Grok Bot can now help build a presentation, show sample slides for you to choose from and deliver the result as a PowerPoint file or Google Slides. SpaceXAI lists that capability in its 7 October 2026 v0.68.1 release note. For a small business, the useful change is a shorter route from an approved brief to a reviewable deck.
The final handover still needs a person. A polished slide can contain an unsupported number, an outdated promise or a private detail that should never leave the team. Start with a limited presentation task, keep the source material traceable and inspect the delivered artifact before sharing it.
This guide explains the update and gives you a practical first test. The official sources were checked on 9 October 2026. The example below is hypothetical; we have not tested the feature or measured time savings. For the wider operating framework, read how to use AI in a small business without losing control.
What the Grok Bot update changes
The release note describes a collaborative deck workflow: work with a Bot, select from sample slides and receive a presentation in one of the two named formats. That gives you a useful checkpoint before asking for the whole deck. Decide whether the sample communicates your message clearly, then continue with the chosen direction.
Keep the product distinction clear. This announcement concerns Grok Bot. Grok already had separate presentation add-ins for PowerPoint, announced on 16 June, and Google Workspace, announced on 24 July; this is not the first time any Grok product has worked on slides. Nor does the Bot release note establish a new presentation API, perfect export fidelity or a universal account entitlement.
Check access before planning a live handover
The official Grok Bot page currently lists access for Cursor Pro, Pro+ and Ultra; SuperGrok, SuperGrok Plus and Heavy; Cursor Standard and Premium Teams; and Enterprise. Its FAQ describes included weekly usage and additional usage charges. Confirm your own entitlement, allowance, app version and administrator restrictions rather than assuming an ordinary free Grok account includes the workflow.
For the first test, choose the output your reviewers already use. A team that checks presentations in PowerPoint should inspect the actual delivered file there. A team working in Google Slides should review the actual presentation and its sharing state. Treat format choice as a handover decision, not a claim that both versions will behave identically.
Build a small source pack before asking for slides
Give the Bot a defined audience, purpose and source boundary. “Make an impressive sales deck” invites more interpretation than “Explain the approved onboarding process to five new staff members using these two documents.” The second brief makes errors easier to spot.
A useful source pack contains the current approved brief, a short outline and any permitted brand guidance. Add a source date and owner. Label an older document as background, or leave it out if it could confuse the current instruction. If two sources disagree, resolve the difference before generation or explicitly ask the Bot to stop and identify the conflict.
Choose one decision the presentation should support. For an internal training deck, that might be understanding the next three actions. For a client update, it might be approving a specific next step. Decide what the deck should leave the audience able to do, then remove slides that do not help that outcome.
The official files-and-results guide explains how to attach source material and request an artifact you can inspect. Use that capability to provide a bounded pack. It is not permission to upload everything in a shared drive. Start with public, synthetic or appropriately redacted inputs under your organisation’s data rules.
Use the sample slide as a review checkpoint
A sample is valuable when it tests more than colour and typography. Look at the main claim, how the source is used and whether the visual hierarchy helps the reader. A sample with a strong headline but an unsupported percentage should be corrected before its style spreads through a full deck.
Give specific feedback: “Keep the two-column layout, remove the invented attendance figure and show the three confirmed steps.” Avoid asking for vague improvements such as “make it more authoritative”, which can encourage confident language without stronger evidence.
Once the direction works, ask the Bot to continue the same artifact. Keep revisions tied to a named version, so the reviewer knows which file or link contains the corrections. A collection of disconnected exports can hide the version that was actually approved.
A compact prompt for a bounded first deck
Use this starting prompt with material you are permitted to share. Replace the brackets and check that every instruction matches your account’s available tools. This defines a review boundary; it does not guarantee that an automated system will follow every instruction.
Create a draft presentation for [audience].
Purpose: [one decision or learning outcome].
Sources: [approved files/links, dates and owners].
Use only those sources for factual claims.
First propose a five-slide outline and sample slides.
Flag missing facts, contradictions and assumptions.
Do not invent figures, quotations or customer results.
After my review, revise the same draft.
Deliver as [PowerPoint file / Google Slides].
Include source references in appropriate slide notes.
Do not email, publish, change sharing or overwrite originals.
Stop for my review of the actual delivered artifact.Name the reviewer separately in your task record. The official approvals guide recommends explicit boundaries around consequential actions. Approval of an operation does not establish that the slide content is accurate. Review the actual target and effect, and keep broad automatic permissions out of a first pilot.
A worked example: a five-slide training brief
Imagine a small training provider preparing an internal briefing for a new workshop. Its approved source pack contains the course outline, joining instructions and facilitator checklist. The audience needs to understand the session’s purpose, preparation, timing, responsibilities and next action.
The owner requests five slides: learning outcome; attendee preparation; the session sequence; facilitator responsibilities; and the next step. Before generation, the owner marks the registration total as unconfirmed. The Bot should flag that gap or omit the number, rather than turn a plausible estimate into a statement.
The reviewer checks a sample against the outline. If it says “95% satisfaction” without an authorised source, the draft fails that check even if the layout is attractive. The reviewer asks for the claim to be removed and checks the revised version. No testimonial, customer name or claimed result belongs in the deck merely because it sounds persuasive.
For a second test, supply a deliberately outdated joining time alongside the current instructions, clearly labelled as a conflict to investigate. A useful result should identify the discrepancy and stop for a decision. This exercises the workflow’s handling of uncertainty; it does not justify using real attendees’ private records as test material.
The team’s acceptance criteria are concrete: five relevant slides, no unsupported figures, the current time and location, references for consequential facts, and no external sharing. The named reviewer approves the delivered version only after opening it. This is an example of a test design, not a report of product performance.
Review the actual artifact, not only the chat preview
Open every slide at a normal presentation size. Read the speaker notes as well as the visible content. Check that a source supports the exact claim, including its date, scope and units. If the deck shows a percentage, confirm the denominator and period; if it shows a chart, compare the plotted values with the approved data.
Check presentation quality separately from factual accuracy. Look for clipped text, unreadable labels, broken links, inconsistent language and images you do not have permission to use. Make sure colour is not the only way to distinguish meaning. An image or chart should help explain a point rather than make an uncertain claim feel proven.
Then inspect the handover. Confirm the file or presentation owner, the intended version, the recipient list and any cloud sharing permissions. An internal draft should not become publicly accessible as an incidental step. Keep sending or publication behind a deliberate human checkpoint; the human approval guide explains where that checkpoint belongs.
If the work uses a shared assistant, apply the access boundaries in the Grok Team Bots pilot guide. Do not assume creating another Bot creates an independent security boundary for your files or signed-in services.
Measure usefulness after review and repair
Compare this bounded task with your normal manual process. Record briefing time, generation time, human review, repairs and final handover. A fast first draft can still create more work if every slide needs fact checking or reformatting. Count the effort needed to reach an approved, usable presentation.
Repeat with a missing fact and a conflicting source. Save the inputs, version and reviewer’s decision. The business AI test-set guide gives a way to keep those cases consistent. Hold the workflow when it invents evidence or shares beyond the approved scope, and repair the same draft before retesting.
You do not need to prove a sweeping return on investment. Decide whether this one presentation task meets your quality threshold with an acceptable amount of review. That is a stronger reason to continue than one attractive sample or an unsupported claim that AI has saved hours.
Reader Q&A
What changed in Grok Bot on 7 October 2026?
The v0.68.1 changelog says Bots can build slide decks with users, offer sample slides to choose from and deliver the result as a PowerPoint file or Google Slides. This is a Grok Bot update, not a new model or a claim that Grok had no presentation tools before.
Is the slide-deck feature available to every free Grok account?
The official Bot page lists eligible paid Cursor and Grok subscriptions, Teams plans and Enterprise. Confirm Bot access, the installed version, your usage allowance and any administrator restrictions in your own account. The release note does not establish universal free access.
Should I upload confidential customer material to make a deck?
Start with public, synthetic or appropriately redacted material. Check your organisation’s approved data rules and the current service terms before using confidential files. Give the workflow only the access needed for the task, and do not treat separate Bots as separate security boundaries.
Does a source link make an AI-generated slide accurate?
No. Open the source and check that it supports the precise claim, date, units and comparison shown. A genuine source can still be misquoted, outdated or used to support a claim it does not make.
What should I inspect after receiving a PowerPoint file or Google Slides?
Open the actual delivered artifact. Check every slide, speaker note, chart, link and image; look for clipping and incorrect figures. For a cloud presentation, inspect its owner and sharing permissions. Verify the version and intended recipients before sending.
How can a small team decide whether this workflow is useful?
Compare one bounded presentation task with its normal manual process. Record preparation time, human review and repair time, source errors and whether the final artifact is usable. Repeat with a missing or conflicting source; do not infer reliable savings from one attractive draft.
Start with one deck you can verify
The new Bot workflow gives existing eligible users another way to prepare a presentation. Begin with a small approved source pack, review the sample, then open the actual output and check its claims and handover. Keep the corrected version and review record together. Only widen the workflow when repeated evidence supports it.
[…] presentation tasks, the Grok Bot slide-deck review guide explains the 7 October update, an approved source pack, sample-slide checks and review of the […]