AI VSL Script Generator: Framework + Corpus-Grounded Outlines

The blank page is the expensive part of a VSL

Writing a video sales letter rarely fails on word choice. It fails on structure: a hook that takes too long to land, a mechanism buried after the offer, a close with no urgency, or seven minutes spent on a problem that should have taken ninety seconds. Those are pacing and architecture problems, and a blank document gives you no help with either.

A generic AI prompt — "write me a VSL for a weight-loss supplement" — makes the architecture worse, not better. It returns plausible-sounding prose with no sense of how long each section should run, no proof that the angle has worked, and no way to trace any line back to a real winner. You end up rewriting structure you should have started with. The VSL script generator attacks that exact gap: it decides the skeleton first, then fills it.

What the generator actually produces

Under the hood this page describes a single tool, generate_vsl_outline. You give it a framework, optionally a niche, a copy kind (vsl, ad, email, or landing), and a target length. It returns a structured outline — not a finished script — where every section carries a purpose line, a target duration in seconds, and a target word count derived from a ~150-words-per-minute voiceover pace.

Crucially, each section also ships with 1-3 corpus exemplars: real extracted hooks, pains, mechanisms, proof lines and offers pulled from the corpus of 4,490+ validated winning VSLs and ads. Each exemplar is cited by the size of the cluster it belongs to (how many corpus products use that pattern) and the product names it was lifted from, so you can see prevalence before you borrow a structure. The agent then writes your prose into that scaffold — it does not invent the skeleton, and it never copies the exemplars verbatim.

Why framework + corpus beats a generic prompt

Two things make a VSL script generator useful rather than decorative. The first is a real framework, which supplies a tested order of operations and a defensible time budget. The second is corpus grounding, which replaces the model's guess about "what a good hook looks like" with hooks that were actually extracted from winning videos and ranked by how widely the pattern recurs.

Generic AI gives you neither. It writes from its training distribution, so it produces the average of all sales copy on the open web — fluent, forgettable, and occasionally hallucinated. The generator inverts that: the framework fixes the architecture, and the corpus exemplars per section show you the specific moves that recur across many proven products in your niche. You spend your effort on voice and offer, not on rediscovering structure that already exists.

The seven-beat VSL spine and how time is budgeted

Most long-form VSL frameworks resolve to the same spine: hook, lead, problem, mechanism, proof, offer, and urgency/close. The hook earns the next thirty seconds; the lead frames the big idea; the problem agitates the pain the buyer already feels; the mechanism explains why this solution works when others failed; proof makes it believable; the offer makes it concrete; and urgency drives the decision now instead of later.

The generator turns that spine into minutes and words. Each framework section declares a proportion of the whole, so when you ask for an eight-minute VSL the tool splits the budget across the beats — a hook might get its share of seconds and a word target, the mechanism gets a larger block, and the close gets a tight one. Defaults are tuned per copy kind: a VSL is eight minutes, an ad one minute, an email about a minute and a half, and a landing page four minutes. You can override the total length, and the per-section budget recalculates against it.

How to generate a VSL script with the agent

The workflow is deliberately short so you spend time writing, not prompting. Each step grounds the next so the final draft inherits both a tested structure and proven angles from the corpus.

1. 1Name your nicheTell the agent your niche so it resolves to a canonical slug and grounds every section's exemplars in that vertical instead of pulling cross-corpus.
2. 2Pick a frameworkChoose a framework such as the 12-step VSL, PAS, PASTOR or RMBC — or ask the agent to recommend one based on your audience's awareness and sophistication.
3. 3Generate the outlineRun generate_vsl_outline to get a time-budgeted scaffold: a target duration and word count per section, each grounded with 1-3 cited corpus exemplars.
4. 4Write into the scaffoldWrite your prose section by section, using the exemplars for inspiration and honoring the per-section word budget so pacing stays on target.
5. 5Audit and scoreHave the agent score the finished draft against the corpus — structural completeness, hook strength, and missing sections — then tighten the weak beats.

Beyond VSLs: ads, emails, and landing pages

The same tool scaffolds more than long-form video. Because it takes a copy_kind argument, you can generate a one-minute ad outline, a short email scaffold, or a four-minute landing-page structure using the same framework-plus-corpus mechanism. Each kind gets its own sensible default length and its own set of expected sections, validated against what the chosen framework supports.

That makes the generator a single front door for an entire funnel. Outline the VSL, then outline the pre-sell ad and the follow-up email against the same niche corpus, and the angles stay consistent across the funnel because every scaffold is grounded in the same set of validated patterns across 57+ niches — not in three separate one-off prompts that drift apart.

Close the loop: score the draft against the corpus

An outline starts the script; an audit finishes it. Once you have written prose into the scaffold, the agent can score the draft against the corpus — checking which expected sections are present or missing for that copy kind, ranking your hook against real corpus hooks, and citing specific exemplars to fix the weak beats. That turns "generate" into a loop: scaffold, write, audit, tighten.

Because the same corpus powers both the outline and the audit, the standards are consistent end to end. The structure you were handed is the structure you are graded against, so a draft that survives the audit is one that actually fills the spine the framework promised — not one that merely reads well in isolation.