Your next customer might be an AI agent

This week, an important issue emerged in the insurance industry. An executive pointed out that Meta’s Muse shopping agent browses insurance websites without any indication of being an agent — something that websites cannot detect using robots.txt. It retrieves quotes but strips away crucial details that matter legally, such as coverage limits, deductibles, eligibility conditions, and state-required disclosures. Unlike purchasing a T-shirt, buying insurance is context-dependent. When an agent simplifies the information, the “quote” that reaches the customer is not a true representation.

This observation is not unique. Marketers are increasingly facing a challenge: when you ask AI tools like ChatGPT or Claude to compare your product with a competitor, they often get critical details wrong. Customers are now describing their needs in terms of specific outcomes — such as “waterproof hiking shoes for Scotland, wide fit, under €180, delivered by Thursday” — rather than navigating through user interfaces. In the payments sector, there’s an ongoing debate about whether agent-led commerce prioritizes payment methods and who ultimately owns the customer relationship. Across these discussions, one theme is clear: agents are starting to search for, compare, and purchase on behalf of customers, yet many businesses are unaware of how they fit into this evolving landscape.

01The gap we’re facing isn't technical; it’s about clarity.

Your website was designed to engage human visitors and rank well on Google, but agents look for different information. They seek structured, trustworthy facts about your offerings, prices, terms, and limits. When this structured layer is missing, agents must improvise. As the insurance example illustrates, important information that is necessary for a regulated purchase can easily get lost in this improvisation.

This is not a call for nostalgia about SEO. Search engines rank your site, while agents represent buyers. Excelling in search rankings and being clear to a buyer’s agent are distinct disciplines, and only one of these has a decade's worth of playbooks.

02We treated our own site as the test case

We decided to use our own website as a test case. Since we provide this kind of service, we made our site the pilot project. We created a llms.txt index of our content, guidance for visiting agents, and machine-readable signals in robots.txt. We then developed a measurement tool; there is a free checker available at /tools/agent-ready. You can type in a URL, and it evaluates what a logged-out agent can see — covering fifteen checks across four categories, from discovery basics (robots.txt, declared sitemap, real metadata) to content that agents can utilize (markdown negotiation, llms.txt index) and stated bot policy, as well as the agentic surface: the emerging protocols (MCP, A2A, signed bot authentication) that enable agents to take action rather than just read. The tool also provides suggested fixes in a format that your coding agent can implement. The last category acknowledges that this is still in its early stages; failing these checks today is normal, and publishing them indicates that agents are welcome.

03The five-minute version

First, identify your most important product and ask two AI assistants to compare it against a competitor. Record any inaccuracies; these represent the hallucination risk every shopping agent carries about your product.

Next, run your domain through the checker to see what machine-readable elements are missing.

By comparing these two results, you will identify the gap: the story agents tell about you, and the facts they cannot verify. Addressing this gap involves unglamorous work — creating structured content, clear terms, and agent-facing signals — and requires re-evaluation each quarter. This work is essential.

Remember, your next customer could be an individual, but increasingly, their first interaction may be through an agent.

Sources include the insurance case regarding Muse, the challenge of using AI for product comparison, the trend of outcome-focused shopping, and the debate surrounding agent-led payments — all observed on LinkedIn between September 23 and September 24, 2026.