Your Website Has a Third Visitor Now — And Most Sites Aren't Built for Them

For twenty-plus years, a website really only had two audiences: the people looking at it, and the search engines crawling it to decide where it ranks. Every good web designer built for both — a site that looks great and reads cleanly to Google.

There's a third visitor now, and it's not here to look or to rank you. It's here to do something. Claude in Chrome, Perplexity Comet, ChatGPT's agent mode — these tools can now open your site, read what's actually on the page, click a button, fill out a form, and complete a task, all without a human touching the mouse. I covered what these tools can actually do — and where it's already gotten weird — in this post on AI browser agents. This one is the flip side of that question: is your website actually ready to be used by one?

For most small business sites right now, the honest answer is no. Not because anything's broken exactly — the site loads fine, a human can use it without thinking twice. It's that nobody built it with a machine trying to act on it in mind, and that's a different bar than looking good or ranking well.

Two Different Jobs: Being Found vs. Being Usable

It's worth separating two things that get lumped together. I wrote about how to show up when someone asks ChatGPT "who's the best [X] near me" — that's about being cited in an answer, a text-generation problem. This post is about something else: what happens once an agent has already decided to visit your site and try to use it — book the appointment, fill out the quote form, compare your price against a competitor's. That's an action-completion problem, and it breaks in completely different places.

What Actually Breaks an AI Agent on Your Site

An agent "sees" a webpage the way a screen reader does — not as a picture, but as a structure of text, labels, and interactive elements. A few of the most common ways that structure falls apart:

Buttons and links with no real label

A button that just says "Click Here," or an icon-only button with no text alternative, tells a human almost nothing either — but a human can guess from context. An agent parsing the DOM sees a button with no accessible name and has to guess blind, or just gives up.

Forms with no actual field labels

Placeholder text that disappears the moment you start typing looks fine to a person who can see the field before they click into it. To an agent reading the page's structure, a field with no bound <label> is just an anonymous text box — it can't reliably tell if it's asking for a name, an email, or a phone number.

Content that only exists after JavaScript finishes doing something clever

If your key content or your form only appears after a complex client-side render, and there's no clean fallback in the actual HTML, an agent that reads the page quickly can hit a blank shell where your content should be.

Information trapped inside an image

Same problem I flagged in the AI-search post — your hours, your price, your service area baked into a graphic instead of real text. A human sees it instantly. A machine sees a file it can't read.

The Overhyped Fix: llms.txt

You'll see a lot of noise right now about llms.txt — a plain text file some sites are adding that's supposed to summarize your site for AI tools, kind of like a sitemap written for a language model instead of a search crawler. Here's the honest state of it as of today: adoption sits around 10% of sites, and according to Ahrefs data, 97% of llms.txt files receive zero bot requests. As of early 2026, none of the major AI companies — not OpenAI, not Google, not Anthropic, not Meta, not Mistral — have publicly committed to actually reading or acting on it in their production systems.

That doesn't make it worthless — it's a low-effort experiment, and standards like this sometimes take a couple of years to get real adoption. But if you're choosing where to spend an afternoon of dev time right now, llms.txt is not it. It is not a substitute for robots.txt, for real structured data, or for clean HTML — it's a nice-to-have on top of those, not instead of them.

What Actually Works (and What We Already Build In)

The good news, same as with AI search: the fixes that make a site agent-ready are almost entirely the same fixes that make it accessible and well-built in the first place.

• Real structured data. Schema.org / JSON-LD markup that labels your business name, hours, services, and pricing as machine-readable facts — not something to re-explain here since I covered it in the AI-search post, but worth repeating that this is the single highest-leverage technical fix and every site we build ships with it from day one.
• Semantic HTML and proper labels. Real <label> elements tied to form fields, descriptive button and link text instead of "Click Here," and landmark regions (nav, main, footer) that tell any reader — human, screen reader, or agent — what part of the page they're looking at. This is the exact same foundation as real WCAG accessibility work — a site built to be genuinely accessible to a person using a screen reader is, almost by accident, already most of the way to being agent-ready, because both are reading the same underlying structure instead of the visual layout.
• Content that exists in the actual HTML, not just after a JavaScript render finishes — server-rendered or pre-rendered content so there's always something real for a fast-reading agent to find.
• A deliberate robots.txt. The real AI crawlers worth knowing by name are GPTBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot, and Google-Extended. Whether you want them crawling your site to train models, or you'd rather they only pull content to answer a live question, is an actual decision worth making on purpose in your robots.txt — not something to leave on default and forget about.

Quick Fixes, Ranked by Actual Effort

If you're deciding where to start, here's an honest ranking of effort versus payoff — not everything on this page needs to happen the same afternoon:

• Low effort, do it this week: add real alt text to images that carry information, replace "Click Here" and icon-only buttons with descriptive labels, and make sure every form field has a bound <label> instead of relying on placeholder text alone.
• Medium effort, worth a dedicated afternoon: add or audit your JSON-LD structured data (business info, services, FAQ, reviews), and write a deliberate robots.txt that names the AI crawlers explicitly instead of leaving it on default.
• Higher effort, worth planning for: if your site leans heavily on client-side JavaScript to render its main content, look at server-side rendering or at least meaningful fallback content, so a fast-reading agent — or a slow connection — isn't looking at an empty shell.

A 5-Minute Test You Can Run Right Now

You don't need any special tools for this. Open your own site and try viewing its page source, or use your browser's reader mode. Ask yourself: is your phone number, your service area, and your main call-to-action actually there in plain text — the kind you could select and copy? Can you tab through your contact form with just a keyboard, in a sensible order, with each field clearly labeled before you type into it? If either answer is no, that's not just an agent problem — that's a human-with-a-screen-reader problem too, and it's the same fix either way.

Think about what that actually looks like for a real local business. Someone asks their AI assistant to "find a Bluffton contractor who does kitchen remodels and book a consultation for next Tuesday afternoon." A well-built site hands that agent a clear services list, a real booking form with labeled fields, and a working calendar link — the agent completes the task in seconds. A site where "Book Now" is an icon with no label, sitting inside a form with three unlabeled text boxes, just fails silently. The customer never even knows your business existed as an option, because the agent moved on to a site it could actually operate.

My Take

Right now, AI agents are still a small slice of the traffic hitting most small business sites — this isn't a five-alarm fire the way, say, mobile-first design was a decade ago. But the pattern is exactly the same one we've seen with every previous shift: the businesses that quietly get the fundamentals right — real HTML, real labels, real structured data — end up ready for whatever reads their site next, without having to scramble. The businesses chasing whatever this month's trendy fix is (llms.txt, right now) usually end up doing real work twice.

Build it right once — for a human, a search engine, and an agent, all three at the same time — and you don't have to keep rebuilding it. That's been true of every site we build from day one, and it's exactly why it holds up as the internet's audience keeps adding new members.

Curious whether your current site would actually survive that 5-minute test, or want a second set of eyes on it? Get in touch and I'll give you an honest read.