How a B2B podcast can earn AI citations: transcripts, YouTube, show notes

Give each episode a form AI products can find and quote: a full transcript on its own web page, a YouTube upload with accurate captions and a detailed description, and show notes with specific claims. We found no independent study proving this earns citations, but AI engines cite pages and videos, and audio alone offers little to cite.

What does the evidence actually say?

It is better to be plain about this first. We looked for a large independent study measuring whether podcast transcripts lead to AI citations and did not find one we could verify. Some firms publish figures on this, but we have left out numbers whose method we could not confirm. Here is what we did read.

• OtterlyAI, a vendor of AI tracking software, published a YouTube citation study covering 100M+ citations over 30 days. It is about YouTube videos in general and says nothing specific about podcasts.
• Ahrefs, which sells AI brand tracking, analyzed 75,000 brands and found that YouTube mentions (about 0.737) and branded web mentions (0.656 to 0.709) correlated with AI visibility. The article itself notes that correlation is not causation, and well-known brands would naturally have more of both.
• Rephonic, a podcast database, argues that titles, descriptions, show notes, transcripts and YouTube pages create text that can surface in AI answers. This is a reasoned argument from a company with something to sell, and it does not report a test.

So the case is logical rather than proven. The steps below cost little if you already produce a show, and most of them help human listeners and Google anyway.

Why does an audio file on its own fall short?

Perplexity describes gathering information from sources such as articles, websites and journals, then linking to them with numbered citations. Google's grounding documentation describes search results and URL citations. Both of these describe retrieval and citation of web sources. Your best insights, if spoken in episode 43 and available only as audio, may never enter that process in a usable way. We cannot say what each engine does with audio, but a page of text removes the uncertainty.

How should you publish transcripts?

Give every episode its own page on your domain, with the full transcript in the page's HTML text rather than behind a click-to-load widget, a PDF or a third-party player. Then do the following:

• Correct the transcript. Fix names, company names and product terms, since speech recognition can get these wrong and a misheard brand name helps no one.
• Add speaker labels so that a quotation can be attributed to the right person.
• Add subheadings at each topic change, phrased the way a buyer would ask the question.
• Put a short summary at the top that states who spoke, what was covered and the main conclusions.
• Link to the relevant product, service or resource pages on your site and from those pages back to the episode.

Keep in mind Google's position that its AI features only require a page to be indexed and eligible for a snippet. Check that your episode pages are indexed in Search Console, because a page that is not indexed cannot be cited through search.

What should you do on YouTube?

Publish full episodes, not just clips. In OtterlyAI's study, long-form videos made up 94% of YouTube citations and Shorts 5.7%, with Shorts appearing mostly in Google AI Overviews and AI Mode. Clips still have a promotional job, but they should not be your only YouTube presence.

The same study has some encouraging findings for smaller shows. Views, likes and subscribers had almost no correlation with citation frequency, at about -0.03, and about 41% of cited videos had under 1,000 views. Description length showed a weak-to-moderate correlation of 0.31, and the average description ran 334 words. The study warns that it only covers videos already cited, so it explains repeat citation and not how a video is first chosen.

Practical steps:

1. Check the captions. YouTube creates automatic captions using speech recognition, and you can upload a corrected transcript or file instead.
2. Write a real description. Say who the guest is, name their company, summarize the main points and link to the episode page.
3. Add chapters. YouTube's rules require a first timestamp at 00:00, at least three timestamps and at least 10 seconds per chapter.
4. Write titles that state the topic. A title such as "Episode 87 with Sarah" tells nobody what is inside.

OtterlyAI found timestamped citations only in Google AI Overviews and AI Mode, so chapters are most likely to matter there. Google's video guidance also says you can add timestamps and labels in YouTube descriptions to define key moments.

What makes show notes useful?

Weak show notes read like an advert: "In this episode, we chat with Jane about all things supply chain." That sentence gives an answer engine nothing to use. Stronger notes contain specifics that could stand alone:

• The guest's full name, title and company, spelled consistently across every platform.
• Three to five concrete takeaways, written as complete statements.
• Any data mentioned, with its source and a link.
• A direct quotation or two, attributed correctly.
• Links to tools, reports and companies named in the conversation.

This lines up, loosely, with the original GEO research, where adding quotations, statistics and cited sources tended to raise visibility in the authors' experiments. Those tests were on rewritten web content and not podcasts, so read it as a direction. Never add a statistic your guest did not actually provide or that you cannot source.

Which episodes deserve the most effort?

You likely cannot give the full treatment to a whole back catalog right away. Start with the overlap between what your show has covered and what buyers ask.

1. List the questions your buyers ask AI assistants, using your sales calls as the source.
2. Search your transcripts for places where a guest or host answered those questions well.
3. Upgrade those episode pages first.
4. Turn the strongest answers into standalone articles that link back to the episode. A focused page answering one question is often easier to cite than an hour-long transcript covering twelve topics.

For future episodes, plan at least one segment around a real buyer question. This can also improve the show, since it forces specificity.

Which engines will this help most?

Expect uneven results. OtterlyAI attributes most YouTube citations to Perplexity (38.7%) and Google AI Overviews (36.6%), with Google AI Mode at 19.6%, ChatGPT at 4.4% and Gemini at just 0.2%. Meltwater's September data, using a different method and consumer prompts, instead found YouTube leading on Gemini and reported YouTube citations down 57.3% from August. Both cannot be read as stable fact. Do not build a plan that depends on one platform.

That is a further argument for transcript pages on your own domain: OtterlyAI's index found company-owned pages made up 53.7% of all citations. Your site and your YouTube channel can work as two separate routes.

How do you know whether it is working?

Pick 20 or so questions your episodes address and test them on each engine before you start. Repeat at a set interval, and record whether your brand is named, whether an episode page or video is cited, and which competitor sources appear. Also check Search Console and look for AI assistant referrals in GA4. Change one thing at a time when you can, and do not expect a clean causal story.

Fame runs B2B podcasts, with 300+ shows and 100K+ episodes behind us, and publishes this article, so we have an interest here. Fame Search can use a client's podcast transcripts as source material, but you do not need a Fame podcast to use it. We have no client results to report on this yet. Treat the steps above as practices to test.