Is your website ready for AI agents?

Telstra scores a perfect 100 out of 100 on Google’s new agent readiness check. So do AGL and Origin Energy. None of the three exposes anything an AI agent can use.

We found that by running Google’s own audit across 181 Australian brands in August 2026. The full data, the sector tables and the method are in our agent readiness benchmark. This piece is the practical version: what the score means, and what to actually do about it.

What is Google grading, exactly?

In May 2026 Google added an Agentic Browsing category to Lighthouse, the auditing tool built into Chrome, and PageSpeed Insights picked it up a fortnight later. It runs six checks. Three cover WebMCP, the new browser standard that lets a page offer an AI agent typed, callable tools instead of making it read the screen and guess. The other three cover page structure, layout stability and whether you publish an llms.txt file.

The idea is sound. An agent that can call a function to check your stock gets a true answer from your server. An agent that scrapes your page can read a stale number and tell a customer something wrong, which makes your business the source of a bad answer.

Why does the score say yes when the answer is no?

Because a missing tool is not treated as a failure. If your site has no agent tools, Google decides those checks do not apply to you, drops them from the report card, and grades you on what is left. For 166 of the 175 sites we scanned, that meant all three agent checks disappeared and the entire score came from page structure and layout stability.

It gets worse than merely flattering. The nine sites in our sample that genuinely are agent ready mostly score 75%, because building real tools switches the missing checks back on and gives you more ways to lose points. The businesses doing the work score below the businesses doing none of it. The number also moved an average of 26 points when we re-ran the same sites hours later, so it is not stable enough to report month over month either.

So how do I tell if my site is actually ready?

Ignore the score and check three things directly. None of them takes longer than a few minutes.

1. Check whether your platform already did it. Every agent ready site in our Australian sample was a Shopify store, and not one of those merchants built anything. Shopify ships ten tools by default, including add to cart and proceed to checkout. If you are on Shopify you are probably already exposed, and you should know that rather than discover it.
2. Load your homepage and look for llms.txt. Visit yourdomain.com/llms.txt. Only 30% of the brands we tested have one. If you get a 404, that is the cheapest item on this list.
3. Run the Lighthouse check and read the individual lines, not the total. The total is the misleading part. The page structure check is the one that matters most today, and 77% of the brands we tested fail it.

What should I actually do about it?

In this order, because the cheap and useful work comes first and the speculative work comes last.

1. Fix the page structure check. It is the same work as fixing accessibility, it helps human visitors and screen readers as well as agents, and it is the single most commonly failed check in our data.
2. Publish an llms.txt. It takes an afternoon and most of your competitors have not done it.
3. Find out whether AI agents are actually arriving on your site. If nothing is coming, building tools for them is premature.
4. If they are arriving, look at what they are trying to do. A booking request an agent cannot complete is a lost customer, not a technical curiosity.
5. Only then consider exposing tools, and start with read only ones. Anything that changes state, takes payment or makes a commitment deserves a human confirming it.

Step three is where most businesses stall, because standard analytics cannot tell an agent from a person. We covered why that measurement is hard, and how to start doing it by hand, in how do Australian businesses track AI search citations. The same logic that decides whether you get cited in the first place is covered in how AI decides who to cite.

Where these numbers come from

Every figure quoted here comes from our own scan of 181 Australian brands on 20 August 2026, using Lighthouse 13.4.1 on Chrome 151. The full findings, the sector by sector breakdown, the named sites and the limitations of the method are published in the Agent Readiness Benchmark.