Does updating YouTube descriptions really boost AI retrievals by 359%?

Updating descriptions probably helps, but the 359% figure is not proof of that on its own. Peec AI changed two things at once (descriptions and timestamps) on just 15 videos over a two-week window, so the stat shows correlation, not a clean cause. The direction is backed by other research on YouTube and AI citations, but the exact number shouldn't be treated as a guarantee.

What Peec AI's 359% study actually measured

Peec AI, an AI search analytics platform, ran an experiment comparing YouTube videos with updated descriptions against videos left unchanged, then checked how often each group showed up in AI retrievals, according to Search Engine Land's coverage of the study. The headline claim is that videos with updated descriptions saw a 359% increase in daily AI retrievals, compared with 17% for unchanged videos and 25% for YouTube overall, per the same report.

Broken down by platform, the lift was uneven. Google AI Overviews retrievals increased 293% compared with 27% overall, and Google AI Mode retrievals increased 129% compared with 11% overall, as reported by Search Engine Land.

The exact numbers behind the headline

The sample was small. Peec AI GEO researcher David Konitzky shared the findings on LinkedIn, and the firm analysed about 60 YouTube videos, updating descriptions and adding manual timestamps to 15 of them on 19 October, according to Search Engine Land. The comparison window ran from 6-21 October against 22 October-4 November, a short stretch for a retrieval metric that can move from run to run. That instability is part of a wider pattern we've written about before: when AI engines barely agree with one another, a single short-window measurement can shift substantially without anything on the publisher's side actually changing.

Why critics say the stat is misleading

Search Engine Land itself flags the key limitation: the study measured changes in AI retrievals but couldn't isolate the impact of the description updates specifically. The confound is baked into the methodology. Peec analysed roughly 60 videos, and on 19 October it updated descriptions on 15 of them while adding manual timestamps to the same 15, according to AEO Sherpa's critique of the study.

One independent critic argues the timestamps may be doing the heavy lifting: "My guess is that the timestamps are doing as much work as the descriptions, because they hand an engine clean, labelled chunks of a video to retrieve," according to AEO Sherpa. To his credit, Konitzky himself doesn't oversell the finding. He wrote that "of course, this doesn't prove that longer descriptions alone caused the increase. Simply filling all 5,000 characters isn't the goal," per the same AEO Sherpa piece.

There's a conflict-of-interest note worth flagging too. The author of that critique runs a competing analytics agency, ViewEngine, and points out that Peec also sells AI search analytics, which colours how the result gets marketed. The same piece notes that Google Search Console data describes performance in Google's own search surfaces, not AI Overviews, AI Mode or ChatGPT, so Search Console can't confirm or refute Peec's AI retrieval counts. The critic's bottom line: Peec showed that videos it edited saw more AI retrievals over a short window, but it didn't show which edit mattered, and one run can't show the lift would repeat, since AI visibility moves around from run to run.

What corroborating research says about YouTube and AI citations

A separate, larger-sample study from OtterlyAI offers a more cautious read on timestamps specifically. It found no timestamped YouTube citations at all in ChatGPT, Gemini, Microsoft Copilot or Perplexity during its study window, according to OtterlyAI's research. It did find a metadata signal, though weaker than Peec's headline number: 31% of all cited videos in its dataset contained timestamp signals, meaning timestamped citations and/or chapter-style timestamps in the description. A third-party summary of that same dataset reports correlation coefficients rather than percentage lifts, finding description length (r = 0.31) and hashtag presence (r = 0.20) are the only metadata variables with meaningful correlation to citation frequency, according to Radyant's guide to YouTube AI search visibility.

On the platform split, Google AI Overviews and Perplexity drive more than 75% of YouTube citations, while ChatGPT contributes just 4.4%, per the same Radyant guide. That matters because it means Peec's result, strongest in Google's AI surfaces, sits in exactly the platforms where YouTube citations are most common anyway.

Konitzky has done separate work suggesting ChatGPT weighs metadata over full transcripts. He believes a ChatGPT algorithm prioritises videos with more views but also considers subscriber count and verification status, with verified channels receiving more citations, and found that a third of ChatGPT citations came from channels with around 1,000 followers and 10 million cumulative views, according to Practical Ecommerce's interview with Konitzky. He also observed a correlation between updating descriptions and AI Mode citations specifically, noting that AI Mode can access complete transcripts, but well-structured, detailed descriptions are easier to retrieve.

YouTube's overall importance to AI answers is well established beyond this one study. In a larger, separate analysis of 30 million citation sources, Peec AI found Reddit ranks as the most-cited domain in AI-generated answers, followed by YouTube and LinkedIn, according to Search Engine Land's coverage of that study, which also notes that YouTube dominates video citations via transcripts and descriptions. Independent researcher Jakob Nielsen's analysis found YouTube is the single most common domain cited by ChatGPT answers, at around 11.3% of all references, per his substack analysis. For Google's AI Overviews, the same analysis put Reddit at the top with around 21% of top citations, with YouTube next at around 19%. If you're already weighing Reddit's role in AI answers, it's worth reading how Reddit's login wall could cut AI search engines off from its content, since the two platforms often sit side by side in citation studies.

Description quality and length correlate with citation frequency in OtterlyAI's larger dataset, Konitzky's separate ChatGPT research shows metadata matters more than transcripts, and Peec's own 30-million-source study confirms YouTube is already a heavily cited domain. What none of this proves is that a description rewrite alone produces a 359% jump: that number came from 15 videos, two changes at once, and a two-week comparison window.

Our view: treat the 359% figure as a directional signal, not a target. Updating stale, thin descriptions is low-risk and plausibly useful based on the correlational evidence, but anyone promising a repeatable 359% lift from descriptions alone is reading more into one small test than the test supports. This fits a broader pattern we keep seeing in GEO: tactics that produce a visible citation are not the same as tactics that produce a result a business can bank on, which is also the thrust of why being cited by AI isn't the same as getting recommended.

How to apply this to your own videos

A GEO vendor's practical guidance, directionally aligned with Peec's test, recommends expanding descriptions to 500+ words with structured sections and key entities, adding 5-8 timestamped chapters with keyword-rich titles, adding topic-relevant hashtags, and reviewing the auto-generated transcript for accuracy, according to Radyant's YouTube AI search framework. YouTube allows descriptions of up to 5,000 characters, though many videos use only a fraction of that space, per Search Engine Land. A separate GEO practitioner guide cautions against writing descriptions as summaries; a good description should be a standalone document that answers the same question the video answers, in readable prose, according to Viewership.ai's guide to using YouTube for GEO.

That emphasis on substance over box-ticking matters generally in GEO. Technical fixes that look good on paper don't always move the needle, which is roughly the lesson from why llms.txt still doesn't boost AI search visibility despite being an easy, low-effort addition. Description rewrites are cheaper and more directly tied to what a video is actually about, which is probably why the correlational evidence here is stronger, but the comparison is a useful reminder not to assume every plausible-sounding GEO tactic pays off equally. It also echoes the finding that AI search engines tend to trust earned, substantive content over anything that reads like marketing copy; a description written as a genuine standalone answer is closer to earned content than a description written to game a character limit.

The limitation worth being honest about: none of this evidence isolates descriptions from timestamps, so if you only have time to do one thing, you don't actually know from this data which lever to pull first. Do both, and don't expect to attribute the result cleanly afterwards.

Frequently asked questions

Did Peec AI prove that descriptions alone caused the 359% increase?

No. Peec AI changed descriptions and added manual timestamps to the same 15 videos on the same day, so the study can't separate which change drove the result, according to both Search Engine Land and AEO Sherpa's critique.

Do timestamps matter more than descriptions?

It's unclear. One critic suspects timestamps do most of the work because they hand engines clean, labelled chunks to retrieve, per AEO Sherpa, but OtterlyAI's larger study found no timestamped citations at all across ChatGPT, Gemini, Copilot or Perplexity, while still finding 31% of cited videos had timestamp signals in the description.

Which AI platforms cite YouTube the most?

Google AI Overviews and Perplexity drive over 75% of YouTube citations combined, while ChatGPT contributes just 4.4%, per Radyant's analysis. Separately, Jakob Nielsen's research found YouTube is ChatGPT's single most-cited domain at about 11.3% of references.

Is there a conflict of interest in Peec's study?

The critique raising this point comes from a competing analytics agency, ViewEngine, whose author notes that Peec also sells AI search analytics. That doesn't make the data wrong, but it's worth knowing when weighing how the result was marketed.

Should I rewrite my YouTube descriptions based on this study?

It's a reasonable, low-risk move supported by correlational evidence from multiple sources, including OtterlyAI's finding that description length correlates with citation frequency. Just don't expect a guaranteed 359% lift, since that figure came from a small, short, confounded test.