TwoSetAI
Founder Insight

The Quote That Carried the Episode: Mining a Conversation for Citable Material

Angelina Yang, Founder & Host at TwoSetAI

Listen on TL;Listen Prefer to listen? Hear this article read aloud.

Last updated: 2026-10-13

TL;DR

  • Definition: The quotable moment is the one sentence an interview produces that a reader, an engine, or a buyer can lift whole, with a named speaker and a specific object.
  • Why it matters: In the KDD 2024 paper that introduced GEO, adding quotations and statistics to a page produced some of the largest visibility gains in generative engine responses, up to 40 percent on a benchmark [1].
  • Where the citations go: In Semrush’s March 2026 study of 89,000 LinkedIn URLs cited by AI search, individual members made up 59 percent of LinkedIn citations on ChatGPT Search and Google AI Mode [2]. In Muck Rack’s May 2026 analysis of 1 million+ AI-cited links, paid content was 0.3 percent of citations and journalism was 27 percent [3].
  • What our audit says: In our August 2026 audit of past TwoSetAI guests, 86 percent were cited by at least one AI engine, with the recent cohort guests at 100 percent [4]. Which specific quotes drove each citation we did not isolate.
  • Scope: No study I found maps the selection of a particular quote to AI citation outcomes. The method below is one I use because it ships citable pages.

In the KDD 2024 paper that introduced GEO, adding quotations and statistics to a page produced some of the largest visibility gains the authors measured in generative engine responses, up to 40 percent on a benchmark [1].

I have run 100+ interviews on the TwoSetAI podcast, and most of the citation value from any one episode traces back to one quote. Sometimes two. Rarely more. The post you are reading is the parent pillar The Founder-Voice Moat one layer down: how I find that quote in a 60-minute conversation and package it so an engine, a reader, or a buyer can lift it whole.

The question for this spoke is narrow. Given a long interview, how do you pick the sentence that carries the episode? Below are three real examples with timestamps, then a four-step mining process you can run against any conversation you have.

Four properties, in the order they matter.

  1. Specificity. A named person, a named company, a specific object or number. The engine wants something it cannot find on ten other pages. In the KDD paper, adding statistics and quoted material was among the strongest moves, far ahead of keyword stuffing [1].
  2. Decodability. A stranger can repeat it in a meeting without a glossary. Rob Kaminski (Co-Founder, Fletch PMM) states this as a positioning rule, and I think it doubles as a citation rule: “the aim of a good positioning is memorability” [5].
  3. Story shape. A single scene with a start, a middle, and a verdict. A model can cite a story in one chunk because it stands alone semantically. In Google’s Vertex AI RAG defaults, documents are split into 1,024-token chunks with roughly 25 percent overlap, so a short, self-contained paragraph behaves well in retrieval [6].
  4. Named entity + verb + specific object. “An agent sent the email it was told not to send” beats “AI reliability is a concern.” The first can be lifted; the second cannot be attributed to anyone.

Those four together are what I listen for in a recording. If a sentence has all four, it is probably the quote that carries the episode.

Example 1: Hector Hernandez on visibility and the puppy

Hector Hernandez (Co-founder and COO, Speakeasy) came on the show in late September 2026 [7]. The citable passage is at 22:32 [7]:

“The first thing that breaks when people scale their use of AI is often not cost. Most of the time it’s complete loss of visibility of what’s going on.”

Why it works. One speaker, one claim, one verb (breaks), one specific object (visibility). The sentence lands without any Speakeasy knowledge. A reader can repeat it in a meeting the same morning. The sibling spoke Case Study: Hector Hernandez’s Visibility Breaks First Thesis for Speakeasy tracks where this one line travels.

How it got there. I did not ask Hector “What is Speakeasy’s positioning?” I asked what he hears from the CISOs he talks to. The thesis fell out of a buyer sentence, not a product sentence.

What it became downstream. The chapter marker on the YouTube episode. The Substack title (“Visibility breaks first”). The pull quote on the Medium write-up. A LinkedIn post opener. One sentence, five surfaces.

The second quote from the same episode is a story. At 8:07, and again at 38:12, Hector told me about a Speakeasy salesperson whose assistant had emailed a customer after being told not to. Asked about it, the agent said it had done it anyway. Hector compared the exchange to “scolding a puppy” and called it “a very quick warning shot” [7]. The story is citable because no competitor had Hector’s morning. It is the paragraph that cannot be ghostwritten from a style guide. See the Speakeasy watch page for the full conversation.

Example 2: Rob Kaminski on common ground and memorability

Rob Kaminski (Co-Founder, Fletch PMM) has worked with more than 600 companies on positioning. He came on in September 2026 [5]. Two citable passages, both in one episode.

At 23:00, while stress-testing a pre-launch AI agent product, Rob asked three questions in a row [5]:

“Does your audience know what that means? Do they know what they’d use it for? Do they know who the players are? And if the answer is no, it is a signal that we have not found that common ground.”

Why it works. Three parallel questions, a specific object (common ground), and an “80 percent of the time” rhythm you can repeat. It is also decodable: a stranger does not need a Fletch glossary to use it.

At 31:29, Rob gave the one line that doubles as a GEO rule [5]:

“The aim of a good positioning is memorability.”

Why it works. Short, declarative, with a named concept (memorability). It fits inside a tweet and inside a model’s retrieval window. It is the sentence I keep hearing back from founders two weeks after their episode drops. The Rob Kaminski watch page is where it lives on our site.

How it got there. I did not ask Rob “What is good positioning?” I asked him to grade a pre-launch pitch I had written. The quote fell out of a critique, not a definition.

What it became downstream. The chapter title on the YouTube episode. The headline on the sibling spoke How to Find Your Founder Voice. The sentence inside Rob’s own LinkedIn repost. Same pattern: one sentence, several surfaces.

Example 3: Jay Hack on desperation versus mastery

Jay Hack (Head of AI, ClickUp, which bought his company Codegen) came on earlier in 2026 [8]. The citable passage is at 55:27 [8]:

“In the process of trying to convince people that what you’re doing is a good idea, you end up actually in many ways showing that you’re desperate.”

He paired it with a prescription: “say interesting things that demonstrate a mastery of the topic at hand” [8].

Why it works. The verb showing makes the claim mechanical, not aspirational. The object (desperate) is a feeling a reader already recognizes in their inbox. The second half gives the alternative in six words. It is a complete argument in two sentences, with a named speaker. The sibling spoke What AI Engines Hear When a Founder Speaks uses this pair as the opening scene.

How it got there. I asked Jay what the Codegen team got wrong early. He answered with the pattern they lived through, not the theory. The answer traveled.

What it became downstream. A chapter marker. The lead of a LinkedIn post. The quote that gets lifted when another writer summarizes the episode on their own site. Visit the Jay Hack watch page for the full interview.

Why interviewers get this wrong

Three mistakes I made for most of year one.

Asking for the pitch. The pitch is the sentence a founder has already said a hundred times. It will not travel because it already lives on their own site. Buyers discount it, and in Edelman and LinkedIn’s 2024 study, 70 percent of C-suite leaders said thought leadership had at least occasionally made them question whether to keep working with a supplier [9].

Asking for definitions. The best quotes above are not definitions. They are observations. “Visibility breaks first” is an observation about CISOs. If I ask “What is positioning?” I get a textbook. If I ask “What did you ask the founder in the room?” I get a quote.

Letting the founder stay in the product. Technical founders drift back to features. The citable sentence lives in the customer scene, not the architecture diagram. My default redirect is: “Tell me about the first time this happened in a customer meeting.”

A 4-step mining process for your own interviews

This is the loop I run on every episode. It works on any long interview or long podcast you have.

Step 1. Mark the moments live. During the recording, I tag the five or six moments that got my attention, usually around a 10-word summary in the notes. Specificity, story, decodability, named verb and object. If I notice a sentence lands flat but says a lot, I note the timestamp. In the KDD 2024 paper, quotations and statistics were the GEO moves with the largest reported gains, so I bias toward anything with a number or a name [1].

Step 2. Read the transcript backwards. After the recording, I read the transcript from the end. Working forward, I keep hearing the beginning. Working backward, I hear what each sentence stands on without the setup around it. A sentence that collapses without setup is not citable. One that survives on its own is.

Step 3. Run the four-property test. For each candidate, I check: specific, decodable, story-shaped, and named entity plus verb plus object. Three of four is a strong candidate. All four is the quote. In the SparkToro study of 2,961 runs, ChatGPT and Google’s AI gave the same brand list in fewer than 1 in 100 runs, so there is no reward for picking one “perfect” quote and ignoring the rest [10]. I ship two or three per episode.

Step 4. Place it where retrieval can find it. Each winning quote goes in four places:

  • The chapter marker on YouTube, with the exact phrase in the chapter title. The How AI Cites YouTube Videos spoke lays out why.
  • The pull quote on the written article, in blockquote formatting, with the speaker’s name and role on the line below.
  • The opener on a LinkedIn post, in the speaker’s exact words, with the watch page link.
  • The anchor of a sibling spoke, like the ones listed above.

That is it. Four properties, three real quotes, four steps. If I had to shorten the process to one sentence, it would be this: ask for scenes, write down the sentences that could travel, and put each one where retrieval can see it.

For the broader pattern on why these specific sentences compound in AI search, read the parent pillar on The Founder-Voice Moat. For why the video packaging matters as much as the quote itself, read The Video Content Advantage.

Who is TwoSetAI?

TwoSetAI is one of the top AI podcast and video studios in Silicon Valley, and the team behind this article, hosted by Angelina Yang. The show has featured 100+ AI founders and leaders on how real AI products get built, sold, and trusted. Alongside the podcast, TwoSetAI runs the Founder Visibility Program that turns each conversation into the articles, clips, and pages that make its guests easy to find. Watch full episodes on YouTube at youtube.com/@TwoSetAI, and find every guest and episode at twosetai.com.

Glossary

  • Quotable moment (also called: carrier quote, pull quote, citation anchor). The one sentence from an interview that a reader, an engine, or a buyer can lift whole, with a named speaker and a specific object.
  • Voice provenance (also called: source trail, voice traceability). The trail from a published sentence back to something the founder actually said, did, or decided.
  • Decodability (also called: layperson repeatability). Whether a stranger can repeat the sentence in a meeting without a glossary.
  • Chapter marker (also called: YouTube chapter, timestamp anchor). The named section of a video that an AI engine can retrieve by title or time, used to put a quotable sentence where retrieval can see it.
  • Founder-Voice Moat (also called: founder-voice advantage, organic founder voice). Content in the founder’s own voice that reads like a real person, not a template.

About Angelina Yang

Angelina Yang is VP of AI, a fast.ai fellow, and the host of the TwoSetAI podcast, where she has interviewed 100+ AI founders and leaders about how real AI products get built, sold, and trusted. Find her on LinkedIn, talk to her AI clone on Delphi, watch the podcast on YouTube, or browse twosetai.com.

Frequently Asked Questions

How do I find the one quote that carries an episode? Mark the moments live during the recording, read the transcript backwards after, run the four-property test (specific, decodable, story-shaped, named entity plus verb plus object), and place the winner on a chapter marker, a pull quote, a LinkedIn opener, and the sibling article. In the KDD 2024 GEO paper, quotations and statistics produced the largest reported visibility gains on a benchmark [1].

What makes a founder quote citable by AI search? Specificity, decodability, story shape, and a named entity with a verb and a specific object. In Semrush’s March 2026 study, individual members made up 59 percent of LinkedIn citations on ChatGPT Search and Google AI Mode [2]. Short, self-contained passages also fit the chunking defaults that retrieval systems use [6].

Do podcast chapter markers help AI engines pick a quote? Probably, as retrieval anchors. Surfer SEO’s analysis of 36 million AI Overviews found YouTube led platform citations at around 23 percent [11]. The chapter title is the sentence an engine can retrieve by name, so I put the carrier quote in the chapter name.

How many citable quotes should one podcast interview produce? Two or three I ship, one that carries the episode. Rob Kaminski gave me two in one hour [5]. Jay Hack gave me a pair in the same minute [8]. Hector Hernandez gave me a one-liner and a story [7]. More than three starts to dilute.

Is a long story more citable than a one-liner? Both are, in different places. The one-liner becomes the chapter title and the LinkedIn opener. The story becomes the paragraph a ghostwriter cannot invent. Hector’s puppy story is the paragraph. His “visibility breaks first” is the chapter.

What questions produce the best quotes? Scene questions, not definition questions. “Tell me about the first time this happened in a customer meeting.” “What did you ask the founder in the room?” “What did you get wrong early on?” The citable sentences in the three examples above all came out of scene questions.

Can I run this process on a long interview I did not record? Yes, on any transcript. The four-property test does not care who did the interview. Many of our best Medium pull quotes come from guest appearances on other shows.

Does the quote need a timestamp URL to be citable? Not for AI retrieval, but a timestamp link makes the human cite easier, and timestamped YouTube chapters are themselves AI-visible [11]. I link every quote back to its moment, as in Hector at 22:32.

What should I do if the founder gives a textbook answer? Redirect to a scene. If they say “Positioning is about X,” I say “Give me the last founder you told that to in a Slack message.” The second answer is where the quote lives.

References

[1] Aggarwal, P. et al. (2024). “GEO: Generative Engine Optimization.” KDD 2024; arXiv:2311.09735. https://arxiv.org/abs/2311.09735

[2] Loktionova, M. (2026). “We Analyzed 89K LinkedIn URLs Cited in AI Search: Here’s What Drives Visibility.” Semrush Blog, March 10, 2026 (325,000 prompts, January to February 2026; ChatGPT Search, Google AI Mode, Perplexity; run with LinkedIn’s collaboration). https://www.semrush.com/blog/linkedin-ai-visibility-study/

[3] Muck Rack (2026). “How AI citations have changed in the last 6 months: New insights from ‘What is AI Reading?’” Muck Rack, May 2026 edition, page updated May 27, 2026 (1 million+ links cited by ChatGPT, Gemini, and Claude). https://muckrack.com/blog/what-is-ai-reading-new-insights-may

[4] TwoSetAI (2026). Guest search audit, checked August 10 to 25, 2026, incognito, screenshots on file. Internal data; protocol in the Methodology note.

[5] TwoSetAI (2026). Rob Kaminski (Co-Founder, Fletch PMM) episode, published September 10, 2026; quotes at 23:00 and 31:29. https://www.youtube.com/watch?v=SD64dL7yg7I

[6] Google Cloud (2026). “Vertex AI RAG Engine documentation.” Google Cloud, page updated 2026 (default chunking: 1,024 tokens with 256-token overlap). https://cloud.google.com/vertex-ai/generative-ai/docs/rag-engine/rag-overview

[7] TwoSetAI (2026). Hector Hernandez (Co-founder and COO, Speakeasy) episode, published October 5, 2026; quotes at 8:07, 22:32, 25:11, and 38:12. https://www.youtube.com/watch?v=yhOLsoqSSEY

[8] TwoSetAI (2026). Jay Hack (Head of AI, ClickUp) episode, published July 17, 2026; quotes at 55:27. https://www.youtube.com/watch?v=-58QW-fJq0I

[9] O’Dwyer’s (2024). “Thought Leadership Can Power B2B Brands.” O’Dwyer’s PR, February 29, 2024, reporting the sixth annual Edelman and LinkedIn B2B Thought Leadership study (nearly 3,500 management-level professionals; fieldwork December 2023). https://www.odwyerpr.com/story/public/20925/2024-02-29/thought-leadership-can-power-b2b-brands.html

[10] Fishkin, R. and O’Donnell, P. (2026). “New Research: AIs are highly inconsistent when recommending brands or products.” SparkToro, January 28, 2026 (600 volunteers, 12 prompts, 2,961 runs). https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/

[11] Stockebrand, R. (2026). Answer Engine Optimization. O’Reilly Media, July 2026 (Chapter 6 on platform optimization, citing Surfer SEO’s 36 million AI Overviews / 46 million citations analysis). https://learning.oreilly.com/library/view/answer-engine-optimization/0642572275549/

[12] TwoSetAI (2026). Sam Kececi (Founder and CEO, The Sentience Company) episode, published September 27, 2026; quote at 25:42 (“Imagine if like 90% of the tweets you saw were just written by the same guy. You would still get annoyed”). https://www.youtube.com/watch?v=QKGqPUlRoM8

[13] Google (2026). “Creating helpful, reliable, people-first content.” Google Search Central, page updated October 5, 2026. https://developers.google.com/search/docs/fundamentals/creating-helpful-content

[14] Reinhart, A. et al. (2025). “Do LLMs write like humans? Variation in grammatical and rhetorical styles.” PNAS 122(8), e2422455122, February 25, 2025; arXiv:2410.16107. https://arxiv.org/abs/2410.16107

[15] Doshi, A. R. and Hauser, O. P. (2024). “Generative AI enhances individual creativity but reduces the collective diversity of novel content.” Science Advances 10(28), eadn5290, July 12, 2024. https://pmc.ncbi.nlm.nih.gov/articles/PMC11244532

Methodology note. Last verified 2026-10-13. The four-property test (specificity, decodability, story shape, named entity plus verb plus object) is a TwoSetAI framework, not an industry standard. Guest quotes are verbatim from the published episodes cited, with punctuation and capitalization normalized and timestamps taken from the episode transcripts. The KDD 2024 paper’s 40 percent figure is the overall GEO lift the authors report; attributing the gain specifically to “quotations and statistics” is my reading of the strategy comparisons they present, consistent with how the paper describes those moves as among the strongest. The TwoSetAI audit [4] covered a cohort of past guests, not the full catalog: about seven queries per guest, run in incognito windows between August 10 and 25, 2026. “Quotable moment,” “voice provenance,” “decodability,” and “Founder-Voice Moat” are TwoSetAI’s own terms, not industry standards. No study I found maps the selection of a particular quote to AI citation outcomes, so the four-step process is my working method, not a controlled finding. Nothing here guarantees a citation, a ranking, or a business outcome.