Presentation research and statistics 2026
Four sourced findings about editing presentations and evaluating AI generated slides. Each number includes its unit, source, and limits so you can trace it back to the original research.
By slAIdes · Reviewed September 30, 2026
Published research
122%
PowerPoint activity rises just before meetings
Microsoft reported a 122% spike in PowerPoint view and edit actions in the final 10 minutes before a meeting. This is an activity comparison across fixed pre-meeting windows, not the share of presentations edited at the last minute.
Method and scope: Microsoft says it calculated a rolling 28-day sum of PowerPoint view and edit actions per meeting participant across fixed windows before meetings. The report does not disclose the underlying action count for this figure.
Original source: Microsoft Work Trend Index, April 23, 2025585
Test examples in a structured slide benchmark
The SlidesBench researchers assembled 585 test examples from a corpus of 310 slide decks spanning 10 domains; they also report about 7,000 training examples. These are benchmark examples, not a count of people or presentations made worldwide.
Method and scope: The authors derived instruction and slide examples from their selected deck corpus, then evaluated image and program generation approaches. Their paper reports that programmatic methods produced higher-quality user-interactable slides in their tests.
Original source: Ge and colleagues, AutoPresent, revised June 19, 2025238
Source-backed slide generation tasks in PresentBench
PresentBench contains 238 evaluation instances, each with background material for creating slides. Its authors wrote an average of 54.1 binary checklist items per instance to check specific qualities of generated decks.
Method and scope: These counts describe the authors' research benchmark and its manually designed rubrics. They do not measure the accuracy or quality of slAIdes decks.
Original source: Chen and colleagues, PresentBench, March 7, 20268,133
Source-grounded checks in an audience-aware benchmark
The X+Slides researchers report 8,133 deduplicated, source-grounded evaluation probes across 113 topics and seven presentation scenes. The probes test generated presentations against the intended audience and source material.
Method and scope: These are evaluation probes created for the study, not measured outcomes from presentation software users.
Original source: X+Slides research paper, June 2026slAIdes measurements
We have not published a measured study of slAIdes customer presentations. The figures above come from the linked publishers; none describes slAIdes usage or performance. We will add original findings only when we can publish the sample, timeframe, calculation, and missing-data rules alongside them.
This is a curated reading page, not a pooled statistical analysis. The studies use different populations and methods, so their figures should not be combined. For the product behind this page, readabout slAIdes.