𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗼𝗻𝘀 𝗙𝗶𝗿𝘀𝘁, 𝗗𝗮𝘁𝗮 𝗟𝗮𝘁𝗲𝗿
Can an AI agent read your website?
I built a tool called Agentis Lux to answer this. It shows you exactly what an agent sees when it visits a site.
Before I run my engine on 50 different websites, I am making seven predictions. I am writing them down now so I cannot change them later to fit the data. This is called pre-registration. It keeps my work honest.
I will scan 10 sites from each group: • E-commerce • SaaS • Content and Media • Government • Indie projects
Here are my bets:
- The rendering cliff is the main divider. Client-heavy sites will struggle more than server-rendered ones.
- Government sites will beat startups in structure.
- Structured data will mostly exist in commerce and media.
- E-commerce sites will show the widest range of scores.
- Some sites will block agents by accident.
- Scores will vary wildly across the web.
- Many sites will have good specs but hide them in the wrong places.
I am scanning the public side of these sites. I am looking at marketing pages, docs, and API specs. I am not looking behind login screens. I am scanning the front doors.
If the data shows my bets are wrong, I will report that. If the data is flat and shows no patterns, I will report that too. A baseline is only useful if it is honest.
I will post the results soon. You can see my hits and my misses.
Which prediction do you think will fail first?