What happened
Vercel announced that AI traffic can now be queried in Vercel Observability. The new AI Traffic view is intended to show how AI bots access a site. It focuses on traffic from verified AI bots rather than presenting all automated requests as one category.
The view exposes five types of information. First, it reports requests by verified AI bot, showing how many requests each bot makes. Second, it identifies the most-crawled pages: pages requested most frequently by AI crawlers for indexing or model training. Third, it distinguishes assistant retrievals, defined as pages fetched by AI assistants in response to user actions or search queries. This distinction matters because a page collected by a crawler and a page retrieved to support an assistant response represent different paths into an AI system.
The feature also reports Markdown requests: pages that AI bots request in Markdown format. Finally, it surfaces errors, identifying pages where requests from AI bots return errors. Together, these categories let an operator inspect both demand and failure modes for machine-oriented visitors without inferring them solely from aggregate logs.
The dashboard entry point is Observability > AI Traffic in a Vercel project. Vercel also makes the metrics queryable through vercel metrics in the Vercel CLI, including use by agents. The changelog gives vc metrics vercel.ai_traffic.count --group-by botName as an example for breaking traffic down by bot name. CLI queries for these metrics require Observability Plus. The source does not state that the dashboard view itself requires that plan, nor does it specify pricing or retention terms.
This is an observability addition, not a bot-control system. It tells an operator what verified AI-bot traffic Vercel records, which pages it reaches, and where it encounters errors. It does not, in the supplied release, offer a policy mechanism to allow, block, authenticate, or otherwise govern those bots.
What the archive adds
This is a new fact in Vercel’s broader observability story. The earlier Field Note, See every agent session in AI Gateway logs., described session-level inspection for traffic flowing through AI Gateway. Vercel Observability adds custom monitors with threshold and anomaly alerts covered query-based monitoring and alerts. The new view applies observability to inbound AI-bot traffic reaching a site.
Trend signal
The release weakly supports Machine readers need operator identity. Vercel’s use of “verified AI bot” shows why a named and attributable machine visitor is more useful operationally than generic bot traffic. But the product only reports traffic; it does not establish an identity or authorization layer for governing machine readers.
Connections to other events
The AI Gateway Sessions view made multi-turn agent activity inspectable as a session rather than as isolated requests. AI Traffic moves in the other direction of the application boundary: it makes machine-originated requests to a site inspectable by bot, requested page, format, and outcome. The custom-monitor release provides a related operational primitive: query-based metrics can be turned into threshold or anomaly alerts.
Why it matters
Teams building AI-facing sites increasingly need to treat AI agents, crawlers, and assistants as a separate class of client. Aggregate request totals cannot reveal whether an increase comes from a particular verified bot, whether a small set of documentation pages attracts retrieval, or whether machine consumers are repeatedly receiving errors.
Vercel’s view gives teams a starting point for those questions. A documentation owner can inspect frequently crawled pages and Markdown demand; an engineering team can locate bot-specific error paths; an operations team can query the same metrics from the CLI. The availability of CLI access also makes the data usable in scripted checks or agent-assisted investigations, subject to the Observability Plus requirement. The immediate value is diagnostic: measure actual behavior before making content, routing, or reliability changes.
What happens next
- Teams operating Vercel documentation or content sites will use bot-level counts and error views to prioritize machine-facing fixes. (3 months, 65%). Watch for: Vercel guidance or product additions around alerting on AI Traffic metrics.
- Demand for clearer machine-client attribution will exceed simple bot labels. (12 months, 55%). Watch for: new controls or identity signals attached to AI traffic reporting.
Limits and uncertainty
The announcement does not explain Vercel’s verification methodology, enumerate the bots included, or state how bots are classified as crawlers versus assistant retrievals. It also does not provide traffic volumes, accuracy measurements, data retention, latency, or examples of error remediation. The source says that CLI queries require Observability Plus, but it does not state the plan requirements or price for the dashboard experience. Most importantly, visibility is not control: the release does not claim to provide access policy, bot authentication, consent management, or a way to block a particular machine client.
Skill to build now
Create a short machine-traffic review for one Vercel project. In AI Traffic, compare the highest-requested pages, Markdown requests, and pages returning errors; then run the bot-name grouping in the CLI if your team has Observability Plus. Turn the result into a small, testable backlog: fix an erroring page, check a heavily retrieved document for accuracy, or validate that its Markdown representation contains the intended material. Keep crawling and assistant retrieval as separate categories during review.
Reader takeaway
AI Traffic makes a previously blended operational signal legible: which verified AI bots reach a Vercel site, what they request, and where they fail. For builders, the useful next step is not to guess what agents consume, but to inspect the new metrics and validate the pages that machine clients actually encounter.
Open question
Will Vercel extend verified-bot reporting into explicit identity, policy, or alerting controls, or will it remain a diagnostic surface?