Your buyer asks AI before they ask you
In April, G2 published a survey of 1,076 B2B software buyers. 51% now begin their software research with an AI chatbot more often than with Google, up from 29% a year earlier, and 71% rely on chatbots for software research at some point (G2, The Answer Economy).
Google is moving the same way from the other side. When an AI Overview sits above the results, the top-ranking page now gets a 58% lower click-through rate on average, according to Ahrefs' re-run of its 300,000-keyword study comparing December 2023 with December 2025 (Ahrefs).
Put those together and the buyer's first impression of your category is increasingly a paragraph written by a machine. The companies named in that paragraph make the shortlist. The companies linked from it get the visit. Everyone else has to earn attention the expensive way.
The visits that do come through are worth more than normal search traffic. Ahrefs reported that AI search sent 0.5% of its site visitors over 30 days but produced 12.1% of its signups, a 23x conversion gap over traditional organic search (Ahrefs). Ahrefs sells to marketers who adopted AI early, so treat the multiple as an upper bound. The direction is what matters: someone who clicks through from an AI answer has already been told you are relevant.
For a B2B company with a podcast, this is either a problem or a large, mostly unclaimed opportunity. Which one depends on a single thing most shows get wrong.
AI does not listen. It reads.
When ChatGPT, Perplexity or Google's AI Mode answer a question, they search the web, pull passages from pages, and write a summary with citations. They work on text. An MP3 on a hosting platform contains no passages. A 45-minute interview can hold the best answer on the internet to your buyer's question and still be invisible, because the words only exist as sound.
Look at what a typical B2B episode gives a machine to read: a Spotify or Apple Podcasts listing with a two-line description, a YouTube upload titled "Ep. 47 with Jane Doe" with auto-generated captions, and maybe a LinkedIn clip. The 7,000 words the guest actually said are nowhere on the web as text.
The domains AI engines cite most show where that text needs to live. Peec AI analyzed 30 million sources cited across ChatGPT, Google AI Mode, Gemini, Perplexity and Google's AI Overviews. The top three were Reddit, YouTube and LinkedIn (Peec AI).
Two of the top three are places a B2B podcast should already be publishing. The podcast apps, where most shows put their effort, are not where answer engines look.
The podcast industry is catching on. At Podcast Movement in New York on September 17 and 18, World Podcast Network founder Bruce Chamoff told the room that Reddit is "the single most cited website across both AI and Google search results," and recommended building each episode around the exact question a listener or an AI tool would type in (Podglomerate's recap).
We asked AI 24 buyer questions. Here is what it cited.
Studies of millions of citations tell you what gets cited in general. We wanted to know what gets cited when a B2B buyer asks a B2B question. So on September 24, 2026 we ran the same 24 questions through ChatGPT, Perplexity and Google AI Mode: 8 asking what experts say about a topic, 8 how-to questions, and 8 vendor comparisons, across cybersecurity, SaaS, fintech, healthcare and B2B marketing. We logged every one of the 727 citations and checked that each cited YouTube video exists.
24 questions x 3 engines = 72 answers and 727 citations, collected on September 24, 2026. ChatGPT and Perplexity were used logged out, with web search on. Google AI Mode ran in a browser signed into a Google account, so its answers may be personalized. AI answers change from run to run. Read this as a snapshot of how the engines behave, not a benchmark.
Three findings stood out.
1. The engine changes everything, and ChatGPT reads company websites
Google AI Mode cited a YouTube video or podcast episode in 16 of 24 answers. Perplexity did in 6. ChatGPT did in none. YouTube alone made up 17.1% of everything Google cited.
ChatGPT cited far fewer sources overall, about 3 per answer against 11 for Google and 16 for Perplexity, and 68% of what ChatGPT cited were company websites. That is the most useful number in the test. ChatGPT is not skipping podcasts because it dislikes audio. It is citing pages on company websites. If your episode lives on your website as text, it is one of those pages.
2. Google cites moments, not videos
26 of the 45 YouTube links in Google's answers pointed to a specific timestamp, not the start of the video. For "How do I reduce churn in a B2B SaaS company?", Google attributed individual recommendations to named YouTube creators, inline, sentence by sentence.
The videos that won were not famous shows. They were small channels whose titles were the question: "How to Reduce SaaS Churn (4-Step Framework)", "SOC 2 for Startups: A Complete Guide to Building Trust", "How to Actually Sell Software to Hospitals". For "Vanta vs Drata for SOC 2", Google cited a comparison video from Vanta's own YouTube channel. (We took apart Vanta's content engine here. It keeps showing up.)
3. Company podcast pages get cited when the words are on the page
Perplexity cited Lithic's Fintech Layer Cake episode page for a question about fintech regulation, Demandbase's podcast page for an account-based marketing question, and UserTesting's podcast page on brand versus demand. Google cited SaaStr's interview page about hiring a first VP of Sales, which is titled, literally, "(Video + Transcript)". Every one of them is a B2B company publishing an interview on its own domain, with the words written out.
One more result, for honesty. We asked all three engines for the best B2B podcast production agencies. None of the answers cited a single podcast episode or video, and none named us. They were built entirely from agency listicles and directory pages. We run a podcast agency and we are not in that paragraph yet either. What follows is the system we are putting to work on our own show and our clients' shows.
The Citation Anchor: a name, a claim, a page
What makes one episode citable and another invisible? The best public data point is a small one. An agency tracked about 70 episodes from three client podcasts over six months after publishing cleaned transcripts on the clients' own sites. Transcript pages showed up in AI citations on roughly 11% of the queries they were relevant to, mostly in Perplexity and Gemini. The pattern mattered more than the rate:
"Episodes that were two co-hosts having a meandering conversation almost never got cited regardless of topic quality. Episodes with a guest making a clear, paraphrasable point cited well."
Tracking podcast transcripts through 4 AI engines over 6 months, DEV Community, May 2026
The author is upfront that three clients cannot support strong claims. It matches what we saw in our own test, and it matches how these systems write. An AI answer needs three things before it can cite you: someone to attribute the point to, a point worth attributing, and a page it can link to. We call that combination the Citation Anchor. Every episode should be built around one.
Name: a named expert the engine can credit
AI answers are full of attribution: "according to", "experts at", "as one CISO put it". A named guest with a title and a company can be credited. Two hosts riffing cannot. This is the structural advantage of the guest-interview format, and it is why your guest list does double duty. The people you want as customers are also the people buyers ask AI about. (If booking those guests is the hard part, here is how we do it with zero audience.)
- Put the guest's full name, title and company in the page title or the first line, in the transcript speaker labels, and in the episode description.
- Have the guest say it on the recording, too: "I'm [name], CISO at [company]." Captions and transcripts then carry it automatically.
- Get the spelling right. In one of our own interviews, YouTube's auto-captions turned the guest's surname, Leshchiner, into "Lesner" and his company, ElectroKare, into "ElectroCare". A machine cannot credit a name it cannot spell.
Claim: one specific point worth quoting
Engines lift sentences they can paraphrase and credit. Vague consensus gives them nothing to lift. The difference is usually one follow-up question.
| A claim no one cites | A claim built to be cited |
|---|---|
| "Security really is a team sport." | "Every AI agent gets treated like a new contractor: its own identity, the least access it needs, and an expiry date." |
| "Pricing is one of the hardest things for SaaS founders." | "We raised prices 30% and churn did not move. What moved churn was the first 14 days of onboarding." |
| "Regulators are paying more attention to fintech." | "Bring your compliance lead into the product spec, not the launch review. Retrofitting controls cost us a quarter." |
The examples above are illustrative. The pattern is real: a number, a named mistake, or a clear opinion with a reason. Plan for it before you hit record. Write three questions into every interview that force a specific answer: "What's the number?", "What do most companies get wrong about this?", and "If you had one sentence for a CFO, what would it be?"
Page: the words, on your domain, as plain text
The episode needs a home a machine can read. That means your own site, with the full transcript as real text, not a PDF, not an email-gated download, and not a tab that only loads with JavaScript. The best business shows already do this. Acquired publishes chaptered transcripts on its episode pages. Lenny's Podcast runs a speaker-labeled, timestamped transcript next to the player.
Neither of these is a B2B company's show, and that is the opening. Most B2B podcasts still stop at the player.
Worked example: rebuilding one episode page
Here is the rebuild on a typical B2B interview. The example is a composite: episode 31 of a cybersecurity vendor's show, where a hospital CISO talks about medical devices that cannot be patched. The names are placeholders. The steps are the ones to follow.
- Find the question the episode answers. Read the transcript for the moment the guest answers something a buyer would ask. Write it the way a buyer would type it into ChatGPT: "How do hospitals secure medical devices they can't patch?" One episode can answer two or three questions. Pick the strongest for the page and save the others for clips.
- Make the title the question. "Ep. 31: Securing Healthcare with [Guest]" becomes "How Hospitals Secure Medical Devices They Can't Patch". Put the guest's name, title and company directly underneath.
- Write the answer first. Open the page with about 120 words that answer the question in plain language and credit the guest by name. This is the block an engine is most likely to lift, so write it like the best paragraph in the article, not like show notes.
- Pull out three key claims. Quote the guest's three most specific sentences, each with its timestamp. These are your Citation Anchors, and they double as your clip list.
- Add chapters. Five to eight timestamped chapters named after what is said in them, not "Intro" and "Part 2". Use the same chapters in the YouTube description.
- Publish the full transcript as text. Clean it, label the speakers, correct every name and company, and keep the timestamps. Put it on the page itself.
- Link it into the archive. Add two or three related episodes at the bottom and link back from them. Chamoff recommended the same at Podcast Movement: it keeps listeners and crawlers moving through the back catalogue.
- Add the markup. Describe the video with VideoObject and mark each key claim as a Clip with start and end times. Google documents Clip markup as a way to show key moments for a video. Add the guest as a Person with a link to their LinkedIn profile.
Step 8, the markup (trimmed)
{
"@context": "https://schema.org",
"@type": "VideoObject",
"name": "How hospitals secure medical devices they can't patch",
"description": "[Guest], CISO at [Health System], explains how to isolate and monitor legacy devices.",
"uploadDate": "2026-09-24",
"duration": "PT46M",
"thumbnailUrl": "https://yourcompany.com/podcast/ep-31/thumb.jpg",
"embedUrl": "https://www.youtube.com/embed/VIDEO_ID",
"contributor": {
"@type": "Person",
"name": "[Guest Name]",
"jobTitle": "Chief Information Security Officer",
"sameAs": "https://www.linkedin.com/in/guest-profile"
},
"hasPart": [{
"@type": "Clip",
"name": "Segment every unpatchable device in 30 days",
"startOffset": 252,
"endOffset": 390,
"url": "https://yourcompany.com/podcast/ep-31?t=252"
}]
}A note on markup, because a lot of old advice is still circulating. Google dropped its podcast features in Search when it shut down Google Podcasts, and it stopped showing FAQ rich results in May 2026. Markup still helps machines understand a page. It will not win you a special search result by itself. The words on the page do the work.
Then test it. Run the page's question, plus a version that names the guest, in ChatGPT, Perplexity, Gemini and Google AI Mode. Save what comes back. Repeat at two, four and eight weeks. If you already turn every episode into a blog post and a newsletter, as we do for clients (here is that whole system), this is the same work pointed at a new reader.
Do the same on YouTube
YouTube is the second most-cited domain in Peec AI's ranking, it was 17.1% of Google AI Mode's citations in our test, and Google linked to exact moments more often than not. For most B2B shows, YouTube is also the platform with the widest reach. It deserves the same treatment as the episode page.
YouTube itself is moving in the same direction. At Made on YouTube on September 23, it announced Custom Feeds, which let viewers build a homepage tab by describing what they want, and "Your podcast lineup", a weekly spoken preview of recommended shows generated by AI. Even inside YouTube, discovery is turning into a conversation with a machine that reads titles, descriptions and captions.
- Title: the question, not the episode number. The winners in our test read like search queries.
- Description: the first two lines are your 120-word answer, trimmed, with the guest's name and title. Link to the episode page.
- Chapters: the same timestamped chapters as the page, in the description.
- Captions: upload your corrected transcript as the caption file instead of relying on auto-captions.
- Clips: cut each key claim as a Short and as a LinkedIn clip, with the quote written out in the post text. LinkedIn is the third most-cited domain on Peec's list.
If you are still deciding whether to film at all, our video versus audio breakdown covers the trade-offs. For AI search, the answer is simple: an audio-only show has one less place to be found.
How to tell if it is working
Being cited is not the goal. It is a leading indicator. "Working" means the same thing it always has for a B2B show: conversations with the right people, and pipeline you can trace back to the content. Track both layers.
| Signal | Where to look | How often |
|---|---|---|
| Citation log | 20 fixed prompts across ChatGPT, Perplexity, Gemini and Google AI Mode: questions naming your guests, the problems your episodes answer, and your category. Record whether your page, show or guest appears. | Monthly |
| AI referral traffic | GA4, Traffic acquisition, filtered to sources such as chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. Some AI visits arrive with no referrer and land in "direct". | Monthly |
| Self-reported source | Add "ChatGPT or another AI assistant" to the "How did you hear about us?" field and ask it on discovery calls. | Every lead |
| Branded search | Google Search Console queries containing your company, show or host name. | Monthly |
| Pipeline | Opportunities where the contact mentions an AI answer, an episode, or a guest. | Quarterly |
Expect the prompts that name your guests to move first, because few pages compete for them. Problem and category prompts take longer and depend on how many strong episodes you have on the topic. For connecting any of this to revenue, our podcast attribution playbook covers the CRM side.
What keeps a show out of AI answers
- Audio only, on podcast apps only. No podcast app made Peec's top ten, and podcast apps were 0.8% of the citations in our test.
- "Ep. 47" titles. An episode number answers no one's question.
- Transcripts nobody can reach. A PDF download, an email gate or a script-loaded tab might as well not exist.
- Show notes that summarize instead of saying something. "In this episode, we discuss the future of security" gives an engine nothing to lift.
- No guest, no claims. Two hosts chatting is the format the small study found almost never gets cited.
- Starting a show just for AI search. The study's author put it well: "launch a podcast if you have something to say and someone you'd want to interview." AI citation is a reward for a good show published properly, not a reason to start one.
None of this requires a new show. It requires treating every recording as source material for a page, a video and a handful of clips that a machine can read. If you are planning the show from scratch, start with our complete B2B podcast strategy guide and build the Citation Anchor in from episode one.
FAQ
Does ChatGPT listen to podcasts?
No. ChatGPT, Perplexity and Google's AI Mode answer questions by searching the web and reading text on pages. An audio file on a podcast app gives them nothing to quote. To be cited, the words from an episode need to exist as text: a full transcript and summary on a web page, plus captions and a description on YouTube.
Which AI engine is most likely to cite a podcast or YouTube video?
In our September 2026 test of 24 B2B buyer questions, Google AI Mode cited a YouTube video or podcast episode in 16 of 24 answers, Perplexity in 6, and ChatGPT in none. ChatGPT cited far fewer sources overall, and 68% of what it cited were company websites, which is why an episode page on your own domain matters most for ChatGPT.
Do I need to publish full transcripts of my podcast?
Yes, if you want AI engines to cite the episode. Publish the full transcript as plain HTML text on your own site, with speaker labels and timestamps, next to a short written answer and the key claims. Do not hide it in a PDF, behind an email gate, or in a tab that only loads with JavaScript. Upload the corrected transcript as your YouTube captions too.
Does podcast schema markup help with AI search?
It helps machines understand the page, but it will not earn a special search result on its own. Google dropped its podcast features in Search when it shut down Google Podcasts, and it stopped showing FAQ rich results in May 2026. The markup worth adding is VideoObject with Clip markup for key moments, which Google still supports, plus a Person entry for the guest.
How long does it take for a podcast episode to get cited by AI?
There is no reliable public benchmark yet. Log the same set of prompts at publish, then at two, four and eight weeks. Questions that name your guest usually move first because few pages compete for them. Broad category questions take longer and depend on how many strong episodes you have on the topic.