AI NewsGo-to-marketReported
An audit of 50 websites scored agent transaction readiness at 2.1 percent
Reza Moaiandin of SALT.agency published an audit of 50 large websites across three layers of AI readiness and reported that the layer covering agent transactions and machine discovery scored an average of 2.1 percent, with 46 of 48 testable sites at zero.
Image: Search Engine Journal
Why it mattersThe work needed to be found by an answer engine and the work needed to be usable by a buying agent are different jobs, and almost nobody has started the second one.
Reza Moaiandin, co-founder at SALT.agency, published an audit on 1 September of 50 large websites drawn from retail, SaaS, travel, publishing and finance. The team scored each site against 27 elements grouped into three layers, using an instrumented browser that captured live HTTP responses, the rendered DOM, the raw server HTML and machine-discovery endpoints. The capture ran on 12 June 2026.
The gap between the three layers is the finding.
What each layer scored
Layer one covers retrievability: can a machine reach the content at all. It averaged 74.4% across the 50 sites, and only 3 of them scored below 50%. JSON-LD structured data was present on 35 of the 50 homepages, which is 70%.
Layer two covers attribution and meaning, and it averaged 38.5%. Within that layer, 5 of the 50 sites had a content signals policy, which is 10%, and 11 had published an llms.txt file, which is 22%.
Layer three covers agent transactions and machine discovery, and it averaged 2.1%. Of the 48 sites that could be tested for it, 46 scored zero.
The overall averages were 56.6% mean and 58.3% median. Airbnb.com scored highest at 79.2%. Ten of the 50 sites scored under 50% overall. The audit also reports that 29 of the 50 sites, which is 58%, showed no deliberate decision either way about AI bot access.
These are SALT.agency's own measurements against SALT.agency's own framework, published through Search Engine Journal, and no independent party has repeated them.
Layer three is the engineering team's job
The three-layer split is the part worth copying, whatever you think of the exact scores. It separates being readable from being understood from being transactable, and those turn out to be three different pieces of engineering with three different owners.
The first layer is mostly solved, and the numbers say so. Serving clean HTML, letting the right crawlers in and publishing valid JSON-LD is work most large sites have already done for classic search, and it works for answer engines with little change.
The third layer is where the score falls to almost nothing, and it is work for engineering rather than for a content team. Publishing machine-discoverable endpoints, describing what an agent may do on your site, and making a transaction completable without a human clicking through a rendered checkout are all product decisions with real implementation behind them. A score of 2.1% across large, well-resourced sites says this has not started in any serious way.
The honest reading is that the field has not settled what layer three should even look like, and the standards competing for that job are new. A team deciding where to spend this quarter should treat layers one and two as maintenance with known answers, and treat layer three as an open question worth tracking rather than a checklist to complete.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
