Is Your Website Ready for AI Agents? Data From 829 Audits
AI agents will reach your website in two ways: by reading your pages, and by talking to systems through protocols. For almost every small business, the first route is the one that matters this year, and our audit data suggests most sites are not ready for it. In October 2026, Google opened a platform for apps inside Search and Gemini built on the Model Context Protocol, launched a cross-retailer shopping cart built on its Universal Commerce Protocol, and its CEO described agents that browse pages as well as call protocols.
Below we summarise what changed, what "agent-ready" means in practice for a business without an engineering team, and what 829 of our own automated website audits show about how far typical sites are from it.
What changed in October 2026
Three announcements in the space of two weeks set the direction.
Universal Cart (15 October). Search Engine Journal reported that Google's Universal Cart lets shoppers add items from multiple retailers into one cart across Search and the Gemini app, live in the United States. Checkout happens either natively with Google Pay or through a cart transfer to the merchant's own site using the Cart API in the Universal Commerce Protocol. In the article's words, "the merchant stays the seller of record either way."
Two routes for agents (22 October). Search Engine Journal reported Sundar Pichai's view that agents will access the web through both browsing and protocols: "I think you have both ways and they're going to develop." He described browsing as "a very elegant way to work too, because that's what you as a user would do", noted "a lot of momentum behind MCP or agent protocols or payment protocols", and saw protocols as the more scalable but slower route.
Connected Apps (28 October). Google announced a preview platform that lets developers bring their services into Gemini and Search. Developers connect through "a cloud-hosted MCP server or Android MCP using the AppFunctions APIs." Google said it had launched more than 50 Connected Apps, with partner integrations live in Gemini and expected in Search within weeks, and access opening "in phases".
What agent-ready means for most businesses
The protocol route (MCP servers, commerce APIs, partner programmes) is for businesses with developers, product catalogues and integration budgets. It is worth watching, and for retailers on platforms that support these protocols natively, worth enabling. For a plumber, an accountant, a dental practice or a regional manufacturer, it is not this quarter's priority.
The browsing route is different. An agent asked to "find a local electrician who can come on Thursday and check their prices" has to read websites in much the way a person or a search crawler does. That makes agent-readiness, for most businesses, an extension of good technical SEO rather than a new discipline. A browsing agent needs:
| Requirement | Why it matters to a browsing agent |
|---|---|
| Pages that load quickly | Agents work under time and cost budgets; slow pages are slower to read and more likely to be abandoned |
| Content in the HTML | Text that appears only after heavy scripts run is harder and more expensive to extract |
| Clear structure | Headings, labelled images and consistent navigation tell a machine what each part of a page is |
| Crawl access | robots.txt rules and sitemaps decide what automated systems can find |
| Explicit facts | Prices, service areas, opening hours and contact details stated as plain text, not implied |
| Machine-readable descriptions | Structured data describes the business, services and products in a standard format |
None of these is speculative. They are the same properties that help Google's crawler, screen readers and human visitors. The difference is that an agent acting for a customer may be the only "visitor" that reads the page before a decision is made.
What our audit data shows
Between 21 February and 16 July 2026, our automated audit engine produced 829 free reports covering 799 distinct domains, all websites we audited. Each report combines a crawl of key pages with Google's PageSpeed Insights tests. The figures below are aggregates; no individual site is identifiable. Where a report could not take a reading, it is excluded from that line.
| Signal | Share of sites affected |
|---|---|
| Homepage slower than 3 seconds (739 with a reading) | 60% |
| Images without alt text on a crawled page | 72% |
| At least one crawled page with no structured data | 65% |
| Missing meta descriptions | 61% |
| Thin content on a key page | 59% |
| Missing canonical tags | 49% |
| Missing H1 heading on a crawled page | 39% |
| Sitemap problems | 39% |
| robots.txt problems | 24% |
| Mixed HTTP/HTTPS content | 22% |
| Broken links | 10% |
Read against the agent requirements above, the pattern is clear.
Speed is the most common barrier
The median homepage in our data took 3.9 seconds to load, and 60% took longer than 3 seconds. In Lighthouse's mobile lab test, 91% of the sites with data failed Core Web Vitals, almost entirely on Largest Contentful Paint. Lab tests simulate a mid-range phone on a throttled connection, so real-world figures can be better, but the direction is unambiguous. An agent that has to wait four seconds for a page, and then for scripts to finish, is paying a cost every time it reads that site.
Most pages describe themselves poorly to machines
Seventy-two per cent of sites had images without alt text on at least one crawled page, 39% had a page with no H1 heading, and 65% had at least one crawled page with no structured data at all. Each of these removes a label a machine could otherwise use. An image of a price list without alt text is invisible to any system that reads text. A page without a main heading leaves an agent guessing what the page is about.
Thin pages leave questions unanswered
Fifty-nine per cent of sites had thin content on a key page. Short service pages often omit exactly the facts an agent would need to complete a task: what is included, where the business works, how pricing is structured and how to book. If the answer is not on the page, an agent acting for a customer moves on to a site where it is.
Access rules are a quieter risk
Around a quarter of sites had robots.txt problems and almost two in five had sitemap problems. Most were not blocking anything deliberately. But misconfigured access rules affect every automated visitor, and as AI providers introduce separate training, grounding and search controls, a robots.txt file nobody has reviewed in years becomes a bigger liability.
An agent-readiness checklist
This is the order we would work in for a typical small-business site. None of it requires a developer to build anything new.
- Fix load time first. Compress and resize images, enable server compression, and defer scripts that are not needed for the first screen. Our Core Web Vitals optimisation guide covers each metric.
- Make sure key content is in the HTML. View the page source of your main service pages. If prices, service descriptions or contact details are missing from the source and appear only after scripts run, raise it with whoever built the site. Our technical SEO audit guide shows how to check rendering.
- Label everything. One clear H1 per page, logical H2 sections, and descriptive alt text on every meaningful image, especially images that contain text such as price lists or menus.
- State the facts plainly. For each service: what it is, who it is for, where you provide it, what affects the price and how to book. Put opening hours, phone number and address in text, not only in an image or a map embed.
- Add structured data. Google's introduction to structured data explains how markup helps it understand a page. At minimum, describe your organisation or local business, and your products or services where relevant.
- Review robots.txt and your sitemap. Confirm you are not blocking pages you want found, that the sitemap is valid and current, and that you have made a deliberate choice about AI training and AI search access rather than inheriting an old default.
- If you sell online, check your platform's agentic commerce options. Retailers on platforms that support Google's commerce protocols may be able to enable them through settings rather than development work.
How agent-readiness connects to search visibility
The work above is not a bet on a single technology. Each item already improves classic search performance and AI search visibility. Ahrefs found in October that pages ranking in Google were about seven times more likely to stay cited in AI Overviews and AI Mode for at least a week, a correlation that suggests the foundations overlap heavily.
That overlap is why we treat agent-readiness as part of technical SEO rather than a separate service. It is also why the timing matters: Google's dashboard shows core updates starting in November in both 2023 and 2024, and our timeline of confirmed Google algorithm updates tracks every confirmed update since. Sites that fix their foundations now benefit whichever way search develops.
A ten-minute self-test
Before commissioning any work, a business owner can get a rough picture of agent-readiness in ten minutes:
- Ask an AI assistant to do a real task. For example: "Find a [your service] in [your town] and tell me their prices and when they are open." Note whether your business appears, and whether the details it gives are correct.
- View the source of your main service page. In most browsers, right-click and choose "View page source". Search for your price, your phone number and your service area. If they are not in the source, a browsing agent may not see them either.
- Run PageSpeed Insights on your homepage on mobile. Note the Largest Contentful Paint figure. Over 2.5 seconds is the threshold Google calls "needs improvement".
- Open your robots.txt file at yourdomain.com/robots.txt and check that nothing important is disallowed.
- Paste a key page into Google's Rich Results Test to see whether it carries any structured data.
If two or more of these checks fail, the checklist above is the place to start, and the first two items will usually deliver the most improvement for the least effort.
If you want to see where your own site stands on these signals, you can run a free audit with just your domain, or check whether AI assistants already mention your business with our AI visibility check.
Key takeaways
- In October 2026 Google launched Universal Cart on its commerce protocol and an MCP-based Connected Apps platform, and its CEO said agents will both browse pages and use protocols.
- For most small businesses, the browsing route matters now: agents need fast, clearly structured pages with facts stated in text.
- Across 829 of our audits, 60% of homepages took longer than 3 seconds, 72% of sites had images without alt text and 65% had a page with no structured data.
- Agent-readiness is mostly good technical SEO: speed, content in the HTML, clear labels, explicit facts, structured data and deliberate access rules.
- The protocol route (MCP, commerce APIs) is worth watching, and worth enabling where your platform supports it, but it is not where most businesses should start.
Where does your own site stand?
Run a free audit: enter your domain and get a dated, sourced record of what is publicly visible about your site. No login, no card. Prefer to talk it through? Book a consultation with our team.
