Designing Websites for AI Agents, Not Just Humans

Picture a customer who never sees your site. They told an AI assistant, "Find me a plumber who can come Thursday and book it." The agent fans out across the web, reads a dozen sites in seconds, extracts hours and pricing and availability, and books the one it can understand and trust. Your beautiful hero video, your brand colours, your clever copy — none of it registered. The agent made a decision based purely on what it could parse.

This isn't science fiction; it's the leading edge of how a growing slice of commerce already works. And it introduces a question the design industry has never seriously had to ask: what does your website look like to a machine acting on a human's behalf?

Why agents change the rules

A human visitor is forgiving and intuitive. They'll squint at a confusing layout, infer that "Get in touch" means contact, and tolerate a price buried three clicks deep. An agent is the opposite: literal, fast, and unforgiving. If your hours live inside an image, the agent can't read them. If your pricing is "contact us for a quote," the agent can't compare you. If your booking flow needs a human to interpret it, the agent moves on to a competitor it can complete.

Agents don't reward persuasion. They reward clarity, structure, and honesty. The qualities that make a site accessible to a screen-reader user turn out to be almost exactly the qualities that make it legible to an agent. Accessibility just became a growth strategy.

What machine-legibility actually requires

1. Put critical facts in text, not images

Hours, prices, addresses, availability, specifications — if it influences a decision, it must exist as real, selectable text in the HTML. Text baked into a graphic is invisible to an agent. This one habit eliminates a huge share of "the agent skipped us" failures.

2. Use semantic, structured markup

Proper headings, lists, tables, and landmarks tell a machine what each part of the page is. Layer on schema.org structured data — Product, Offer, LocalBusiness, Service, OpeningHours — and you hand the agent a clean, unambiguous datasheet instead of forcing it to guess from layout.

3. Make pricing and availability explicit

"Contact us for pricing" is a dead end for an agent comparing options. Even a range, a starting price, or a clear "from $X" lets you enter the comparison at all. The same goes for availability: if a machine can't tell whether you're an option for Thursday, you're not an option.

4. Keep navigation predictable

Standard patterns, clear link text, and logical flows let an agent move through your site the way it expects to. "Click here" tells a machine nothing; "Book a Thursday appointment" tells it everything. Predictability beats novelty when your visitor is a program.

5. Be honest and consistent

Agents cross-check. If your site says one thing and your Google listing says another, the inconsistency reads as unreliability, and unreliable sources get filtered out. Consistent name, address, hours, and pricing across the web isn't just good hygiene — it's how an agent decides you're safe to act on.

Agents don't fall for marketing. They reward the businesses that are clear, structured, and telling the truth — which is a refreshing thing to optimise for.

The emerging standards worth watching

A few conventions are forming to make the agent web work better, and they're worth keeping on your radar:

• llms.txt — a simple file that gives language models a clean, curated summary of your site, much like robots.txt does for crawlers. A low-effort way to spell out who you are and what matters.
• Richer structured data — as agents transact, schema for products, offers, and services becomes the interface they actually read. Investing here pays off directly.
• Agent-facing endpoints — some businesses are beginning to expose clean, structured ways for agents to check availability or pricing, rather than making them scrape a human UI. Early, but the direction is clear.

A practical starting checklist

• Move every decision-critical fact out of images and into real text.
• Add structured data for your core entities (business, products, services, hours).
• State pricing as a number or range wherever you possibly can.
• Rewrite vague links and buttons into literal, descriptive actions.
• Audit consistency across your site, Google listing, and key directories.
• Publish an llms.txt and keep it current.

The same structure that wins here also wins citations in AI search — see our GEO guide — and fits into the broader picture in our 2026 field guide. Machine-legibility is quietly becoming the through-line of the entire AI-era web.

The agent web is still early — but the businesses preparing for it now will be the ones quietly winning sales that their competitors never even see happening.