ChatGPT Slides can make an editable presentation from a conversation, outline or source files, but the deck still needs a source check. OpenAI announced collaborative Slides on October 16, 2026. Eligible Pro, Business and Enterprise users with Space access can create and edit a deck in Work or Space; rollout is gradual. On the web, OpenAI lists export to PowerPoint, PDF and Google Slides. OpenAI release notes Slides help
OpenAI says you can change text, shapes, tables and charts directly, request edits to a selected element, share a deck with editors and leave comments. Those features make correction possible; they do not verify a figure or claim on your behalf. This article uses a fictional source brief and authored answer key, not a deck AIViewer produced in ChatGPT.
The source notes: a fictional project update
A community learning group needs a six-slide update for a funder. These short notes are the entire source for the exercise:
- Quarter 1: Eight workshops; 96 attendances in total.
- Quarter 2: Ten workshops; 110 attendances in total.
- Budget: $1,800 approved; $1,650 spent by the end of Quarter 2.
- Not measured: Unique people, learning outcomes, satisfaction or the reason attendance changed.
“Attendances” counts visits. A person attending twice may be counted twice. The notes do not justify a claim about unique people reached.
The two useful calculations are 96 ÷ 8 = 12 attendances per workshop in Q1 and 110 ÷ 10 = 11 in Q2. The total rose by 14 attendances, while the per-workshop average fell by one. The unspent budget is $1,800 − $1,650 = $150. None of these numbers explains why the change occurred.
A six-slide plan with an evidence trail
This is an authored reference plan. It is the result to request and check, not a screenshot or observed ChatGPT deck.
| Slide | Intended message | Source or calculation | Review limit |
|---|---|---|---|
| 1. Scope | Two quarters of community workshops | Source notes | Call them attendances, not unique attendees. |
| 2. Activity | Workshops rose from 8 to 10 | Q1 and Q2 workshop counts | Use the same scale for both bars. |
| 3. Total attendance | Attendances rose from 96 to 110 | Q1 and Q2 attendance totals | Label totals as visits/attendances. |
| 4. Per-session view | Average moved from 12 to 11 | 96/8 and 110/10 | Do not say every session had exactly 12 or 11 people. |
| 5. Budget | $1,650 spent from $1,800 approved; $150 unspent | Budget notes and subtraction | “Unspent” is not a cost saving claim. |
| 6. Next questions | Check unique people and outcomes before claiming reach or impact | “Not measured” note | Present as a proposed measurement step. |
A deliberately flawed authored caption for slide 3 would be: “Engagement improved 15% in Q2.” The total attendance increase is 14 ÷ 96 = 14.58%, or about 14.6%, but “engagement” was not measured. More workshops could produce more total visits even as the average per workshop falls. A faithful caption is: “Total attendances rose from 96 to 110 across two additional workshops; the average per workshop fell from 12 to 11.”
Request the deck, then request corrections
For an eligible account, OpenAI’s help page says to ask ChatGPT in Work to “create ChatGPT slides” from the notes, or use Create → Slide in Space. Tell it the audience, length and required source material. A bounded request for this exercise is:
Create six editable ChatGPT Slides for a funder update from the fictional notes below. Use the six-slide plan in this article. Put the source figure or formula in a visible small source line for each chart. Keep total attendances separate from average attendances per workshop. Do not claim unique reach, outcomes or a cause for the change. Leave the deck for review; do not share or export it yet.
If a slide says “engagement improved,” select the caption and request a precise change: “Replace this with the verified total and average comparison. Remove any claim about engagement or unique people.” OpenAI documents direct editing and element-level edit requests; whether the actual deck follows them accurately still has to be checked. Slides editing guidance
Check the file people will receive
Read all six slides at presentation size and scan the tables/charts after each edit. Verify that each number and unit matches the notes or the displayed arithmetic. Then inspect the exported PowerPoint, PDF or Google Slides file, because formatting and labels can change in a handoff. OpenAI specifically advises reviewing content and formatting after export. Slides export guidance
Try one final exercise: suppose Q2 had 120 attendances instead of 110, with ten workshops. What changes? Answer: Q2 average becomes 12 per workshop, matching Q1’s 12; total attendances rise by 24, which is 25% of Q1’s 96. The budget figures do not change. A deck that updates only the total chart but leaves “average fell” on slide 4 would now be inconsistent.
If Slides is absent, check plan, Space access and rollout before writing a workaround as though it were the same product. OpenAI notes a Windows desktop distinction: some local environments create a PowerPoint file instead of a collaborative Slides deck; web or cloud mode can provide the collaborative route. The source-to-slide ledger above remains useful with any presentation tool. For source fidelity before making slides, see our report-summary exercise.