If you ask ChatGPT or Google's AI Overviews a question your podcast answers well, your show probably won't come up. These systems answer by reading text on the web, and a show that exists only as audio inside Spotify and Apple Podcasts gives them almost nothing to read. The fix is to publish crawlable pages with the content of your episodes in plain text, clear answers near the top, and an obvious description of who you are. No one can guarantee a citation. Being readable is the part you control.
Why podcast AI search can't hear your audio
Most AI answer engines work with text. Ask one a question and it either draws on what it learned in training or fetches fresh web pages, pulls out the relevant passages, and writes an answer with links to its sources. The second route is growing. Google calls its version AI features in Search, and it says the same basics that apply to regular Search apply here: your page has to be indexed and eligible to appear in results.
An audio file is a poor input for that process. Even when a crawler can fetch an MP3, it can't quickly work out whether minute 41 contains a good answer about sourdough hydration. A Spotify or Apple Podcasts listing doesn't help much either. Those pages are thin, they're built around the app's own interface, and they carry a few lines of description at best.
Take a baking show with a 20-minute segment on fixing a flat sourdough starter. The advice is good. But if the only public text is "Ep. 87: Starter troubleshooting, plus a listener Q&A," there is nothing for an AI system to quote.
Video fares a little better. A YouTube upload has a title, a description and captions. Your own site is still the only place where you control the wording, the structure and the sources.
How podcast AI search picks sources to cite
Nobody outside these companies knows the exact ranking logic, and it changes. Public documentation and a lot of watching citations turn up the same ingredients again and again.
They need a page they can fetch
The page has to load without a login, without being locked behind a script that only runs in an app, and without being blocked in robots.txt. A page that returns the text in the initial HTML is the safest bet.
They want a direct answer, close to the question
Answer engines favor passages that respond to a question plainly. A heading that reads like the question, followed by a two-sentence answer, is easy to lift. A rambling intro followed by the answer in paragraph nine is not.
They need to know who is speaking
This is entity clarity. Is "Dana Ruiz" the host, a guest, or someone else? Is the show a weekly interview program about small-batch baking, or a general food podcast? If your site, your Apple listing, your LinkedIn and your guest bios all describe you differently, the system has a blurry picture. A consistent description makes you a safer source to cite.
They like structure they can read as data
Structured data (schema.org markup in the page) tells crawlers what a page is: a podcast episode, its title, its date, its show. Google's structured data introduction explains the basics. It isn't a magic switch, and Google is clear that markup alone doesn't guarantee any special treatment, but it removes guesswork.
They notice when other sources agree
Being mentioned, quoted or linked elsewhere (a guest's site, a newsletter, a directory, a Reddit thread that links your episode) makes a source look more established. You can't fake this, and you shouldn't try. You can make it easy by giving guests a clean episode URL they will want to share.
Which AI crawlers matter?
Several bots now visit sites on behalf of AI products, and they do different jobs. People mix them up constantly.
| Bot (user agent) | Run by | What it does |
|---|---|---|
| Googlebot | Crawls for Search, which also feeds AI Overviews | |
| Google-Extended | Controls whether your content is used for Gemini training and grounding; it is a robots.txt token, not a separate crawler | |
| GPTBot | OpenAI | Collects content that may be used to train OpenAI models |
| OAI-SearchBot | OpenAI | Finds pages to show in ChatGPT search results |
| ChatGPT-User | OpenAI | Fetches a page when a user asks ChatGPT to look at it |
| PerplexityBot | Perplexity | Indexes pages for Perplexity's answer engine |
| ClaudeBot | Anthropic | Collects public web content for Anthropic's models |
| CCBot | Common Crawl | Builds the open web archive many AI models are trained on |
Check each company's own documentation before you rely on this list, because names and behaviors get updated. OpenAI's bot documentation is a good example of a page worth bookmarking.
What matters is whether a bot trains models or fetches pages to answer a live question. Blocking a training crawler and blocking a retrieval crawler have very different effects on whether you appear in answers.
Should you block AI crawlers in robots.txt?
That depends on what the show is for, and the trade-off is real.
A host who publishes free episodes to grow an audience wants to be found, cited and linked. For that host, blocking the retrieval bots (OAI-SearchBot, PerplexityBot, and Googlebot itself) backfires. The system can't read the site, so it can't cite it.
A publisher who sells transcripts, or who objects to them being used as training data without payment, has a legitimate reason to block training-oriented bots such as GPTBot, Google-Extended or CCBot. The cost is that you may be less represented in what those models know. And robots.txt is a request, not a lock. Well-behaved bots respect it; others don't.
Two points trip people up. Google-Extended does not affect whether you appear in regular Google Search results, and Google says AI Overviews draw on the same Search index, so blocking Google-Extended is not how you opt out of AI Overviews. And a blanket rule written years ago, like a Disallow for every unknown bot, may be blocking things you now want.
A split policy that allows search and retrieval and blocks training looks like this:
User-agent: OAI-SearchBot
Allow: /
User-agent: PerplexityBot
Allow: /
User-agent: GPTBot
Disallow: /
User-agent: Google-Extended
Disallow: /
User-agent: CCBot
Disallow: /Open yoursite.com/robots.txt and read what is already there. If a platform manages the file for you (Squarespace, Wix and Webflow all do to some degree), check what that platform lets you change before assuming you have full control.
What should an episode page contain to get cited?
Give every episode its own page with real text on it. That is the same groundwork that helps regular Google rankings, and our guide to podcast episode pages for SEO covers the basics. For AI citation, a few details matter more.
A short, direct summary at the top
Open with two or three sentences saying what the episode answers, in the words you'd use across a table: "Dana explains why a sourdough starter goes flat after a feeding, and the three fixes that worked for her bakery." A summary that states a finding is more quotable than a teaser.
A transcript, cleaned up
The transcript is where the actual answers live. Publishing it in full gives an AI system the passages to quote, and it gives Google long-tail text to rank. Auto-generated transcripts are fine as a start, but fix names, jargon and obvious errors, because a misspelled guest name weakens the entity signal. Our page on podcast transcript pages goes through formats and trade-offs.
Headings shaped like questions
Break the transcript or show notes into sections with headings such as "Why does my starter smell like nail polish?" and "How often should I feed it?" These match the way people phrase prompts to an assistant, and each section can be pulled out on its own. Put the short answer directly under the heading, then the detail.
Named people, with context
On first mention, write "Dana Ruiz, owner of Ruiz Bread Co." rather than "Dana." Link guests to their own sites. List sources you cite. This reads naturally for humans and gives machines something to anchor on.
Structured data
Mark up each page as a podcast episode and the show as a podcast series. If you use WordPress, an SEO plugin may handle some of this. If you build by hand, the schema.org types are PodcastEpisode and PodcastSeries.
How do you make your show an obvious entity?
Entity clarity is dull work, and it pays off because it applies to every episode at once.
- Write one sentence that says what the show is, who hosts it, and who it is for. Example: "Crumb Theory is a weekly interview show about small-batch baking, hosted by Dana Ruiz."
- Put that sentence on your About page, in your RSS feed description, in your Apple Podcasts and Spotify descriptions, and in your social bios, with the same wording or close to it.
- Pick one spelling of the show name and each host name and keep it. Not "Crumb Theory," "The Crumb Theory Podcast" and "CrumbTheory" in different places.
- Make an About page that states the host's background, how long the show has run, and where to find it. Include links to profiles that confirm who you are.
- Add a hub page that lists every episode, so a crawler can find all of them from one place.
- Link each guest page or episode page back to the show's main page, and the main page to them.
Checking whether podcast AI search results are improving
Test by hand and by measurement. Each has blind spots.
By hand: ask ChatGPT, Perplexity and Google a handful of questions your episodes answer, phrased the way a listener would ask them. Note whether your site or a competitor shows up as a source. Repeat monthly. The answers vary between runs, so look for patterns rather than single results.
By measurement: look at your analytics for referrals from AI assistants, and at server logs or a bot report for which crawlers actually visited. If GPTBot and PerplexityBot never request your pages, you have a technical or blocking problem. If they visit and you never get cited, the problem is more likely content or authority.
The ShowHub Podcast SEO Audit is a free check. It looks at what Google can find on your site and flags gaps. For the ongoing view, ShowHub builds a page per episode from your feed and, on the Pro plan, includes analytics that separate traffic from search, social, podcast apps and AI assistants. The Business plan adds transcripts on episode pages, structured data and a report of which AI crawlers and bots have indexed the show. If you already run a WordPress site with good plugins and the discipline to maintain it, you can build most of this yourself, and that is a fair choice.
What you can't control
Anyone promising guaranteed AI citations is selling something. The limits:
- You can't force a citation. Each system chooses sources by its own logic, and that logic changes with model updates.
- Results vary by prompt and by day. The same question can cite different sources an hour apart.
- Some answers won't link out at all. An assistant may use what it learned in training and name no source, and you'll never see that traffic.
- The rules are still being written. Bot names, robots.txt tokens and opt-out options have changed repeatedly, and they will again. Recheck the documentation a couple of times a year.
The same pages also serve ordinary search and human visitors, so none of this depends on one assistant staying popular.
FAQ
Does having my podcast on Spotify or Apple Podcasts help me appear in AI answers?
Not much on its own. Those listings are thin, and they live inside someone else's platform. They do help listeners find you and they confirm your show exists, but the text AI systems quote usually comes from pages like your own episode pages, so publish those too.
Should I add my transcript if I already have show notes?
Yes, if you can. Show notes summarize, while a transcript contains the actual wording of your answers. Keep the short summary at the top and put the full transcript below it, with headings that break it into questions.
Will blocking GPTBot keep me out of ChatGPT answers?
Not necessarily. GPTBot relates to model training, while OAI-SearchBot and ChatGPT-User are used for search and for fetching pages on request. Blocking one does not automatically block the others, so set each rule on purpose and check OpenAI's current documentation.
How long before I see AI citations?
There's no reliable timeline. Pages first have to be crawled and indexed, which can take days or weeks, and after that citations depend on how well your content matches questions people actually ask. Treat it as a slow build and track it monthly.
Where to start this week
Pick your five most useful episodes. Give each a page with a plain-English summary, a cleaned-up transcript and question-style headings, then check your robots.txt. That is a couple of afternoons of work. If you'd rather not build the pages by hand, see what ShowHub's features generate from a feed.
End of No. 02 · Oct 28, 2026