Cursor has now given three different answers to what Grok Bot bills against. Users are still asking.
The marketing email promised a dedicated Grok Bot usage pool, separate from your Cursor usage. The forum shows what it took to find out whether that was true, and the answer is still: it depends which part of Grok Bot you mean.
When Cursor rolled Grok Bot out to paying customers, the email said the plan "now comes with a dedicated Grok Bot usage pool, separate from your Cursor usage." That sentence is the reason a user posting as TownDrin opened a thread on the Cursor forum on August 28 titled "Grok Bot spend cursor usage, I can't accept it." They had watched the always-on agent chew through their Pro plan's model allowance and could not even choose which model it used. "I thought Cursor Bot would have a separate usage pool," they wrote, "but only after using it did I realize that it shares the same usage pool with Cursor."
The first official answer came within hours from Dean Rie, a Cursor staff member and forum moderator, and it confirmed the user's fear. "Grok Bot is billed to your Cursor account and uses your plan's weekly usage on Pro, not a separate limit." Another user replied by pasting the marketing email with its promise of a dedicated pool. Six hours after his first post, Rie posted a correction. "On Pro, Grok Bot has its own separate included weekly pool, separate from your Cursor plan usage," he wrote, adding that once that pool is used up, further Grok Bot work "will start billing from the shared On-Demand pool" if on-demand spending is turned on. Answer one: same pool. Answer two: separate pool, with an overflow into your wallet.
The dashboard disagreed
Users kept checking their own numbers. One, posting as icetique, ran nothing but a Grok Bot automation and watched the spending page attribute the work to claude-opus-5-thinking-low and a fast Grok 4.6 variant. It "used 7.1% of my included allowance of Cursor Models during a short run," they wrote. Another, PeterShanxin, took the question to Cursor support on September 1 and reported back that support "explicitly confirmed" the Grok Bot rows were included in the monthly Cursor Models and Other Models percentages shown on the page, while also confirming that the weekly Grok Bot allowance "is supposed to be separate." Both things cannot be true unless the dashboard is wrong, and on September 2 that became answer three. Support told icetique that Grok Bot usage showing up under Cursor Models "is a display issue, not actual consumption of the monthly Cursor allowance," and that the requests were really being recorded against the separate weekly pool.
The thread ended there, with 820 views and an open question about what happens when the displayed Cursor usage hits 100 percent on the strength of entries that supposedly do not count. Cursor's own help page is not much more precise. It says paid Grok Bot access "includes usage that resets weekly" and that when the weekly amount runs out, extra usage "can continue on your account's shared on-demand spend if on-demand is enabled." It describes the Pro, Pro+ and Ultra tiers only relative to one another, with no figures.
Monday's seven questions
On Monday a second user, a marketing agency posting as Publicidade Já, opened a new thread that shows the confusion did not end with the display-issue explanation. It quotes two further statements from a Cursor staffer named Mohit. On September 3: conversations use the Grok Bot weekly pool, but routines and automations count against the Cursor plan, and so do browser actions and cloud-agent work. On September 7: the grok-bot rows under Cursor Models are "display-only," but "cloud agents launched from Grok Bot do count toward your Cursor plan." The agency then lists seven questions, and they are the right ones. Does a Grok Bot automation draw from the weekly pool, the Cursor plan, or both at once? Which actions count as a "cloud agent launched from Grok Bot"? Can a single task drain both allowances simultaneously? Is the quota token-based or does every agent step count? Which features keep working when one pool is empty and the other is not?
None of those has an answer on the record as of Monday evening. We have laid out the dates because the sequence is the story. Between August 28 and September 7, Cursor's staff and support gave four public descriptions of how one included feature is metered, two of them from the same person on the same day, and a paying customer still cannot say with confidence what a routine costs.
Why this is a pricing story, not a support story
"Included in your plan" is becoming the default way agent features are sold. Copilot has its credits, Claude Code has its hours, and Cursor, under new ownership since the SpaceX deal, has made Grok Bot its headline product. An always-on agent that launches sub-agents, runs routines and drives a browser is the hardest possible thing to meter against a plan designed for a person typing in an editor, because one instruction fans out into many billable actions across several pools. That is exactly the gap the Cursor thread exposes. The company appears to have shipped the feature before it finished deciding, or at least before it finished explaining, how the meter runs.
The contrast with OpenAI is instructive. When Codex spending got unpredictable, OpenAI shipped hard budget caps so a customer could at least bound the damage. Cursor's equivalent today is a forum thread and a promise that the scary number on the dashboard is cosmetic. For a $20-a-month individual that is an annoyance. For a team lead deciding whether to turn on on-demand billing for forty seats, it is a reason to wait. We have asked nothing of Cursor that its own customers have not already asked in public, and we will update this piece when the company answers them.
