• Open regular Grok and confirm that ordinary chat works for your account.
• If you want private bookmark access, review the current Grok Bot plan rules. On The X Subscription, Premium, Super, and Ultra include Grok Bot; Free and Plus do not.
• Open the Grok Bot setup page and sign in with the eligible account you intend to use.
• In the latest X app or at x.com, open X Chat and look for Bots. If the Bot is not available there, use the Grok Bot app instead.
• If you do not have Grok Bot, skip to the free text-and-URL fallback after Step 6.

Grok Bot is an agent that can use connected tools and a persistent cloud computer. Some linked legacy, business, or Cursor plans can also qualify for Bot, but X Chat has narrower account rules.

The official X Chat guide says X Chat access is rolling out gradually. It requires a Bot-eligible plan, a Grok account connected to a Cursor account, and the latest X app on mobile. Connecting a Bot from X Chat is for personal accounts. Company-managed email domains, Cursor team accounts, and legacy request-based plans are not eligible for that connection route. The Grok email must be verified, Legacy Privacy Mode must be changed, and mismatched Grok and Cursor accounts can also prevent the Bot from appearing.

• Regular Grok works but Grok Bot does not: Your account may not include Bot. Use the free fallback instead of treating regular Grok as connected bookmark access.
• The Bot is missing in X Chat: Update the X app, verify the linked account and email, and check the account restrictions above. Use the Grok Bot app while X Chat is unavailable.
• You are unsure which account is linked: Stop before connecting X. Confirm the Grok, Cursor, and X account identities first.

• Open the Grok Bot app and sign in with the eligible account from Step 1.
• Open Marketplace or Settings > Plugins.
• Select the official X for Grok Bot plugin.
• Sign in to the X account that owns the bookmarks and review the permissions if X asks. If you previously authorized this connection, the plugin may reconnect immediately without showing a new consent form.
• Before reading any bookmarks, verify that the connection belongs to the intended X account. If you cannot verify the account, disconnect and resolve the account identity first.
• Check the connected X developer account's API-credit balance. Private bookmark requests need X API credits. If the balance is $0, add prepaid credits or use the free fallback.

Grok Bot can search public X posts, profiles, mentions, trends, and news without connecting your X account. The plugin is for account-private data such as bookmarks, DMs, the home timeline, and information about who liked posts.

Bookmarks are private to the X account that saved them. Connect only the intended account, use the narrowest permissions available, and revoke the plugin after the exercise if you no longer need it. The Grok Bot troubleshooting guide explains that reconnecting or changing plans does not add X API credits.

An immediate reconnect means an existing X authorization was reused. It does not confirm that the correct account is connected, so verify the account before asking the Bot to read private data.

• The X plugin is missing: Check the Bot Marketplace and update the app. Do not install an unofficial plugin with a similar name.
• The plugin connects but bookmarks are unavailable: Confirm that you connected the X account that owns them and that bookmark permission is included.
• The API-credit balance is $0: Add prepaid X API credits or use the free fallback. Reconnecting the plugin will not create credits.
• Sign-in opens the wrong X account: Cancel, sign out of the unintended account, and reconnect before sharing any private data.

• Choose a focused topic. For practice, use AI agents for everyday work, including coding and design workflows.
• Write an exact start date, end date, and timezone for post publication, such as October 28 through November 3, 2026, in Eastern Time.
• Search X using two or three focused terms and open posts published inside that window.
• Bookmark 8 to 12 posts that add distinct examples, updates, questions, or first-hand accounts.
• Save the topic, publication window, timezone, and bookmark count in a note for the next step.

The publication date on a post is not the date you saved the bookmark. Do not infer a bookmark save date from the post date or its position in your bookmark list.

A defined topic and publication window make the sample repeatable. The sample still reflects only what you chose to save. Likes, reposts, and repeated claims do not prove that a claim is true or common across X.

• A post's publication date is unclear: Leave it out of the dated sample or mark it unavailable.
• You find fewer than eight useful posts: Continue with the smaller sample and record the actual count. Do not add weak posts just to reach a target.
• Most posts repeat one announcement: Keep the clearest source and look for distinct examples, questions, or user problems.

• Open Grok Bot in its app or X Chat and confirm that the official X plugin is connected.
• Keep Auto Review on and add an Ask first rule for posting, replying, liking, reposting, following, messaging, or changing bookmarks. Never choose Always allow for those actions in this workflow.
• Replace the bracketed topic and date fields in the prompt below, then send it to the Bot.
• Check the reported pages or batches, total bookmarks inspected, matching posts, date and topic exclusions, and unread or truncated source content.
• Open the linked source posts and verify every theme before keeping it.

Copy this research prompt:

Review the X bookmarks I selected for this exercise.

Topic: [TOPIC]
Post publication window: [START DATE] through [END DATE]
Timezone for the publication window: [TIMEZONE]

Use the official X connection. Read only bookmarked posts about this topic that were published inside the stated window. Do not infer when I saved a bookmark. Do not use my likes, general timeline, DMs, or unrelated topics.

Inspect no more than the first 100 bookmarks returned. If the connection returns pages or batches, request only as many as needed to reach 100 and then stop. If the first page contains fewer than 100, stop after that page. Do not silently retrieve my full bookmark history.

Before the analysis, report:
- Every page or batch requested, the number returned in each, the total inspected, and whether more pages were available when you stopped.
- The number published inside the stated window.
- The number excluded because the publication date was outside the window or unclear.
- Among the in-window posts, the number excluded as off-topic.
- The number of matching posts retained for analysis.
- The number of returned posts marked unavailable.
- The number and URLs of matching posts with truncated text or linked pages, articles, or videos that you did not read.

Return:
1. Up to five useful posts, with author, publication date, and direct post link.
2. Three themes that appear across multiple posts, when the sample supports them.
3. One practical workflow people in this sample are trying.
4. One recurring problem or limitation in this sample.
5. Two claims that need independent verification.
6. Three questions for follow-up research.

Do not treat likes, reposts, or repeated claims as proof. Do not invent a post, author, date, quote, or link. If you cannot read a bookmark, list it as unavailable. Keep fact, opinion, and your own inference separate. Do not claim that this sample represents all of X.

Do not like, repost, bookmark, follow, reply, post, or message anyone. Return the brief in this chat only.

A useful brief links every important statement back to a source post. A theme is a way to organize your reading, not proof of a platform-wide trend.

In the observed practice run, the Bot inspected only the first 100 returned bookmarks. For posts published from October 28 through November 3, 2026, in Eastern Time, 17 fell inside the date window. Ten matched the AI-agents topic, including coding and design examples, while seven of those 17 were excluded as off-topic. None of the returned posts were marked unavailable, but some post text was truncated and no linked pages or videos were read. The 10 matches are a bounded sample, not all bookmarks and not complete evidence for every linked claim.

The prompt sets a clear boundary, but the Ask first rule provides the technical review stop for matching X actions. Deny any unexpected action card and review the approval controls before continuing.

• The Bot sees no bookmarks: Check the connected X account, bookmark permission, and API-credit balance.
• The Bot gives themes without links: Ask for the author, publication date, and direct URL for each supporting post. Remove unsupported themes.
• The Bot includes posts outside the window: Ask it to separate them and rerun the analysis using the exact publication dates.
• An unexpected action card appears: Deny it. Check Auto Review and the Ask first rule before sending another request.

• Choose one theme from the brief and open every source post linked to it.
• Confirm that the sources support the theme and fall inside your publication window.
• Replace the bracketed theme field in the prompt below and send it in the same conversation.
• Check every factual claim and remove anything the sources do not support.
• Keep the result in chat. If you later decide to publish, copy and review it yourself instead of asking the Bot to post.

Copy this drafting prompt:

Use the theme [PASTE ONE THEME] from the bookmark brief to draft one original X post.

Audience: people who are curious about using AI agents for everyday work.
Goal: explain one useful lesson in plain English.
Length: no more than 240 characters.
Tone: direct, useful, and specific.

Use only facts supported by the linked posts. Do not copy another person's wording, imply that one example proves a broad trend, invent numbers, or claim that I tested something I did not test.

After the draft, list the source post links that support it and identify any wording that needs my review. Do not publish, schedule, reply, like, repost, or message anyone.

In this example, the first draft claimed that review could happen inside a Slack thread, although its sources did not establish that. The revised draft removed the claim, attributed the announcements to their sources, and ended with a question about who approves an agent’s changes. It contains 226 characters.

Keep your own experience separate from something another person posted. Verify claims that need outside proof before sharing. The prompt requests a draft only, but it does not technically disable writes. Keep the Ask first rule in place and deny any action you did not request.

• The draft overstates the sample: Replace broad phrases such as “everyone” or “across X” with language about the saved posts you reviewed.
• The draft contains an unsupported fact: Remove it or add a verified source before using it.
• The draft exceeds 240 characters: Ask for a shorter version without dropping the central supported claim.
• The Bot proposes publishing: Deny the action and keep the text as a draft.

• Save the first brief with its topic, publication window, and bookmark count.
• Choose a second non-overlapping publication window and timezone for the same topic.
• Bookmark a new set of relevant posts published inside the new window.
• Paste the earlier brief and the comparison prompt below into Grok Bot.
• Open every linked example and save the comparison yourself. Do not create an automatic posting routine.

Copy this comparison prompt:

Compare my current X bookmark sample with this earlier brief:

[PASTE EARLIER BRIEF]

Topic: [TOPIC]
Current post publication window: [NEW START DATE] through [NEW END DATE]
Timezone for the current publication window: [TIMEZONE]

Use only bookmarked posts about the topic that were published inside the current window. Do not infer when I saved a bookmark. Link every new example.

For the current sample, inspect no more than the first 100 bookmarks returned. If the connection returns pages or batches, request only as many as needed to reach 100 and then stop. If the first page contains fewer than 100, stop after that page. Do not silently retrieve my full bookmark history.

Before comparing the samples, report:
- Every page or batch requested, the number returned in each, the total inspected, and whether more pages were available when you stopped.
- The number published inside the current window.
- The number excluded because the publication date was outside the window or unclear.
- Among the in-window posts, the number excluded as off-topic.
- The number of matching posts retained for comparison.
- The number of returned posts marked unavailable.
- The number and URLs of matching posts with truncated text or linked pages, articles, or videos that you did not read.

Identify:
- Themes supported in both samples.
- Topics that appear only in the current sample.
- Earlier themes with little or no support in the current sample.
- Claims that still need independent verification.

Do not claim that a theme is growing or shrinking across X as a whole. You are comparing only two bounded samples I chose to save. Do not change bookmarks or publish anything.

This creates a personal reading record. It does not measure all of X or prove that a topic is becoming more or less popular. Differences can come from what you searched for, whom you follow, and what you chose to bookmark.

• The two date windows overlap: Adjust the dates before comparing so each post belongs to one sample.
• The Bot compares posts without links: Ask for the direct URL and publication date for every example.
• The sample sizes differ: Keep the actual counts visible. Do not present raw theme counts as a growth rate.
• The Bot claims an X-wide trend: Remove that claim and restate the result as a comparison of your saved samples.