Evidence · strength 1 of 5 · searched to 2 Nov 2026
Does AI help people make better slides?
Not enough evidence yet to say whether AI helps people make better slides. 0 of 2 studies found a clear gain in quality of slides.
What little evidence exists on AI slides: a very small government test with blind marking, and one university course. No trial in a firm was found.
VerdictNot enough evidence yet
No good evidence
ImpactNo clear effect
0 of 2 studies found a clear gain in quality of slides.
Evidence strength1 of 5
We have very low confidence in this estimate.
Cost per user a year$$
These studies mostly used general AI assistants; business plans list about $20–$30 a user a month (checked 2 Nov 2026).
Pilot rating under our v0.3 rules. The rules are published on our method page but not yet independently reviewed, so ratings may change as the method is checked and new studies arrive. Our method.
What the studies found
Quality (headline)
No clear effect- Business and Trade (2025): −1 points on a 1-5 scale Observed task: quality score of the slide, blind-marked 1-5 (Copilot 1 vs non-user 2) (no interval; 3 vs 3 people). Not statistically clear.
PowerPoint 1.5 5 1 2 10:47 18:30
- Business and Trade (2025): +54% Diary study: share of slide-deck tasks users were satisfied or very satisfied with (54%, against about 80% across all uses) (81 task records). Self-reported. Not statistically clear.
only 54% and 57% of recorded cases respectively reporting being satisfied or very satisfied
- Leinonen et al. (2026): +0 points on a 7-point rating scale, AI minus instructor Student rating of slide segment quality, AI-generated (edited) vs instructor-made, 7-point scale (mean 5.1 for both; median 5.0) (Mann-Whitney U = 25600, p = 0.84). Self-reported. Not statistically clear.
we failed to find evidence that the slides were of different quality
- Leinonen et al. (2026): +0 no significant difference in guesses between AI and human slides Students' guess whether slides were AI-made (1 definitely human to 7 definitely AI): could they tell? (Mann-Whitney U = 28171.5, p = 0.1). Self-reported. Not statistically clear.
students seemed not to be able to guess the source (AI vs. human) of the slides
Errors
No clear effect- Business and Trade (2025): −3.5 points on a 1-5 scale Observed task: accuracy score of the slide, blind-marked 1-5 (Copilot 1.5 vs non-user 5) (no interval; 3 vs 3 people). Not statistically clear.
PowerPoint 1.5 5 1 2 10:47 18:30
Time
No clear effect- Business and Trade (2025): −41.7% Observed task: time to produce one PowerPoint slide from an Excel analysis, Copilot users vs non-users (10:47 vs 18:30 mm:ss) (no interval; 3 vs 3 people). Not statistically clear.
PowerPoint slides over 7 minutes faster on average, but to a worse quality and accuracy than non-users.
- Business and Trade (2025): +0 hours saved per task Diary study: mean time saving per slide-deck task, adjusted for unused outputs (no interval; 78 task records). Self-reported. Not statistically clear.
Again, this conflicts time savings reported in the diary which showed neutral PowerPoint time savings.
- Checking time not measured: none of these studies counts the time spent checking the AI’s output separately.
A minus sign on time means the task took less time. Bands are provisional; see how impact is banded. Effects are not pooled yet: each band is set from one result per study, and where studies use different units we show only the direction.
How sure are we?
- Starting level1 of 5, from 2 studies, including grade D.
- Consistencywith 2 studies, never downgraded on this test alone.
- Publication biasnot testable with fewer than 10 studies.
- Result1 of 5. We have very low confidence in this estimate.
Behind the average
None of these studies reports results for different groups of people or tasks.
How, where and why it works
Why it works
1 of 4. AI turns text or data into a slide layout; people edit it.
For whom and where
0 of 4. No study tests who gains.
What it takes
0 of 4. Neither study follows use over time.
The full cost
0 of 4. No study counts cost.
Four separate scores from 0 to 4, never added up: how much the studies tell us about why it works, for whom, what it takes and what it really costs (the EMMIE method).
What to do next
Measure your pilot →If you are trying it yourself
- Write the argument yourself, then ask for the layout.
- Check every figure against the source.
If you are rolling it out
- Blind-mark slides made with and without AI on a task of your own.
What it takes
- Check every number on a slide against the source.
- Plan time for editing what AI produces.
What can go wrong
- In a test of six civil servants, Copilot users were faster, but their slides scored lower on quality and accuracy.
- These results rest on very few people, so they could be chance.
The studies
| Study | Grade | Design | People | AI model (released) | Company role |
|---|---|---|---|---|---|
| Evaluation of the M365 Copilot pilot in the Department for Business and Trade: observed PowerPoint task and diary-study slide-deck results UK Department for Business and Trade, 2025. UK Department for Business and Trade evaluation report (government report) | D | other quasi-experiment, realistic task | PowerPoint observed task: 3 Copilot users vs 3 non-users (control colleagues named by diary respondents); 11 people took part across all observed tasks. Diary study of the 1,000-licence pilot: 78 task records on slide decks for time, 81 for satisfaction | Microsoft 365 Copilot in PowerPoint (model not stated) | none |
| AI-Generated Slides: Are They Good? Can Students Tell? Juho Leinonen, Lisa Zhang, Arto Hellas, 2026. Western Canada Conference on Computing Education (preprint) | D | other quasi-experiment, deployed work | Course of 88 enrolled students (51 attended the first lecture). Students rated 22 lecture segments: 83 ratings for Cursor-made and 116 for Claude Code-made segments, plus instructor-made segments (16-28 ratings each) | Claude Sonnet 4.5; NotebookLM; M365 Copilot; Cursor; Claude Code (models behind the coding assistants not stated) (Sep 2025) | none |
Grades: A, a well-run randomised trial in real work; B, a strong natural experiment in real work or a well-run trial on realistic tasks; C, a trial with clear limits or a matched comparison; D, before-and-after, surveys or case studies. How studies are graded.
Does the rating hold up?
- Run without company-run studiesNo change: no such studies.
- Run without working papers and preprints1 studies: strength 0, verdict no good evidence, headline band no clear effect.
How up to date is this evidence?
Method →Evidence searched to 2 Nov 2026. Studies cover models released Sep 2025 to Sep 2025. Rating computed 5 Nov 2026 under Module 15 v0.3 (pilot).
How this page was made: AI research agents searched for studies and read each paper; a second pass checked every number on this page against the paper itself, with the quote and where it appears. The rating is computed by fixed rules, not chosen. No person outside AI Adoption has reviewed it yet. Editorial owner: AI Adoption editors. Spot a mistake? Corrections.
Disclosure: AI Adoption uses AI tools, including Claude from Anthropic, in its research and editorial work. No company pays for or has any say in a rating.