On the busiest shelf, the model and the surface agreed within a month

Twenty-eight distinct editors surfaced across the questions, and the reason is that "photo and video editing" is really two markets wearing one label. CapCut and Snapseed win the phone-first, social-video questions; DaVinci Resolve and LumaFusion answer the desktop-grade ones. The assistants sort by the job, not the app-store shelf, so a single leaderboard ends up mixing a free TikTok editor with a tool used in Hollywood colour suites.

On the chat models, CapCut leads, unsurprising for a ByteDance app built for exactly these questions. In June, Google's AI Overview, asked for the best video editor, didn't lead with CapCut. It led with Splice: "Splice is the most recommended all-around video editor due to its intuitive, desktop-style tools and direct social sharing," with CapCut named second as "the undisputed king of AI templates."

A month later, the same query: "CapCut is the top mobile and social media choice, DaVinci Resolve is the best fully free professional tool." Splice isn't mentioned.

I ran the capture again in July, same query, same method. Splice went from the Overview's lead answer to zero mentions across all six of the video and photo queries I track there. It didn't just drop behind CapCut, it dropped out of the Overview's answer entirely. Across the wider board Splice's score fell from 10 to 2 and Picsart, the other app that lost the most ground this month, fell from 14 to 2, both declines spread across three or four engines rather than one, so this isn't a single odd sample. Where that share went is diffuse: CapCut, Snapseed, Adobe Lightroom, Canva, iMovie, Premiere Pro and VSCO all picked up a few points each, no single winner.

I keep finding surface-versus-model splits that don't last. In the AI category, Google's Overview led with Gemini and dropped ChatGPT in June; a month later it led with ChatGPT and Claude. Here it's the same pattern in reverse: a gap that looked like a stable disagreement between the API and the box people actually see closed in a single cycle. Six queries a month is a small sample for a surface that regenerates its answer per search, so I'd be careful reading either the June gap or the July agreement as permanent. What it does argue for is checking back rather than trusting one snapshot to describe "the AI answer" for good.

The board has gaps the surface exposed

Crowded categories are also where a fixed roster strains hardest. Capturing Google's Overviews surfaced strong apps the chat models underweight or that I hadn't tracked at all: Adobe Express, Pixelcut, remove.bg, Luminar Neo, Photomator. I widened the roster from that signal mid-run. It's a reminder that in a deep category, what you're not measuring is its own finding. The leaderboard is only as complete as the names it knows to look for.

What this is

Thirty-four questions put to four AI assistants, plus Google's AI Overviews captured from the web for six core questions, run in June and again in July on the same frozen prompts. US English. The prompt set was validated against real Google autocomplete demand before it was fixed. See the photo & video index for the current board.