BriefBot — AI Proposal Writer

What it does & who it's for

The proposal is where deals die: three days late, priced by vibe, written in a voice that isn't yours. BriefBot turns the discovery call you already had into the proposal it should produce — through a pipeline built on one principle: the model drafts, you control the facts. Stage 1 reads the raw transcript — crosstalk, [inaudible] tags, the budget mumbled as "twenty-five to thirty, all in" — and produces a one-page fact sheet: goals with verbatim evidence quotes, pain points with the client's own numbers, scope signals tagged requested/implied/deferred, budget signals quoted exactly (never normalised into a number they didn't say), timeline gates with their consequences, open questions, and red flags for your eyes only. You correct anything it misheard and approve. That two-minute read is the step that catches the wrong number before the client does. Stage 2 drafts the proposal from your approved facts, under non-negotiable rules. Every price comes from your rate card with the maths visible ("Template design × 4 @ 1,450 — 5,800"); work with no rate-card match gets [TBC], never a market guess. Three tiers by default — leanest-viable, recommended, stretch — with subtotals that exactly equal their line items. Budget signals position the tiers but never move a price: a tier above the client's stated range says so plainly instead of hiding. Scope only enters from the call. And red flags shape the document silently — a story about a bad former agency becomes staged milestone payments presented as standard practice, with the story never mentioned. The voice is yours because you build it: the calibration worksheet extracts a voice profile from your own past proposals, and every sentence applies it. The pack's worked example runs a realistic 29-minute e-commerce discovery call (Saltbush Supply Co.) through the full pipeline — messy transcript, fact sheet, three-tier proposal in the sample seller's voice, and the same proposal as validated JSON — so you can reproduce the samples and know your setup is exact before the first real call. The Workbench makes it a tool rather than homework: a single local HTML file with the system prompt embedded — paste transcript, load rate card, one click to copy the assembled prompts into Claude or ChatGPT, or run directly with your own Anthropic API key (browser to API, key never stored). A one-click handoff moves the approved fact sheet into Stage 2. For pipelines, the JSON schema and renderer prompt turn BriefBot into an API step with machine-checkable output. The live demo is the actual Workbench preloaded with the sample call — press Run and watch both stages produce the shipped sample outputs.

Version history

1. v1.0Latest5 Aug 2026

Licence & terms

These terms are snapshotted at purchase — what you see here is what applies to your order.

Reviews