arXiv:2611.23289v1 Announce Type: new Abstract: Reviewers rely on pull request (PR) descriptions to understand what a change does, yet some changes, such as UI updates or runtime behavior, are hard to convey in text alone. PR authors attach visual artifacts to fill this gap, but how often they do so, what they show, and whether they change how a PR is reviewed remain unknown. We present a large-scale study of visual artifacts in GitHub PR descriptions. From 6,505,113 closed PRs in 5,012 repositories, we built VisPR, a dataset of 2,473,892 English, human-authored PRs from 647 repos where visual artifacts are an established practice. Of these, 329,185 (13.3%) are visual PRs, i.e., their descriptions contain at least one visual artifact (479,974 in total). By manually labeling a sample of 1,113 visual artifacts from 891 merged visual PRs, we derived two taxonomies: 11 content types capturing what developers show and six purposes capturing why they include visual artifacts. The dominant content type is Interaction & State, which shows how the interface responds to user actions. its share rises from 37.0% of images to 63.5% of GIFs and 79.7% of videos, whereas about a quarter of images (23.4%) show text such as logs, code, or error messages . Compared with 1,782 merged non-visual PRs matched on repository, PR type, and change size, visual PRs receive 43% more review comments and 17% more reviewers, and take 72% longer from first response to merge, time to merge and churn-normalized review efficiency do not differ significantly. LLM-assisted classification of 10,031 review comments in two groups shows that review centers on code in both, but comments on visual PRs address how the UI looks or behaves twice as often (6.4% vs.3.3%), and visual-output fixes after review are about four times as common (4.7% vs. 1.2% of PRs). Visual artifacts are thus associated with closer scrutiny of what users see, not with an overall faster review.