GitBook says AI systems are no longer a side audience for developer documentation. In a new Vercel customer story, the company said 41% of traffic to its hosted docs network now comes from AI crawlers and automated systems, while the platform serves 30,000 documentation sites and roughly 120 million monthly page views.

What changed

The more concrete infrastructure detail is how GitBook keeps those sites current. According to Vercel's write-up, GitBook uses Next.js with the use cache directive and a tag-based invalidation model tied to merge events. When content is updated through GitBook, GitHub, or GitLab, only the affected cached units are revalidated rather than purging the whole multi-tenant deployment.

GitBook said that approach now handles about 40,000 cache invalidations per day, with updated content becoming visible globally in under 300 milliseconds. The company's public GitHub repository separately confirms that GitBook's published content renderer is open source and built on Next.js.

Why it matters

The headline is not just that docs hosting is large. It is that machine readers are starting to behave like a primary production workload. Human readers usually touch a few pages at a time; crawlers can sweep entire doc sets across thousands of sites and hit cold cache paths that people rarely visit.

That makes documentation infrastructure look more like search or API infrastructure: it needs immediate consistency after each merge, targeted cache invalidation, and costs that stay predictable even when automated agents become a major share of the audience.