AI Search
Ask a question, get back exactly where it's answered
AI Search reads a podcast transcript for you and reports back with verbatim quotes and timestamps, for whatever you're looking for: a brand, a person, a claim, a recommendation. It's not a keyword search, it understands the instruction. Below is a real result, not a mockup.
"books recommended or referenced by title, with the author if mentioned"
-
“Elizabeth Gilbert is the author of Eat Pray Love Big magic. And city of girls.”
Tim Ferriss introduces Elizabeth Gilbert by listing several of her authored books. 1:54
-
“Joy Carol Oates has written more than seventy books, including Blonde And we were the Mulvanies. and won the National Book Award for her novel Them.”
Tim Ferriss introduces Joyce Carol Oates by naming several of her well-known novels. 13:06
-
“Mary Carr wrote the best selling memoirs. The Liers Club? Cherry and Lit And her book on writing. The Art of Memoir.”
Tim Ferriss introduces Mary Karr by listing her prominent memoirs and her book on writing. 26:29
-
“Seth Godin has written twenty-one bestsellers published in forty languages, and Including Purple Cow? Lynchpin. And the practice.”
Tim Ferriss introduces Seth Godin by listing several of his bestselling business and creative books. 45:07
Captured 2026-08-31 from the live API, four of eleven results shown. Read the full transcript at /t/a7c18c255b33d645 to verify these are real quotes from a real episode.
This isn't RAG
The usual way "AI search" works is retrieval: chop the transcript into chunks, embed them, find the handful that look most similar to your query, and only show the model those. Anything phrased differently than your query, or split awkwardly across a chunk boundary, can get missed, and there's no real way to know that happened.
AI Search doesn't retrieve, it reads. The whole episode transcript goes to the model in one pass, and it's instructed to find every genuine match, not the closest-sounding pieces. That means a "not found" result actually means the model looked at the entire episode, not just the part a similarity search happened to surface.
This is what runs Bookmenti
Bookmenti is Podmenti's sibling product: a searchable index of every book mentioned across a curated set of podcasts. It's built by giving an AI model an instruction like the one above, "find every book mentioned in this episode, with a verbatim quote", and running it against every new episode as it's transcribed. That pipeline already processes thousands of episodes. AI Search is that same extraction engine, made available on demand, pointed at any episode and any instruction you give it, not just books.
What else it's for
every time our product, or a named competitor, comes up
whenever a specific person's name is mentioned
any statistic or claim about a specific topic
tools, books, restaurants, anything a host recommends
A common pattern: monitor every episode of a show for mentions of your company, your product or your name, and find out the moment it happens instead of searching for it yourself. Continuous monitoring with a webhook isn't built yet, one query at a time is, see the API docs.