Why video summaries bury the 10 percent that matters
Typical video summaries are 80–90 percent filler. If you search those summaries, the model is reading noise and forgetting the hook, friction, SKU, and claim that actually change a brief.
Video summaries bury the insight because 80 to 90 percent of them is filler: a girl is standing in a bathroom talking to the camera; she holds a product; she recommends you try it. That sentence can describe a thousand clips and distinguish none of them. If you then run AI search over those summaries, the model is reading noise. The useful span — hook, friction, 9-hour wear, shade that oxidises — is short and easy to forget. We stopped summarising. We extract atomic truths (format, hook, friction, utility, SKU, claim) and search those, not the bathroom narration.
Summary bloat is a retrieval bug
Long context does not fix this. Models retrieve the wrong clip because the filler is what the embeddings actually resemble. A query for “wear test with purchase intent” should not return the 15 million view morning routine whose summary happens to mention a product in passing.
What we keep versus what we throw away
Filler in a typical summary versus atomic truths| In a typical summary | Keep as an atomic truth? |
|---|
| A person is standing in a bathroom talking to camera | No |
| Holds a product / recommends you try it | No — too generic |
| Format (haul, GRWM, wear-test) | Yes |
| Hook / opening promise | Yes |
| Friction (oxidises, dry-out, splotchy) | Yes |
| SKU and claim | Yes |
Atomic truths, not synopses
The 90 percent fluff is discarded on purpose. What remains is the 10 percent that can change a brief, a briefing, or a media dollar. Search then runs on those atoms.
How we know this
These notes come from building Noodle4’s review and listening pipelines, not from a published academic sample. Cost curves, intent splits, and ASR collisions are from internal R&D and client-shaped tests. They will not match every category. We include them because the failure mode is structural — not because one campaign is universal.
Alex Gray is the founder of Noodle4 and a former technical AI product owner. He works with agencies to apply multimodal AI to creator screening, draft approval, and post-publication monitoring.
Frequently asked questions
Why is AI search on video summaries so inaccurate?
Typical summaries are 80–90 percent filler. Search and RAG then read noise, so the model forgets the short span that contained the actual behaviour or claim.
What does Noodle4 extract instead of a video summary?
Atomic truths: format, hook, friction, utility, SKU, and claim. The fluff is thrown away so search and reporting run on the 10 percent that drives revenue.