar dynamics could arise.

What is GEO (Generative Engine Optimization) - and do I need to tackle it now?

GEO is the term for optimizing content specifically for AI models. The basic principles are not new: clear structure, question-answer logic, good readability for machines. This has already been hinted at with voice search. The practical start: start by formulating your most important content in the way your customers would ask questions.

From classification to implementation

Agentic commerce does not start with the autonomous checkout. It starts with clean data, connectable systems and clearly prioritized use cases. Diconium supports companies in laying precisely these foundations - and in using agent-based AI where it delivers real added value in commerce.