Case studies
What actually changes when a website is made readable to AI agents. Each study below is grounded in something published, so you can read the original and check our reading of it.
None of these companies is a customer. We collect them because the evidence is public and testable, which is worth more to a skeptical buyer than a testimonial. Where a finding cuts against something we sell, we say so on the page rather than leaving it out.
Software • SerpApi
Why software company SerpApi decided to use llms.txt, and their results
SerpApi published a clear argument we share: llms.txt helps AI agents understand a website faster and more accurately once they arrive. Their write-up starts from a developer scenario — an agent told to “use SerpApi to search Google” has to work out what the company is, find the right endpoint, and learn the parameters — and treats that as a documentation problem. The same shape applies to any business an agent is asked to evaluate.
Why it matters here: llms.txt is one of the 13 files we install, and it is the file most often dismissed as optional. This is a company that adopted it deliberately, published the reasoning, and measured Markdown against HTML.
Fintech • Ramp
What Ramp learned when they marketed to AI agents
In early 2026 Ramp published a realization we agree with: “Every company’s website, including ours, was built for humans. Marketing to agents is going to become just as important as marketing to humans.” They tested marketing to AI agents across roughly 50 pages and found Markdown was the only format that reliably surfaced in LLM responses.
Why it matters here: it is a controlled test by a company with no product to sell in this space, and its central finding is about format rather than volume. One experiment does not predict every model or every industry, which we say on the page too.
Read the studyIn progress • Our own domain
What third-party scanners say about agentreadywebsite.io
We run the same package we sell, so our own domain is fair game. We are auditing agentreadywebsite.io against two independent free scanners — Cloudflare’s isitagentready.com and the Ora and Vercel is-agentic.com — and fixing what they find.
This one will be published with the score we actually earn, including the checks we do not pass and the honest reason why. Some gaps exist because a static package genuinely cannot close them, and a scanner cannot tell the difference between a missing file and a deliberate boundary. Saying which is which is the point of the exercise.
Not published yet. In the meantime, our detailed disagreement with one of those scanners is already public: AgentReadyWebsite vs Cloudflare’s scanner.
How we choose what goes here
A case study earns a place here if the underlying evidence is public and someone else can check it. That means a published write-up, a measurable result, or a scan anyone can re-run against a live domain. A vendor claim with no method behind it does not qualify, including ours.
We also record findings that argue against parts of our own model. Independent research has found, for example, that sitemap.xml is fetched in only a small share of agent runs and that JSON-LD showed no measured improvement in controlled probes — both of which land on checks we score. We keep those checks for reasons we can defend, and we would rather you hear the counter-evidence from us.
See where your own site stands
One number, four scoring dimensions, and a clear list of what’s missing, in seconds. No signup.
Run My Free Pre-Check