Evidence for every finding
Each step links to the page, ad or file it came from, and is marked seen or inferred. The evidence is saved next to the report.
// REG by Parsly · Free AI competitor research
Ask your AI assistant how a company wins its customers and get the whole funnel back, from their ads to their upsells, with the evidence for every finding. Free. Runs in Claude Code today, with more AI tools on the way.
In Claude Code: the terminal, or the Code tab in the Claude app.
The funnel
// What you get
Ad spy tools stop at the ad. REG follows it: where it sends people, what happens next, and how the business makes its money back.
Most competitor analysis gives you a list of a rival's ads, or a traffic estimate, or a generic SWOT. None of them tells you how the business actually works. REG does the research a good marketer would do by hand: it reads the company's Facebook and Instagram ads in the Meta Ad Library, its Google, TikTok and LinkedIn ads, follows each ad to its landing page, reads the sign-up or booking form, works out when the card goes in, and traces the upsells and referral schemes that keep customers paying. The result is a competitor funnel analysis you can act on: which channels bring their customers in, which offers and angles they keep paying to run, and where their funnel is weak.
Every ad on Meta, Google, TikTok and LinkedIn, how long each has run, the angles, and where each one leads.
The offer, the proof, the price, and every version they're testing.
The form or quiz, what it asks, and where each answer sends you.
Sales calls, chase-up pages, countdowns, what gets tracked and reported to the ad platforms.
When the card goes in, the trial, every plan and price.
What the product really runs on: their own software, a white-label, a team.
Higher plans, add-ons, lock-in, and what happens when you try to cancel.
Affiliate and referral schemes, and what they pay per customer.
The founders, their audience, and the other things they sell.
Their spend, conversions and churn stay private, and the report says so.
// A real report
We pointed REG at AG1, the greens powder, and published what came back: every step of the funnel, with the evidence.
It's a fair test of AI competitor research: a big brand with thousands of ads, dozens of landing pages and a sophisticated site. REG found the shape of their paid social, the pages built for each podcast and audience, a visitor classifier in their code, and the places where their own numbers disagree.
In their code
AG1's site sorts each visitor into one of nine groups by where they came from, including people arriving from competitors' sites.
In the ad library
205 of the 418 US Meta ads we loaded ran as partnership ads across 31 creators, many linked through a creator-ads tool.
Claims to check
The same free welcome kit was worth $43 on one AG1 page and $72 on another, on the same day.
// Evidence and limits
The risk with AI for competitor research is that it sounds confident about things it made up. REG works the other way round: it only reports what it can point to, keeps what it saw apart from what it inferred, and is plain about the limits of public evidence. A competitor's ad spend, conversion rates and email sequences are private, and the report says so instead of inventing numbers.
Each step links to the page, ad or file it came from, and is marked seen or inferred. The evidence is saved next to the report.
Customer counts that change from page to page, "limited spots" that never run out, testimonials that look like placeholders: listed, with where each appears.
Every report ends with what's private (spend, conversion rates, follow-up emails) and what would show it.
// How it works
REG is a skill for your AI assistant: a set of instructions that tells it how to research a business properly, in what order, and when the job is done. You add it once, then ask about any company in plain words. Your assistant does the browsing and reading on your computer, then writes the report into your folder, with the evidence it collected saved beside it.
The offer, the price, the main ad angles and the funnel in outline.
Every step of the funnel with its evidence: every link, the login and sign-up pages, the site's own code, every ad and where it leads.
Medium, plus the people behind it, company records, archived pages, other ad networks, size estimates and a skeptic's check of every claim.
// Why not funnel hacking?
The usual way to reverse engineer a competitor's funnel is funnel hacking: buy the product, join the email list, book the call and take notes. It works, but it's slow, it puts you in their sales pipeline, and it pollutes their data. REG gets most of the same picture from what any visitor can see, and goes further in places a visitor wouldn't look, like the code the site sends your browser.
Sign up, fill in their forms, sit through their emails and calls. Slow, it puts you in their sales pipeline, and it skews their data.
Paste a link, get a generic SWOT. Fast, but it's mostly guesswork, with nothing to check it against.
Reads only what's public, goes deeper than a visitor would (the site's own code, every ad and where it leads), and shows the evidence for each finding.
// Who it's for
Before you build, see how the businesses already making money find their customers, what they charge, and what they'd struggle to match.
Competitor research for your next campaign: the ad angles a rival keeps running, the landing pages behind them, and the offer that closes the sale.
A sourced competitor analysis for a client in an hour, not a week, with every finding linked so you can show your working.
It sits alongside the other AI skills for marketing: where most help you write copy or plan content, REG does the competitor research that should come first.
// Also in REG
// Questions
Yes. REG is free and open source (MIT licence). You need an AI assistant that can run it: today that's Claude Code, which comes with a paid Claude plan.
Claude Code today. REG is written as a skill, a plain set of instructions rather than code, so it can run in other AI assistants too; support for more of them is on the way.
Yes, if it's told how. Left alone, an AI assistant will guess. REG gives it a method: which public sources to check, in what order, how to read them, and when the job is done, so the competitor analysis comes back with evidence instead of guesses.
No, never. It reads what any visitor can see: their pages and every link on them, the code their site sends your browser, public ad libraries, domain records and reviews. It looks at forms and login pages but never submits them, signs up or logs in, and it doesn't get past bot checks.
The Meta, Google, TikTok and LinkedIn ad libraries are public on purpose, so anyone can see who is advertising what. REG only reads public pages, gently, and treats what it finds as research: learn the patterns, don't copy anyone's ads or words.
Not directly: US and UK ad libraries don't show spend, and revenue is private. REG estimates size from public clues (team, reviews, prices, ad volume), labels it as an estimate, and shows how it got there.
Meta's Ad Library shows every ad a page is running, free and without logging in. REG reads it for you, notes how long each ad has run, and follows each one to its landing page, so you see the ads and the funnel behind them together.
Those each show one piece. REG follows the whole path, from the ad to the page, the sign-up, the checkout and the upsells, and puts it together into how the business actually works, with the evidence for each step.
Anything behind a login or a sign-up: the follow-up emails and texts, the sales call, real conversion rates and churn. Every report ends with a list of what couldn't be seen and what would show it.
Businesses that advertise and sell online: software, online shops, agencies, services with a sign-up or booking form. Companies that sell mainly in person leave less to find.
In Claude Code: the terminal, or the Code tab in the Claude app.