The problem
Every prospect needed a 20-plus-page competitive analysis: demographics, competitor discovery and verification, review analysis, AI-search visibility, assembled into a branded deck, followed by a proposal priced in a spreadsheet and built by hand. It took a skilled person most of a day, depended on the few who knew the method, and quality varied. Figures went unverified, template sample content leaked into client decks, and pages that should have been removed shipped anyway.
What I built
Competitive analysis runs the full SOP as code: competitor discovery through the Google Places API, six-point verification per competitor including API-verified driving distances and AI-assistant visibility across Google AI Overviews, ChatGPT, and Perplexity, Census pulls, generated positioning visuals, and a 22-page report delivered into the agency template. The proposal generator runs the agency's own pricing calculator verbatim as the source of truth, cross-checked every build against an independently tested port, and renders the money pages pixel-identical through a headless pipeline so no AI restyles or recomputes a number. The strategy playbook produces wins, fixes, and easiest-wins reviews for existing clients.
The trust framework
Every claim lands in one of three tiers. Durable verified data is checkmarked. Perishable snapshots, like what an AI assistant answered today, are date-stamped and never ranked. Adverse claims about named parties are hard-blocked until a named human reads the primary source and signs off. That last rule came from a real caught failure: an early build asserted a legal claim, based on coverage of coverage, that the primary source did not support.
A pre-push gate shows a manifest of every research step as done, fallback, or disclosed unavailable. A post-generation read-back diffs the finished deck against the source data line by line. Each failure found in live runs, sample pricing leaking, AI restyling breaking brand formatting, image-only pages defeating name-based removal, became a named rule with its fix.
Result
A day of skilled work became about 90 minutes, mostly unattended, and anyone on the team can run it. First production run: 54 competitors verified, 10 deep dives, 14 API-verified distances, and a data-backed answer to the client's central question. Roughly 20 versioned releases followed, each driven by a defect from a live run. Setup for a new person is three files and ten minutes.
In brief
- What is AI Report Automation?
- Three packaged, versioned AI skills that produce a marketing agency's competitive analysis reports, priced proposal decks, and internal strategy reviews, with a trust framework that decides what the system is allowed to claim.
- What did Brian Powell build for AI Report Automation?
- Competitive analysis runs the full SOP as code: competitor discovery through the Google Places API, six-point verification per competitor including API-verified driving distances and AI-assistant visibility across Google AI Overviews, ChatGPT, and Perplexity, Census pulls, generated positioning visuals, and a 22-page report delivered into the agency template. The proposal generator runs the agency's own pricing calculator verbatim as the source of truth, cross-checked every build against an independently tested port, and renders the money pages pixel-identical through a headless pipeline so no AI restyles or recomputes a number. The strategy playbook produces wins, fixes, and easiest-wins reviews for existing clients.
- What was the result of AI Report Automation?
- A day of skilled work became about 90 minutes, mostly unattended, and anyone on the team can run it. First production run: 54 competitors verified, 10 deep dives, 14 API-verified distances, and a data-backed answer to the client's central question. Roughly 20 versioned releases followed, each driven by a defect from a live run. Setup for a new person is three files and ten minutes.