Build for customers and their agents
Your next customer may send software to research your business before they visit it themselves. For established merchants, being ready means connecting the experience people see with the information and systems their agents need.
The customer is still a person.
The journey is changing. Imagine a procurement manager looking for a new supplier. Instead of opening a dozen tabs, they ask an AI assistant to compare options, check product specifications and prepare a shortlist. Your first interaction with that potential customer may now happen through software.
The underlying questions are familiar. Do you stock the right product? Can you supply our locations? What are your trading terms? Can you support our requirements? What changes is who asks, how the answer is assembled and whether your sales team ever sees the research happening.
A business can have an impressive website and still be difficult for an agent to evaluate. Important answers may sit in images, inconsistent product descriptions or documents disconnected from the catalogue. If the agent can't establish whether you're a fit, the customer may never reach the conversation where your team could explain it.
Our view is simple: build systems for customers and their agents. Keep the human experience strong, while making the business easier for authorised software to understand and work with.
The numbers, and what they measure.
There is evidence behind this shift, but different kinds of traffic shouldn't be treated as the same thing.
Imperva's 2025 Bad Bot Report put automated traffic at 51% of web traffic in its 2024 dataset. That includes malicious bots and other automation. It doesn't mean AI shopping agents have overtaken people. [1]
A more specific signal comes from Cloudflare. In September 2026, it reported that daily requests from AI agents on its network had grown by more than 1,700% over the preceding year. That is a network-level measure of requests, not a count of customers or completed purchases. [2]
Adobe measures something different again: people clicking through from AI sources. Its October 2026 report found that this traffic to U.S. retail sites grew 71% year on year in September. Those are AI-referred visits, not autonomous agents visiting or buying. [3]
These datasets aren't a forecast for Australian B2B commerce, and rapid growth doesn't establish a channel's share of sales. They do show why it's worth measuring AI-assisted discovery and agent access separately, rather than assuming every valuable journey starts with a person browsing your homepage.
Complex commerce is a systems problem.
For an established retailer, distributor, wholesaler or manufacturer, a useful answer often depends on more than published marketing content. Product suitability may involve technical attributes, pack sizes and compatibility. Availability may depend on a warehouse. Pricing may depend on a trade account, quantity or negotiated agreement.
An agent that can read a public product page hasn't automatically earned access to account pricing. An agent that can request a quote hasn't automatically been authorised to accept one. Making a business agent-ready means being precise about both information and permission.
That is why a chatbot alone isn't the strategy. The customer experience depends on what sits behind it: commerce, content management, product information, operational data and business rules. If those systems disagree, adding a conversational interface gives the disagreement another place to appear.
The opportunity is to connect what the business knows with what the customer needs, without exposing what the customer isn't entitled to see.
Make the business understandable first.
Start with a clear, dependable information layer. Explain what you sell, who it suits, its limitations and the next useful step. Give products consistent attributes and units. Keep delivery, returns and service information current. Separate confirmed facts from estimates and conditions.
Then make that information accessible in forms the relevant systems can use. Well-structured pages, readable text and appropriate structured data are practical foundations. For particular agents, plain-text or markdown representations can offer another route to the same information. The goal is consistent meaning across interfaces, not a second version of the truth. We explain one such format in why your website needs an llms.txt file.
There is no magic file that guarantees an AI recommendation. Google's guidance explicitly says llms.txt and special AI text files aren't required for its generative search features, and don't improve Google Search visibility. Useful, reliable content and sound technical foundations still matter. [4]
Helping an agent retrieve your information is a design choice. Claiming that a particular format will make every AI system recommend you is a promise no responsible business should make.
Our own website is a working example.
We've applied these principles to Matter's website, which runs on Matter Content Studio. A visitor sees the designed experience. An agent can also read plain-text versions of pages and use an llms.txt index to find relevant information. Service pages open with a direct explanation of what the service is, and case studies set out the platforms, integrations and work involved. [5]
The contact experience has a conversational entry point too. Matter's AI assistant takes initial notes before the team picks up the conversation. It is a practical example of moving from reading about a business to beginning an interaction with it. It is not evidence that every external agent can complete every step without difficulty.
The wider lesson is the architecture: a designed experience and a machine-readable representation serving the same business information. Good design still matters. So does making the facts beneath the design clear enough to use.
This is an example of how we approach the problem, rather than a claim that an agency website has solved the account permissions, pricing and fulfilment complexity of enterprise commerce.
Move from answers to actions, with boundaries.
Once information is reliable, choose where an agent should be able to act. A sensible first step might be finding a suitable product, preparing a quote request or checking an order after authentication. Each action needs a defined scope, the right permissions and an unambiguous result.
Reading a public specification and placing an order shouldn't carry the same level of access. Commercial commitments need the customer's authority. Sensitive data needs account-level controls. Uncertain requests need a route to a person. Recording what happened matters as much as making the interaction convenient.
The tools are evolving. Chrome's WebMCP work, introduced in early preview in February 2026, illustrates an approach in which websites expose structured actions rather than leaving agents to infer every interaction from a visual interface. That announcement is a direction of travel, not a reason to assume universal support. [6]
Avoid making your strategy depend on a single protocol winning. Clear data, controlled actions and connected systems remain useful as interfaces change, as we describe in what the Universal Commerce Protocol changes for brands.
Keep control of the capability you build.
Matter is a product-powered services business. We bring commerce, content management, data enrichment, design and digital strategy together around these problems. Matter Content Studio supports publishing for people and agents. Matter Data Studio connects business information, systems and rules, with access and approvals shaped around your requirements. Each Data Studio client runs its own licensed instance. [5]
Where they fit, these technologies connect the information customers discover with the systems that can support an answer or action. They aren't compulsory components of every project, and they don't require replacing every system a business already relies on.
For us, "AI you own" starts with a business question: where will the information, rules and capability live, and who controls how they're used? Ownership, licensing, portability and operating responsibilities should be explicit in the engagement. A slogan is no substitute for those decisions.
Start with one useful piece of the journey.
The case for acting now isn't that humans are about to disappear from commerce. It's that improving product information, content governance and system connections takes work, and those improvements can benefit customers and staff before agent-led transactions become a material channel.
Choose a real customer task. Test whether a person and an agent can find the answer, understand its conditions and reach the next step. Fix the information gaps, then connect one bounded action. Measure answer accuracy, successful handoffs, qualified enquiries and completed tasks. Keep request volume separate from commercial value.
For complex merchants, this is an evolution of digital strategy, not a separate AI sideshow. The business still needs to earn trust, explain its value and deliver what it promises. It now needs to do that through more than one kind of interface.
If you're considering what that means for your business, start the conversation. Bring a customer journey that is harder than it should be. We can look together at what your customers need, what their agents need and where a practical first improvement could make a difference.
Sources
- Imperva (Thales): 2025 Bad Bot Report announcement, 15 April 2025.
- Cloudflare: The Internet has a second audience, 30 September 2026.
- Adobe Digital Insights: AI traffic and U.S. retail, 29 October 2026.
- Google Search Central: Generative AI optimisation guidance, updated 10 July 2026.
- Matter: Matter Content Studio, checked 21 November 2026.
- Chrome for Developers: WebMCP early preview, 10 February 2026.