Grok Bot on X: what tagging @bot actually does

#Quick answer

Tagging @bot in a post or reply on X hands that post to your own Grok Bot as a task. @bot posts a short public confirmation and nothing else: the work runs on your Bot's cloud computer and the result lands in that Bot's conversation in the Grok Bot app. It needs SuperGrok access, an X account connected to your Grok account, at least one Bot of your own, and an account based outside the UK, the EU, Australia, Iceland, Liechtenstein and Norway.

Last updated: October 2026

On 8 October 2026 X expanded what these Bots can do with the platform itself, pitching trend insights and brand-mention monitoring: "Ask it to track feedback on your product, follow a breaking story, or send a weekly roundup of your industry." That is the part social teams care about. It is also the part where the coverage stops, because almost nothing written about Grok Bot explains the X side of it, and the X side has sharp edges that produce total silence rather than an error.

#What Grok Bot is, and what it costs to have one

Grok Bot launched on 11 August 2026 as a separate product from the Grok assistant most people met inside X (formerly Twitter). Each Bot is a named teammate with a job, a persistent cloud computer carrying a browser, a filesystem and a terminal, and memory that survives between sessions. The design goal is that work finishes in your real tools instead of arriving as a chat draft, and it keeps running while your laptop is shut.

Access rides on SuperGrok rather than on Grok Bot being sold separately. That means the Premium+ tier on X, which is $40 a month or $395 a year on the web, or a direct SuperGrok, SuperGrok Plus or SuperGrok Heavy subscription. If you are weighing tiers, the full X Premium price breakdown has the per-tier numbers and the app-store markup. The one billing detail worth internalising: Grok Bot comes with its own weekly usage, separate from your Grok usage, and every tagged request spends from that pool. X sends you a notification the first time, telling you exactly that.

For the difference between this and the assistant in the app, what Grok does inside X covers the in-app side, including how to turn it off.

#Tagging @bot: what actually happens

Mechanically it is simple. Reply to a thread, or write a post, and tag @bot with an instruction:

@bot summarize this thread and send the key points to my Slack

@bot answers on X with one short line confirming your Bot has the request. Your main Bot then starts working on its cloud computer. Nothing else appears publicly. The output, the intermediate steps and any file it produces stay in that Bot's conversation in the app.

What your Bot receives is broader than the post you wrote, and this is the detail that changes how you use it. It gets your post, the post you replied to, and any post that either of them quotes, including still photos. So tagging @bot one level deep into a quote chain hands over the whole chain. That is useful when you want a competitor's thread analysed, and it is a reason to think before pointing it at anything sensitive.

Four things have to be true before any of it works:

• A Grok account with your X account connected. On grok.com, that is Settings → Account → Connect next to 𝕏 Account.
• Grok Bot access on that same Grok account.
• At least one Bot you own. A Bot someone shared with you cannot take tagged requests.
• An X account based in a region where tagging has launched.

#The three replies, and what each one means

@bot's public reply is a status code in prose. There are three documented outcomes.

The second and third are easy to confuse from the outside and they mean different things. One is an account-linking problem. The other is an empty roster.

#When @bot says nothing

A missing reply is the failure mode that wastes the most time, because there is nothing to debug and no error to search for. If @bot does not answer, your Bot never got the request. The causes are documented, and they are worth memorising:

• Your account's region is excluded. Tagging is unavailable for accounts based in Australia, the United Kingdom, the European Union, Iceland, Liechtenstein and Norway. Accounts with only a broad region on file, such as "South Asia", or no country at all are also out. The check uses the country X has on record, so your current location and any VPN are irrelevant.
• You replied in a thread that already includes @bot without tagging @bot yourself. Presence in the thread is not a subscription. Every request needs its own mention.
• The post also tags @grok. The two are mutually exclusive in one post. Pick one.
• A video or a GIF is involved. Not just in your post: a video or GIF in the post you replied to, or in a quoted post, also stops it. Still photos are fine. This is the trap in practice, because the threads worth analysing are often the ones with a clip in them.
• The post did not pass moderation.
• Grok Bot is not linked to your Grok account, or your weekly usage is spent.

Notice that four of those six produce identical symptoms. If you are testing, start with a plain text reply in a thread with no media and no @grok, and work up from there.

#What your Bot can see on X without a connector

This is the capability that matters for anyone running an account, and the October expansion made it broader than most people realise.

Reading public X is built in. No connector, no plugin, no authorization step. A Bot can search posts, open a post from its link, read an account's posts and mentions, see who quoted or reposted a post, count recent posts on a topic, and check trends and news. That access is read-only and it does not sign in as you, which is why it needs no permission grant.

What needs the X plugin and a sign-in is your own private surface: your DMs, your bookmarks, your home timeline, and who liked your posts. Those are the four things the docs name.

That split is clean in practice. Competitive research, mention monitoring and trend work need nothing but a Bot. Anything about your own account's inbox or saved items is a deliberate authorization you make once, in Connect apps, and connectors install account-wide rather than per Bot. Worth knowing if you run several Bots and assumed each had its own permissions: they share one cloud computer per user, and the docs say plainly not to treat separate Bots as a security boundary.

#Where this is genuinely useful, and where it is not

The honest framing is that Grok Bot is a research and drafting layer, not a publishing one.

It is good at the work that happens before a post exists. A weekly roundup of what your niche argued about. A read on whether a complaint about your product is one loud account or forty quiet ones. Pulling the thread structure out of a post that performed well. For that kind of job the no-connector read access plus a scheduled routine is a real improvement on checking search columns by hand, and the output lands in a conversation you can correct.

It is not a scheduler and it is not an engagement tool. Sensitive or consequential actions stop for your approval by design, and passwords, two-factor codes and CAPTCHAs are handed back to you rather than worked around. Anything that publishes on your behalf, from either a Bot or any other tool, is automation under X's automation rules, which set expectations about human involvement and require the automated account label on accounts that post autonomously. Browser-driven automation of your own account sits in a gray area that X has never blessed, and no tool, ours included, can state otherwise. The lower-risk setup is the one where the machine does the research and a human owns the publish, which is how we build X-Autopilot: your session stays on your own Mac and your own residential IP rather than in a shared cloud pool, which lowers the detection surface without eliminating it.

If you want to go the other direction and build something on this, Grok Bot templates are now a first-class listing type in the X Developer Exhibit, and there is an invitation-only rewards pilot paying template creators through X Money. For wiring your own tooling to X instead, the hosted X MCP servers are the sanctioned API path.

#Before you tag it

1. Check your account's country, not your location. If X has the UK, the EU, Australia, Iceland, Liechtenstein or Norway on file, tagging will not answer and nothing you change locally will fix it.
2. Create one Bot and give it a narrow job. A Bot named "Mention Watch" with a described boundary produces better work than a general helper, and it is the only kind of Bot that can accept tagged requests.
3. Test on a text-only thread. Media anywhere in the chain is the most common silent failure.
4. Decide what it is allowed to do in the Bot's description before you grant anything. "Never send external messages without approval" belongs in the description, not in a prompt you have to repeat.
5. Watch the weekly usage. Tagged requests spend the Grok Bot pool, and a routine that runs daily spends it faster than you would guess.

The feature worth the setup is not the agent theatre. It is that reading public X properly, at a scale nobody does by hand, now takes one tagged reply and no API key.