LLMs Are Lazy. Jessica Hennessey on Microdosing Video.
- Dane Frederiksen
- May 2
- 5 min read
Jessica Hennessey is the founder of Resonate Online and BetterSites, a platform that helps companies show up in LLM-driven search. She sat down to talk about why LLMs are lazy, why CEOs need to script their video answers, and how to break long interviews into citation-ready clips.
Key Takeaways
Jessica Hennessey founded BetterSites, a platform that helps companies appear in LLM answers.
LLMs don't summarize. They pull pre-packaged Q&A pairs from transcripts, so feed them ready answers, not loose conversation.
Cut long-form video into 2-3 minute clips. Each clip gets its own title, description, and transcript.
The title is the question. The description is the answer. The transcript is the backup.
Real human video will beat AI avatars as the synthetic content flood gets worse.
Why are LLMs lazy when they read your content?
Jessica's first counterintuitive claim: LLMs don't actually summarize content. They pull. "LLMs are lazy. The more clear, pointed, nicely tied in a box kind of content that you give them, the more likely they will be to repeat that fact," she said.
That changes everything about how you make video. Most thought leadership advice says: be authentic, talk like a human, let the magic happen on camera. For human viewers, sure. For LLMs, no. The LLM is scanning your transcript looking for a clean question, a clean answer, a quotable fragment, a chart it can extract. It is not synthesizing your nuanced 25-minute conversation into a paragraph it can cite later.
So the question isn't "did I sound smart on camera?" The question is: can a lazy reader pull a clean answer out of my transcript without doing any work?
If your video transcript reads like a meandering conversation, the LLM gives up and moves on to the next source. If it reads like Q&A with answers tied in a bow, it gets cited.
Should CEOs script their video answers?
Yes. This was Jessica's most counterintuitive take. "The CEO needs to know what the answer is ahead of time. Not answering it off the cuff, but having written answers that have been memorized to a degree, or reading off of a prompter," she said.
That is the opposite of what most video coaches teach. The standard advice is to be loose, conversational, authentic. Jessica's argument: that flow doesn't translate well to a transcript an LLM is going to chunk. A meandering answer becomes a meandering transcript becomes nothing for the LLM to grab.
Her version of the workflow:
Pre-write the answer to each question
Use a teleprompter or bullet points on screen
Read or paraphrase tightly so the transcript stays clean
Trust that the human viewer will still feel the conversation, the discipline is on the structure, not the warmth
This isn't about killing personality. It's about making sure the spiky ideas, the specific terms, the numbers, and the original framings actually survive the trip through the transcript and into the LLM.
How do you microdose video content for AI search?
Cut your long-form interview into 2-3 minute focused clips, then publish each clip with its own page. Jessica described the approach as making "micro videos" instead of long-form. (The "microdosing" framing in the title is host shorthand for the same idea: many small, focused tries instead of one long sprawling video.)
Take a long interview, break it into focused segments of 2-3 minutes each, then publish each one with three things wired together:
The title is the question (the actual question a real person types into Google or ChatGPT)
The description is the answer (a clean, written-out version of what the guest said)
The transcript is the backup (the verbatim words the LLM can chunk)
Each clip becomes its own attempt at a citation. Ten clips means ten chances to get pulled into an AI overview. One long-form video means one chance, and the LLM has to do too much work to find the answer inside it.
"You get more tries to get into the LLM without it having to sift through 20 minutes of conversation." — Jessica Hennessey
The long-form video doesn't go away. It still exists as the human validation layer, the third-party proof that real people had this conversation. But it's not the main GEO play. The microdosed clips are.
Does recency matter more than evergreen for AI search?
Yes. This is where AI search splits hard from old SEO. "LLMs like recency," Jessica said. "It isn't about the way we used to do for SEO where you create a video, you post it, and it gives you two years worth of SEO and every month you're kind of adding onto it."
The implication: you can't post once and coast. The LLMs reward steady, recent output more than they reward old high-authority content. A six-month-old article is already losing ground to a fresh one that answers the same question.
That changes the math on content investment. Instead of producing one prestige asset per quarter, you need a series. Weekly. Biweekly. Whatever cadence you can hold without dropping quality. Jessica framed it as a call to arms:
"The longer you wait to do video or the longer you wait to create quality content, the more behind you're going to be." — Jessica Hennessey
The cost of inaction isn't a flat penalty. It compounds. Every month of no new content is a month other companies are filling the citation gap with theirs.
Will AI video replace real human video?
No. Jessica's bet is the opposite. As more companies flood feeds with AI-generated video and avatar-driven content, real humans on camera will stand out more, not less.
"Real video, live video, not AI-driven video, I think is going to be a standout as more people are creating all of these AI videos and using avatars to say things. I think it's not going to get the job done," she said.
Her reasoning ties back to the recency point. LLMs are pulling content into AI overviews now, and AI overviews are surfacing answers about real people, real conversations, real specifics. Avatar-generated content can fake the words. It can't fake a real transcript from two named people in a Riverside studio on a specific date.
That makes high-quality, real-person video a harder asset to fake and an easier asset for an LLM to validate as authentic. The companies that keep producing real video, with real people and proper transcripts, will have a citation moat that the AI-avatar shops can't cross.
Frequently Asked Questions
What does it mean that LLMs are lazy?
Jessica Hennessey, founder of BetterSites and Resonate Online, says LLMs don't actually summarize content. They scan transcripts for clean Q&A pairs, quotable fragments, or charts they can pull. If your content isn't pre-packaged for them, they skip it and move on to the next source.
How should CEOs prepare for video interviews about LLMs?
Script the answers ahead of time. Jessica recommends using bullet points or a teleprompter so each answer is clear, written, and quotable. Off-the-cuff conversations feel natural, but they don't translate well into the transcripts that LLMs chunk for citations.
What does it mean to microdose video content?
Cutting a long-form video into 2-3 minute clips, each focused on a single question. Each clip gets its own title (the question), description (the answer), and transcript (the backup). The strategy gives you many chances to get cited by an LLM instead of one.
Does recency matter for AI search the way it did for SEO?
Yes, but more so. Jessica argues LLMs reward fresh content far more than old SEO did. A weekly or biweekly cadence keeps you in the citation pool, while a "post once and coast" approach loses ground every month.
Will AI avatar video replace real human video on camera?
No. Jessica's bet is the opposite. As AI-generated and avatar-driven video floods the zone, real human video shot on a platform like Riverside, with real people and real specifics, will become harder to fake and easier for an LLM to cite as authentic.
Comments