Vercel Observability adds AI traffic insights for tracking crawler and assistant requests

What Vercel added

Vercel has added dedicated AI traffic insights to its Observability product, giving site owners a direct view of how AI bots access their deployments. According to the Vercel changelog post announcing the feature, users can now query this traffic to see which bots are making requests, which pages they target, and where those requests fail.

The new view lives in the Vercel dashboard under Observability > AI Traffic.

What the metrics show

Vercel groups the insights into five categories:

• Requests by verified AI bot: the request volume each identified bot generates
• Most-crawled pages: the URLs that AI crawlers hit most often, whether for indexing or for gathering training data
• Assistant retrievals: the pages an AI assistant fetches when reacting to a user action or search query
• Markdown requests: the content bots ask for in Markdown form
• Errors: pages where a request from an AI bot returns an error

The split between crawling and assistant retrieval is the notable part of that list. Crawling happens on the bot's own schedule and at scale, while an assistant retrieval is a targeted fetch triggered by an end user, closer to a single visit that begins inside an AI product. Separating the two lets owners see both how their site is being collected in bulk and which specific pages surface in assistant-driven lookups.

Dashboard and CLI access

Beyond the dashboard view, the same metrics can be queried from the Vercel CLI using vercel metrics. Vercel frames this as something "you and your agents" can do, suggesting automated agents can pull the data as part of their own workflows. The changelog offers an example:

vc metrics vercel.ai_traffic.count --group-by botName

That query returns a site's AI traffic broken down by bot name. There is one tier restriction: according to the post, CLI queries require Observability Plus. The announcement does not state an equivalent requirement for the dashboard view.

Why it matters

AI traffic is a growing part of the web, but most site analytics are built around human sessions: page views, visits and referrers. A crawler collecting training data and an assistant answering a user's question both hit your origin server, yet neither appears in conventional analytics in a way that explains what actually happened. Vercel's breakdown makes that traffic legible, per bot, per page and by intent.

That legibility supports several practical decisions. Per-bot request counts give owners a factual basis for choosing which crawlers to allow or block. The error view catches pages that fail for bots, which matters for anyone who wants their content represented in AI answers or indexes. Markdown request data shows which pages are being consumed in a form geared toward language models. And assistant retrieval figures point to the content actually being used to answer people's questions, a rare window into AI-mediated readership rather than browser visits.

The feature also signals where hosting platforms are heading: AI traffic is becoming a first-class concern for observability rather than noise to filter out. For teams already on Vercel, the dashboard view makes this data available with no setup, while the CLI integration, gated behind Observability Plus, extends the same metrics to scripts and automated agents.