For the people who already made the demo, interview, webinar, podcast, or customer story. Framesite turns the exact moment into a source visible to AI.
"With middle-out it is a fraction of that now."
Richard Hendricks, exact clip attached
See why a video link leaves useful moments hidden, what the citation research found, and how Framesite makes those moments visible to AI.
Otterly's study found a sharp divide between a video URL and a structured video. Timestamps give an AI system smaller, specific units it can retrieve and cite.
A single useful video can answer more than one question.
YouTube visibility changes sharply from one AI platform to another.
Long-form video carried nearly all YouTube citations in the study.
Source: OtterlyAI YouTube Citation Study 2026. These are study findings, not universal rates. The detailed YouTube analysis covers videos already cited during its study window, so it is strongest for repeated citation behavior rather than initial eligibility.
Otterly observed timestamped YouTube citations only in Google AI Overviews and Google AI Mode. Other assistants may know the video exists without carrying the exact quote, speaker, timestamp, and source into the answer.
Otterly describes chapters, timestamps, and transcript segments as connected structure. Framesite extends that same model into a visible page with visible sourcing and schema.
The customer story, webinar, demo, or episode stays intact.
video_id
duration
publish_dateOne asset is divided into distinct questions and topics.
00:00 Setup
02:15 Storage
05:24 ResultEach chapter maps to the exact words and speaker.
Richard Hendricks
"a fraction of that"The moment gets a page, clip, source context, and structured data.
VideoObject
Clip
FAQPageAn answer engine has a specific passage to retrieve instead of one opaque file.
Source: Richard
Clip: 02:15This is the product mechanic, not a citation guarantee. Framesite packages the same source-fact chain Otterly's findings suggest is useful: a specific segment, connected to words, context, and a source an agent can inspect.
See the finished page →Within the already-cited video set, familiar popularity metrics had near-zero linear correlation with how often a video was cited again. Otterly's conclusion was reference selection, not recommendation ranking.
Correlation does not imply causation, and this dataset cannot tell us what made a video eligible for its first citation. It shows that popularity did not predict repeat citation frequency within the cited set.
The study points toward a different job for video in AI search: make a complete reference, divide it into precise moments, and give those moments a visible place to live.
Long-form represented 94% of YouTube citations. Tutorials, demos, interviews, comparisons, webinars, and case studies give an answer engine enough context to select a source.
Use properly formatted timestamps and clear chapter labels. They let Google address one section of a video the way it can address one section of a page.
FramesiteExact audio-timeline offsets stay attached to every chapter.See how it works →YouTube was meaningful in the study, but platform behavior varied sharply. A page built from your video gives the quote, speaker, timestamp, clip, and schema a standard web page beyond the video URL.
FramesitePublish a page, enrich the one you have, or send it to your CMS.See publishing paths →Views and subscribers did not predict repeat citations within the cited set. Publish the clearest answer you already have, then make its reference structure explicit.
Recency had a weak positive relationship with repeat citation frequency, around r = 0.3. That supports meaningful refreshes in fast-moving topics, not publishing volume for its own sake.
These recommendations are Framesite's application of Otterly's observed patterns. They are not promises that a platform will cite a particular video or page.
Different teams feel the same problem at different moments. The work is finished, the answer exists, and AI still cannot find it.
A testimonial, product demo, webinar, or case study can carry the exact answer to a commercial question. Framesite makes that moment visible without turning the video into another generic blog post.
Otterly found that 94% of cited YouTube items were long-form and that views had near-zero linear correlation with repeat citations inside the cited set. Reference value can matter more than channel size.
A video can become a customer story, answer page, quote source, and structured layer on the client's existing site. That creates a measurable after, not another one-time handoff.
Interviews and podcasts are full of specific, attributable answers. Framesite turns those answers into visible pages with the person, context, and moment still attached.
The video stays a video for people. Framesite adds a visible customer story, exact quote, timestamp, clip, and schema around it, so an agent can answer from the real moment.
Open the live example →
piedpiper.com / customer-story"We were spending a fortune to store everything. With middle-out it is a fraction of that now." 2:15
Otterly's 90-day experiment found that only 84 of more than 62,100 AI bot visits touched llms.txt. The average content page received about 265 visits. Framesite focuses on a standard visible page, semantic structure, and visible sources, then adds structured data around it.
llms.txtThis was one site over 90 days, so it is a useful signal, not a universal law. The practical lesson is simple: do not hide the answer in a special file and expect every agent to use it. Make the page itself worth reading.
We use these studies to frame the opportunity. We do not turn a correlation or a platform snapshot into a guarantee.
llms.txt as a ranking lever. It does not prove the file never matters. Read the source.Bring the finished video or archive. We’ll show you the rollout for your company or clients.