Peec AI, an AI search analytics platform, reports that YouTube videos with rewritten descriptions were retrieved by AI search far more often afterward. Daily retrievals for the updated videos rose 359%. Unchanged videos gained 17%, and the platform as a whole gained 25%, according to a Search Engine Land report published November 6, 2026.
The study is small, and its authors say it cannot prove the edits caused the gain. The sample covered roughly 60 YouTube videos, and on Oct. 19 Peec AI rewrote descriptions for 15 of them. Those edits also added manual timestamps, so the description text and the chapter markers are bundled into one treatment. The comparison pits Oct. 6-21 against Oct. 22-Nov. 4. Peec AI tracked retrieval shifts, yet it could not isolate what the description updates contributed.
Google’s own AI surfaces showed the same direction. In AI Overviews, retrievals for updated videos grew 293%, compared with 27% across YouTube. In AI Mode, Google’s conversational search experience, the figures were 129% and 11%. Percentage gains on a low starting base can look dramatic, and the report does not give absolute retrieval counts.
Two design details deserve a closer look. The baseline window runs through Oct. 21, two days after the edits went live, so some post-change activity may sit in the “before” period. Peec AI, which sells AI search analytics, also chose which videos to update, and the report does not say how. If the 15 were picked for topics already gaining attention, the lift would partly reflect selection rather than description quality.
None of that makes the result useless. It fits a plausible mechanism: AI systems retrieving video need text to work with, because they read descriptions and timestamps far more easily than they interpret footage. The 5,000-character description limit on YouTube leaves ample room, and the report notes that most videos fill little of it. A thin description is a cheap thing to fix, which lowers the cost of testing the idea.
The contrast with typical SEO evidence is worth stating. A correlation study from a vendor with a commercial interest is a hypothesis generator, not a ranking factor. Peec AI’s data does not show that any particular wording, structure, or length drives retrieval. It also does not show whether retrievals translate into views, clicks, or citations that bring viewers to a channel.
For a search team, the practical step is a controlled version of the same test. Pick a set of comparable videos, change descriptions on half of them, hold the rest constant, and track AI retrievals across both groups for at least a few weeks. Until a test like that exists, treat the 359% figure as a reason to run the experiment, not a number to put in a client forecast.
Search Engine Land, reporting by Danny Goodwin on a Peec AI study, published November 6, 2026.
