Why Gemini cites YouTube so often and what to do with your episodes

Nobody outside Google can say exactly why, and it is not settled that Gemini does cite YouTube often. Meltwater ranked YouTube first on Gemini in September 2026, while OtterlyAI found Gemini almost never cited it. The practical advice is the same either way: publish full episodes with accurate captions, detailed descriptions and chapters, and pair each with a page on your site.

What does the data show?

Here are the two main datasets side by side, along with a third that tracks direction.

All three companies sell in or around this market. Note also that the first and second measure different things: Meltwater ranks domains within Gemini, while OtterlyAI shows how YouTube's citations split across engines. Even so, the two pictures are hard to reconcile, and timing may explain some of it, since the studies are six months apart and Evertune saw a rising trend in between. A later OtterlyAI experiment with a German broadcaster's 610 videos also found no citations from Gemini, although its Gemini data only began on July 25 and it covered German finance prompts alone.

The fair conclusion is that YouTube looks important on Google's AI surfaces generally, while the result for Gemini specifically depends on when and how you measure.

Why might Gemini favor YouTube?

Google has not published an explanation in anything we reviewed, so these are hypotheses and not facts.

• Shared ownership. Google owns YouTube, so it would not be surprising if its products found YouTube content easy to reach. This is plausible, but ownership alone does not explain the changes over time.
• Search-based retrieval. Google's grounding documentation says the model can generate queries, run them on Google Search and cite the results. If YouTube pages rank for a query, they could then be available for citation. That describes the developer tool, not necessarily the consumer app.
• Coverage. Long video pages include titles, descriptions and possibly captions on topics where few written pages exist. How-to and explainer questions would naturally benefit.
• Model changes. Evertune links its shifts to model versions, noting that newer versions can behave quite differently from their predecessors, and that domains can affect sourcing by changing scraping permissions.

The last point is the most useful one for planning. If model updates can move the numbers this much, you should not rebuild your content program around one month of data.

How stable is this pattern?

Not very. Meltwater reported that YouTube's total citations fell 57.3% from August to September 2026, and that its share declined across all six models with meaningful YouTube activity, including Gemini. It still remained the single most-cited source overall. So YouTube may be both the largest source and a falling one in the same report.

Also keep in mind what a citation does and does not prove. Evertune cautions that not every prompt triggers a web search and that a cited URL does not necessarily shape the response. A link in a source list is a sign of visibility, and it may not mean the video changed what the answer said.

Finally, Meltwater's prompts covered consumer categories such as banking, fast food and streaming. We did not find a comparable public dataset for B2B questions, so test your own.

What should you do with your episodes?

If you already produce a podcast or video series, these steps are low cost and useful beyond Gemini. Most also reflect what OtterlyAI saw among cited videos across engines.

1. Upload full episodes. Long-form videos took 94% of YouTube citations in OtterlyAI's study, against 5.7% for Shorts.
2. Correct the captions. YouTube generates automatic captions with speech recognition, and you can supply a transcript instead. Check names and product terms carefully.
3. Write a full description. Description length had the strongest correlation in the study at 0.31, weak to moderate. The average among cited videos was 334 words. Length alone is not the goal, so write a useful summary rather than padding.
4. Add chapters. YouTube requires a first timestamp of 00:00, at least three timestamps and chapters of at least 10 seconds. OtterlyAI found timestamped citations only on Google AI Overviews and AI Mode, with none on Gemini, so chapters may do more for those surfaces.
5. State the topic in the title. Name the question or subject, not just the episode number and guest.
6. Keep publishing. OtterlyAI reports a correlation of about 0.3 with recency, so older catalogs may need refreshed companions.

Should you make a page for every episode?

Yes, where the episode matters to buyers. Google's video guidance recommends a dedicated watch page where it makes sense for your business, with a unique title and description, and VideoObject structured data that matches the video. It also notes that a blog post with an embedded video does not count as a watch page, so decide which role each page plays.

Your own page also protects you against platform swings. Yext found 52.15% of Gemini's citations came from brand-owned websites in its dataset, and OtterlyAI's index found company-owned pages were 53.7% of citations across engines. Those are vendor studies with their own limits, but they make a clear case against relying on YouTube alone. If you publish the transcript and summary on your site, you have two chances at a citation rather than one.

Do view counts matter?

Less than you might expect, at least among videos already cited. OtterlyAI reports correlations of about -0.03 for views and subscribers, about 41% of cited videos under 1,000 views, and a median cited channel with 41 videos. That is encouraging for a B2B channel with a niche audience, since a specialist B2B topic may never attract a large audience. The study warns that this explains repeat citation among already-cited videos and not how a video gets chosen in the first place. Do not read it as "views never matter."

How can you test this for your company?

The studies above cannot tell you what happens for your category, but a small test can, and it costs little beyond an hour or two.

1. Choose 10 to 20 buyer questions your episodes actually answer.
2. Ask them in Gemini several times, and also in Google AI Mode, ChatGPT and Perplexity.
3. Note any YouTube links, and whether they go to your channel, a competitor or a third party.
4. If videos do show up, look at what those cited videos have in common.
5. If none show up on Gemini, do not spend more on it for that engine's sake. Look at what it does cite, and repeat in a month.

Fame Search tracks ChatGPT, Perplexity, Claude and Gemini weekly, and Fame also produces B2B podcasts, so we have an interest in this topic. Fame has no client results to report on it. For a broader comparison, see how ChatGPT, Perplexity and Gemini differ and how a B2B podcast can earn AI citations.