4,490+
Validated winning VSLs & ads in the corpus
57+
Direct-response niches covered
BM25 + vector
Hybrid search, reranked for relevance
Why the traditional swipe file is broken
Every copywriter keeps a swipe file — a folder of screenshots, saved PDFs, and bookmarked ad-library links. The problem is that it is dead the moment you save it. You can only find a swipe again if you remember the brand, the date, or the exact folder you filed it under. There is no way to ask it a question. When you sit down to write a fear-of-relapse hook for a blood-sugar offer, your swipe file cannot surface the ten ads that already nailed that angle — it just sits there as a pile of images.
A static folder also rots. Ad-library links break, screenshots lose context, and the file never tells you which examples actually worked versus which you saved on a whim. It captures what caught your eye, not what converted. Over a few years it becomes an archive you scroll past rather than a tool you reach for under deadline.
What an AI swipe file is instead
An AI swipe file flips the folder inside out. Instead of files you browse, it is a searchable corpus of 4,490+ validated winning VSLs and ads across 57+ direct-response niches — each one transcribed and broken into structured pieces. You no longer file and recall by hand; you ask it questions in plain language and it returns the most relevant passages, ranked.
Inside Daily Intel Service, this surface lives in the AI Copy Agent. Its search is grounded: every match it returns is a real passage from a real winner in the corpus, cited back to the product it came from. The agent paraphrases what the search surfaces — it does not invent example hooks. That grounding is the whole point. A swipe file you can search by meaning, that only ever shows you copy that actually ran, is a categorically different tool than a folder of pictures.
How semantic search beats keyword matching
The agent searches with hybrid retrieval: BM25 keyword scoring runs alongside vector (semantic) search, and the two result sets are fused with reciprocal-rank fusion (RRF) into a single ranking. BM25 catches the exact phrasing a winner used; the vector side catches the meaning even when the words differ. Search 'fear of being a burden on your family' and you also surface ads framed as 'don't leave them with the bill' — same emotional angle, zero shared keywords.
On top of the hybrid retrieval, a Voyage reranker re-scores the top candidates for true relevance to your query, so the best matches rise to the top rather than merely the most keyword-dense ones. The result is the search experience a folder can never give you: describe the angle, the pain, or the mechanism in your own words and get back the passages closest to that idea.
Two ways to search: extractions vs. raw chunks
The agent exposes two distinct search tools, and knowing which to reach for is most of the skill. search_extractions runs over curated extractions — the hooks, pains, tactics, mechanisms, and offers that have already been lifted out of each transcript and tagged by kind. It is the fast path: tight, deduplicated, ready-to-study units. Use it when you have a concept in mind and want clean examples of that concept, optionally filtered to a single unit kind or niche.
search_chunks runs over the RAW transcript chunks — full prose, not extracted units. You reach for it when extractions are not enough: when you need to see the actual sentences, the transitions, the pacing of a section in context. It is filterable by source_kind (vsl or ad) and by chunk_kind — opening, big_idea, problem, agitation, mechanism, social_proof, offer, urgency_scarcity, close, and the ad-specific sections. Raw chunks cost more tokens than extractions, so the agent searches extractions first and drops to chunks only when full prose is what the job needs.
Search a corpus of winning VSLs & ads — included on Pro.
A Daily Intel Service membership unlocks the catalog; upgrade to Pro to unlock the AI Copy Agent and its swipe-file search. Cancel anytime.
Studying a single product end to end
Search is for discovery across many winners; sometimes you want to dissect just one. When a search surfaces a product whose structure you want to learn from, list_products_by_niche gives you the products in that niche with their transcript IDs, word counts, and extraction counts — your map of what is in the corpus for that vertical.
From there, get_transcript_section pulls every chunk of that one transcript, optionally filtered to a single section. Want to see exactly how a top blood-sugar VSL handles its mechanism reveal, or how a memory-niche ad bridges from agitation to promise? Pull that section and read the real sequencing — the order, the transitions, the sentence-level craft. It is the difference between collecting a swipe and actually reverse-engineering why it worked.
A research session with the AI swipe file
A typical session moves from broad to narrow. You name your niche; the agent resolves it to the canonical slug. You run search_extractions for the angle you are working — say, 'mechanism hooks about hidden inflammation' — and read the cleanest matching units. If you want to feel the pacing rather than just the idea, you re-run the same query through search_chunks filtered to the mechanism sections of VSLs.
When one product keeps showing up in your results, you list its niche, grab its transcript, and use get_transcript_section to study its opening and close in full. Every step is grounded in the same 4,490+-asset corpus, and every result is cited — so the swipe you walk away with is traceable to a real winner, not a half-remembered screenshot.
Who the AI swipe file is for
It is built for direct-response operators who treat swipe as raw material, not decoration: copywriters drafting VSLs and advertorials, media buyers stress-testing angles before they spend, and offer owners studying how competing products structure their pitch. If you have ever wished your swipe folder could answer a question, this is that folder made queryable.
It is not a generic image gallery or a feed of trending creatives. The corpus is transcribed and structured specifically so it can be searched by meaning and read by section — tuned for the moment you are at a blank page and need to see how proven winners solved the exact problem in front of you.
The bottom line
An AI swipe file turns a dead folder of screenshots into a corpus of 4,490+ winning VSLs and ads you can interrogate. Search curated extractions for clean examples, drop to raw chunks when you need full prose and pacing, and pull a single transcript to reverse-engineer one product end to end — all hybrid-ranked, reranked, and cited to real winners. That is the swipe file finally working like a tool instead of an archive.
Frequently asked questions
What is an AI swipe file?
It is a searchable, grounded corpus of validated winning VSLs and ads — not a static folder. You query it in plain language and it returns the most relevant real passages, ranked and cited to the product they came from, instead of making you remember where you filed a screenshot.How is the search ranked?
Hybrid retrieval: BM25 keyword scoring runs alongside vector (semantic) search and the two are fused with reciprocal-rank fusion. A Voyage reranker then re-scores the top candidates for relevance, so the best matches surface rather than the most keyword-dense ones.What is the difference between searching extractions and raw chunks?
search_extractions returns curated, deduplicated units — hooks, pains, mechanisms, offers — and is the fast path when you have a concept in mind. search_chunks returns RAW transcript prose for when you need full sentences, transitions, and pacing in context.Can I study one product in depth?
Yes. list_products_by_niche shows the products in a niche with their transcript IDs and counts, and get_transcript_section pulls every chunk of a single transcript — optionally one section — so you can read its real sequencing and craft end to end.Can I filter searches by VSL vs. ad or by section?
Yes. Raw-chunk search filters by source_kind (vsl or ad) and by chunk_kind — opening, big_idea, problem, agitation, mechanism, social_proof, offer, urgency/scarcity, close, and ad-specific sections — so you can search just the part of the funnel you are working on.Which plan includes the AI swipe file?
It is part of the AI Copy Agent, included on the Pro and Premium plans. A Daily Intel Service membership unlocks the catalog; upgrading to Pro unlocks the agent and its corpus search tools.