Luke ByrnePractical AI resourcesWatch the tutorial ↗

Grok Bot · guide + prompts

Give your Bot a repeatable job.

Turn three public sources into a brief you can check. Save the method as a skill, then run it again without repeating old stories.

Download the guide + prompts ↓Open Grok Bot ↗

A practical companion to the recorded lesson, with the original prompts.

In this guide

These are the retained prompts from Luke’s Grok Bot tutorial. Work through one at a time and inspect the result before moving on. Start with the three sources below, then adapt the workflow to a job you already do.

What was tested: the briefing, saved private skill and a manual replay. What remains untested: the schedule firing at 08:00. The demo routine is still paused.

1. Give the Bot one job

Open Grok Bot, check your account access and usage, and use a conversation with one Bot. The recorded setup was an Apple Silicon Mac. These prompts read public pages; no extra website accounts were needed.

You are my AI Tool Briefing Bot. Help me keep up with product changes using only public information. Start from these three pages:
1. https://x.ai/changelog/bot
2. https://ollama.com/blog
3. https://blog.google/innovation-and-ai/products/gemini-notebook/
You may open the original announcement links on those sites to verify details. Keep factual changes separate from your interpretation. Remember this role for future work. Do not connect private accounts, contact anyone, publish or share anything, buy anything, or schedule work unless I explicitly ask. First, summarise your role and source list; do not start research yet.

2. Check one source

Open the original changelog yourself and compare its date, version and change. The recorded test returned 23 October 2026, version 0.75.5. That is a dated example, not the release you should expect today.

Open only https://x.ai/changelog/bot. Find the newest dated release you can actually read. Return its date, version if shown, direct link and one factual change in your own words. Do not infer missing details. If the page is inaccessible or you cannot verify a date, say so and stop. Do not create a Page or schedule anything.

3. Build the briefing

The brief should contain at most three dated updates with original links, practical uses clearly labelled as interpretation, and a status for each source. A quiet day can produce fewer updates.

Create my first AI Tool Brief using the three agreed public sources. State the current date and Europe/London time at the top. Look back seven calendar days, including today. For each publisher, select at most one substantive product update within that window and open its original announcement to verify it. Return up to three updates total; do not fill a quiet source with old news.
For each update include: the product, publication date, what changed in one plain-English sentence, one possible practical use clearly labelled as your interpretation, and a direct source link. Keep the total brief under 250 words. Add Source checks at the end, reporting each source as checked, inaccessible or no dated update found. Never describe an inaccessible source as having no new updates.
Return the result only in this conversation. Do not create a Page, share, publish or schedule anything.

4. Review the result

Open the cited announcements yourself. Asking the same AI to check its answer is useful, but it is not an independent fact check.

Review the brief you just produced against the original announcement pages. For every factual claim, check that the cited page supports the wording and date. Remove or qualify unsupported details. Keep interpretation visibly separate. Make each practical-use sentence specific enough for a beginner. Return one revised brief in this conversation, under 250 words including source checks, then list changes separately. If no factual correction was needed, say that; do not manufacture one. Do not create a Page or schedule anything.

5. Save the private skill

Inspect the saved skill: check the source list, format, word limit, inaccessible-source rule and the record of prior checks. A skill saves the method; it does not schedule itself.

Save the method we have just tested as a private skill called AI Tool Brief. Include the three exact source URLs, original-article checks, output structure, under-250-word limit including source checks, and the rule that an inaccessible source is not the same as a quiet source. For an initial run, use the previous seven calendar days including today. On later runs, use the last successful run time as the start of the window and avoid repeating the same announcement. Use URL plus release version/date when one changelog URL holds several releases. If there is no reliable prior run record, state that and use the initial seven-day window. Keep a short record of checked times and covered URLs. Return results only in this conversation. Do not connect private accounts, contact anyone, publish or share output, buy anything or create a Page. Show me the saved skill so I can review it. Do not schedule or run another research task yet.

6. Replay the skill manually

This is the tested replay step. On 26 October the Bot reported no new dated items and did not repeat the earlier announcements. It also reported using Google’s blog feed after a listing problem; that feed access and its internal run-record update were not independently inspected.

Invoke the saved AI Tool Brief skill exactly once as a manual task in this chat. Keep AI Tool Brief Demo paused/disabled. This is a saved-skill replay, not a timed routine test. Preserve the public-source-only, no-sharing, no-Page and no-contact boundaries. Show the checked time window and source statuses; do not repeat already-covered announcements.

7. Optional: create a paused routine

Optional: only add a schedule once you want this result regularly. The recorded AI Tool Brief Demo routine was paused. Creating a schedule is not proof that it fires successfully.

Create one routine named AI Tool Brief Demo for this Bot to run the saved AI Tool Brief skill daily at 08:00 Europe/London. Create it PAUSED or DISABLED and do not enable it. If you cannot create it paused, stop instead. Return results only in this conversation. Do not connect accounts, create a Page, contact anyone, send external notifications, publish or share. Show the owning Bot, saved time zone, actual timing behaviour and current paused state. Do not run it.

8. Optional: inspect exact timing

Optional: inspect the owning Bot, time zone, timing and paused state. If your account lacks an exact-time setting, keep it paused and review the limitation. The video did not test an unattended 08:00 run.

Edit AI Tool Brief Demo only: keep it paused/disabled. If an exact-time scheduling option is supported, set it to exactly 08:00 Europe/London with flexible timing disabled. If exact timing is not supported, keep it paused and explain that limitation. Do not run or enable the routine. Confirm the saved state.

Check your result before using it regularly

  • Each factual update has a readable original source, date and direct link.
  • The brief stays under 250 words and does not pad quiet sources with old stories.
  • Inaccessible sources are reported separately from sources with no new dated update.
  • A replay starts from the last successful check and avoids already-covered announcements.
  • The saved routine remains paused until you deliberately choose to enable and test it.

If the result misses something you need, tell the Bot to revise the saved skill, then run another small manual check. Keep the workflow tied to work you actually want to replace.

Sources and scope

This guide preserves the recorded public-source workflow. It does not demonstrate private app access, multiple Bots handing work to each other, or unattended reliability. Product controls and account eligibility can change.

Start with one small job you actually repeat.

Open Grok Bot ↗