4,490+
Validated winning VSLs & ad creatives mined for hooks
57+
Direct-response niches the hooks span
Weighted
Recurrence ranking: recency × member count × niche diversity
What a hook actually is in direct response
In direct-response copy, the hook is the opening — the first one to three sentences of an ad, or the first few seconds of a VSL — whose only job is to earn the next moment of attention. It is not the offer, not the mechanism, not the proof. It is the wedge that interrupts a scroll or a skip and makes continuing feel involuntary. Everything downstream is wasted if the hook fails, which is why operators test more hook variants than any other element of a campaign.
Because the hook is doing one narrow job, it tends to fall into a small number of recurring shapes. A hook either provokes a question the reader needs answered, names a fear they were already carrying, promises a discovery they haven't heard, or contradicts something they assumed was true. These shapes recur across niches and across years because they map to how attention works, not to any single product. That recurrence is exactly what a corpus can measure — and what a single creative session, however clever, cannot.
Two tools: clusters for patterns, extractions for examples
The AI Copy Agent surfaces hooks through two distinct retrieval tools, and the difference matters. get_top_clusters returns recurring hook patterns — groups of similar hooks that have been clustered together because many real winners express the same underlying angle. A cluster answers 'what kind of hook keeps showing up' and is the right tool for 'what are the most common hooks' or 'what hook angles work here.'
get_top_extractions returns individual hooks — the actual single lines pulled from real VSLs and ads, ranked by a per-item confidence score. An extraction answers 'show me real examples of this' rather than 'what is the abstract pattern.' You typically move from clusters to extractions: identify the recurring angle first, then drill into concrete instances of it to study sentence-level craft. Both tools accept a unit_kind, so the same machinery that surfaces hooks also surfaces pains, mechanisms, promises, villains, and other extracted units.
Why recurrence beats one-off cleverness
A clever hook you wrote once might convert, or it might be a fluke that never repeats. The agent's clusters are ranked by a weighted score that is deliberately built to suppress flukes: it blends recency, member_count (how many distinct products in the corpus use the pattern), and niche_diversity (how many different niches the pattern appears in). A pattern only rises to the top when it is being used by many winners, recently, across more than one vertical.
That cross-product, cross-niche recurrence is the signal worth copying. One product nailing a hook tells you little; thirty products independently converging on the same hook shape, across weight-loss and finance and survival, tells you the shape is structurally strong rather than situationally lucky. Ranking by recurrence is also what keeps the agent honest — it elevates what the market is actually validating, not what is merely novel or what a language model finds aesthetically pleasing.
The recurring hook archetypes
When you browse hook clusters, the patterns tend to organize into a handful of recurring archetypes. A curiosity hook opens a loop the reader needs closed — a question, an unfinished claim, a 'the reason why' that withholds the reason. A fear or loss hook names a threat the reader is already worried about and implies it is closer or worse than they thought. A mechanism or discovery hook leads with a novel cause or a 'newly discovered' explanation, reframing a familiar problem around something the reader hasn't seen.
A contrarian hook attacks a belief the audience holds as obvious — 'stop doing the thing everyone told you to do.' An identity hook speaks to who the reader is or wants to be, so the message feels addressed to them specifically. A social-proof hook leads with adoption or results others are getting, borrowing credibility before making a claim. These archetypes are descriptions of structure, not scripts — the agent surfaces which archetypes recur in your niche so you can write within a proven shape rather than guess at one.
Mine winning ad hooks from real winners — with the AI Copy Agent.
A Daily Intel Service membership unlocks the catalog; upgrade to Pro to unlock the AI Copy Agent, its recurring hook clusters, and individual extractions. Cancel anytime.
Narrowing with a semantic query
Pure popularity is the right default for browsing, but sometimes 'most common hooks' is too broad. Both get_top_clusters and the cross-niche pattern tool accept an optional semantic query. When you provide one — and the reranker is enabled in settings — the top candidates by weighted score are reranked by relevance to your phrasing. So instead of the globally most common hooks, you can ask for the recurring hooks closest to 'fear of being a burden as you age' or 'the one-ingredient cause angle' and get patterns matched to that specific concept.
This is the difference between a swipe file and an index. A folder of screenshots lets you scroll; a query-able corpus lets you ask for the recurring pattern nearest to the angle you are already chasing. The semantic query does not invent hooks — it reorders real, recurring patterns by how well they fit what you described, so the examples you study are both validated and on-topic.
Grounded, not generated: where the hooks come from
The agent does not write hooks from a language model's imagination and present them as proven. Every hook it shows you — whether a cluster pattern or an individual extraction — was lifted from a real VSL or ad in a corpus of 4,490+ validated winners across 57+ niches, then ranked by recurrence or confidence. When the agent paraphrases a pattern for you, it is describing what its tools surfaced, and it can cite the products and clusters it drew from.
That grounding is what separates studying winning ad hooks here from prompting a generic chatbot for 'ten Facebook hooks.' A general model produces plausible lines from its training distribution; it cannot tell you whether any of them ever ran, let alone won. The agent retrieves first and writes second, so the archetypes you adapt are the ones the market has already been validating — and if the corpus has nothing relevant to your angle, the agent tells you that instead of fabricating filler.
A practical hook-mining workflow
Start broad. Ask the agent for the top recurring hook clusters in your niche to see which archetypes dominate the vertical. Read the cluster patterns as structure — note which shapes recur and how many products and niches back each one. Then narrow: hand the agent a semantic query describing your specific angle and let it rerank the clusters by relevance so you are looking at the recurring hooks closest to your idea.
Once a pattern looks right, pull individual extractions for it to study real sentence-level execution — phrasing, rhythm, the exact word that lands the turn. From there the rest of the agent's toolkit takes over: scaffold the full piece with a framework, score a candidate headline on the 4U axes, or audit your finished draft's hook strength against the corpus. The hook tools are the front door; they put you inside a proven shape before you write a single line of your own.
The bottom line
Winning ad hooks are not lucky lines — they are recurring shapes that many validated winners keep converging on. The AI Copy Agent makes those shapes browsable: get_top_clusters ranks recurring hook patterns by recency, member count, and niche diversity, get_top_extractions shows the real individual examples behind them, and a semantic query narrows either to your exact angle. Everything is pulled from a corpus of 4,490+ winners across 57+ niches and cited — never invented. It is included on the Pro and Premium plans.
Frequently asked questions
What makes a hook a 'winning' hook here?
Not one clever line. The agent's hook clusters are ranked by a weighted score blending recency, how many corpus products use the pattern, and how many niches it spans. A hook ranks high only when many recent winners across verticals keep using its shape.What is the difference between clusters and extractions?
Clusters are recurring hook patterns — groups of similar hooks ranked by weighted score, answering 'what shape recurs.' Extractions are individual real hook lines ranked by confidence, answering 'show me concrete examples.' You usually go from cluster to extraction.Are the example hooks real or AI-generated?
Real. Every hook the agent surfaces was extracted from a validated winning VSL or ad in the corpus, then ranked by recurrence or confidence. The agent retrieves real patterns first and can cite the products and clusters it drew from.Can I find hooks for my specific angle?
Yes. The hook tools accept an optional semantic query. When provided and the reranker is enabled, the top patterns by weighted score are reranked by relevance to your phrasing — so you get recurring hooks closest to your concept, not just the globally most common ones.Does this only work for ad hooks, or VSL hooks too?
Both. The same retrieval surfaces hooks from VSLs and ads in the corpus, and the unit_kind input also covers pains, mechanisms, promises, and villains — so you can mine the full opening, not just the literal first line.How do I get access to the hook tools?
They are 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 hook clusters and extractions. Cancel anytime.