For most of the commercial web’s history, a website had two important audiences: people and search engines.
A third audience is arriving. AI agents can now research products, compare suppliers, assemble shortlists, fill shopping carts, and complete purchases with a person’s approval. Some buyers will still visit the website. Others may make most of the decision inside ChatGPT, Gemini, or another agentic interface.
This changes what website visibility means. A company can have an attractive homepage, strong brand photography, and reasonable Google rankings while remaining difficult for an AI agent to evaluate. The agent needs clear facts. It needs to know what is available, what it costs, who it is for, where it can be delivered, which conditions apply, and why the seller is credible.
If those facts are missing or contradictory, the agent has an easy alternative: recommend somebody else.
This is Part 1 of After the Chatbot, a four-part series on the AI changes small companies need to operationalize now.
Agentic commerce is already producing real traffic
Agentic commerce describes purchases in which an AI agent helps a buyer discover, compare, select, or buy a product or service. The agent may complete one step or manage most of the journey.
Shopify reported that AI-referred orders grew nearly thirteen times year over year in the first quarter of 2026. Referral sessions from AI chatbots grew more than eight times. OpenAI has expanded its Agentic Commerce Protocol from checkout into richer product discovery. Google’s Universal Commerce Protocol is designed to connect AI shopping interfaces such as AI Mode and Gemini with merchant systems from discovery through checkout and order management.
These systems are still developing. The commercial direction is clear enough to act on. AI interfaces are becoming another place where customers discover and evaluate businesses.
AI agents do not browse like people
A person can infer meaning from layout, images, tone, and context. An agent can use some of those signals, but it performs better when important information is explicit and consistent.
Consider a small manufacturer selling specialist components. Its product pages may use internal product names, omit delivery windows, hide technical details inside PDFs, and show different specifications in the catalogue and FAQ. A human buyer might call the company and resolve the ambiguity. An agent preparing a shortlist may remove the supplier because it cannot verify fit.
The same problem affects service companies. A consultancy may describe its work using broad language while leaving scope, price, delivery time, prerequisites, and expected outputs unclear. An AI system cannot confidently recommend an offer it cannot define.
Agent-ready content makes the commercial facts easy to retrieve:
- a precise product or service name
- a clear description of the problem it solves
- who it is suitable for
- price or a useful pricing model
- availability and delivery conditions
- prerequisites and exclusions
- evidence, reviews, case studies, or certifications
- support, returns, privacy, and cancellation policies
This information helps humans too. Machine-readability work often exposes the same vague copy and inconsistent data that can reduce conversion.
SEO and Generative Engine Optimization now overlap
Traditional SEO helps a page get discovered in search results. Generative Engine Optimization, often shortened to GEO, improves the chance that AI systems can retrieve, understand, cite, and recommend the company.
The disciplines share a foundation:
- crawlable pages
- descriptive titles and headings
- stable internal links
- original and useful content
- structured data
- consistent entity names
- trustworthy external references
GEO adds a stronger emphasis on extractable answers and verifiable claims. Because AI discovery products may use different sources and retrieval methods, companies should treat reviews, directories, industry pages, press coverage, and customer discussions as supporting evidence rather than assume the homepage speaks alone. Contradictory facts make any recommendation harder to trust.
Small companies should therefore treat their web presence as a connected evidence system. The website remains the main source of truth, but it needs support from consistent external information.
Product data is becoming marketing infrastructure
Product feeds used to sit behind advertising and marketplace operations. They now influence whether AI shopping systems can find and recommend an item.
A useful record needs more than a product name and an image. Agents benefit from structured attributes such as dimensions, materials, compatibility, use cases, certifications, stock, delivery regions, variants, price, and returns.
Service businesses need an equivalent structure. A clearly packaged service can expose delivery window, team size, who to involve, prerequisites, outputs, and next steps. That makes the service easier to compare without reducing it to a commodity.
This is especially valuable for a less familiar supplier that cannot rely on buyers already knowing its name.
A ten-point agent-readiness audit
Start with ten practical questions:
- Can an agent identify exactly what the company sells?
- Are product and service names consistent across the site?
- Can price, availability, and delivery conditions be found without opening a PDF?
- Do important pages contain specific facts rather than promotional adjectives?
- Are policies written in plain language and linked from the relevant commercial pages?
- Does structured data match the visible page content?
- Are reviews, case studies, and certifications easy to verify?
- Do external profiles describe the company consistently?
- Can an agent reach the next step, such as booking, enquiry, cart, or checkout?
- Is there a reliable way to measure traffic and conversions from AI sources?
An audit will usually reveal a mixture of content, data, technical, and operational work. Some fixes are small. Others expose a website that has become difficult for the company itself to manage.
Frequently asked questions
Does every small business need agentic commerce?
No. Companies selling standard products online will feel the change first. B2B services, specialist manufacturers, hospitality, and local businesses will still benefit from clearer machine-readable information because AI systems increasingly influence research and shortlisting.
Is an llms.txt file enough?
No. llms.txt is a voluntary proposal with uneven platform support. It may provide a useful index for systems that choose to read it, but it cannot repair vague pages, inconsistent facts, poor structured data, or weak external evidence.
Will AI agents replace websites?
They may reduce the number of visits needed before a decision. Websites will continue to provide authoritative product information, policy, proof, fulfilment, support, and a direct customer relationship.
What should a company fix first?
Start with the pages closest to revenue. Make the offer, price, audience, conditions, proof, and next step explicit. Then improve structured data and external consistency.
Prepare for a customer that reads differently
Start with the pages closest to revenue. Make the offer, price, conditions, proof, and next step clear enough for people, search crawlers, and agents to use without guesswork.
XYZ’s SEO Manager can turn this into a recurring SEO and GEO workflow. If the website structure itself is the constraint, an Agentic Website project can make the content and commercial data easier to operate.