Frequently asked questions
Money & expectations
How much do I need to spend to find out if this works for me?
$300–$1,000 in Facebook ad spend over the first 1–4 weeks is the honest minimum. Plus a $5–10/month hosting bill if you do Module 11. Plus a few dollars of Anthropic API spend. The AI operator does not reduce your test spend — it reduces your test time and the number of bad tests you fire before catching a winner.
How long until I make money?
Paid traffic: best case, week 2. Realistic case, weeks 4–8 to find a profitable combo. Plenty of beginners go 2–3 months. Organic (SEO + YouTube): 3–9 months before meaningful traffic, 12–18 before consistent revenue.
How much do top affiliates make?
One profitable offer with one scaled adset: $40–$120/day net profit is typical. Top affiliates run dozens of offers across multiple ad accounts and clear $5k–$50k/month. The path from $0 to $40/day is the hard part; from $40/day to $1k/day is mostly horizontal duplication (Module 09).
Why is everything free? What's the catch?
There isn't one. The model + browser stack is generic and easy to copy; the moat for affiliate marketing is taste in offers and angles, not the automation. Giving away the tool floor-raises the niche.
The AI operator
What does the operator actually cost to run?
Pennies. Each agent step is ~$0.005–0.02 in Claude API. A full week of testing is typically under $5 in model spend on top of your ad budget.
Does it work on Windows?
Yes. WSL2 is the smoothest path, but native Windows works too (use the .zip download). Playwright + uv both support Windows natively.
Does it need a GPU?
No. Everything runs on CPU. The ML inference (Claude) happens server-side at Anthropic.
Can I run it on a Raspberry Pi / cheap VPS?
The agent logic, yes. The headed-browser layer needs a real desktop environment, so a headless VPS won't work for the browser-driven tools. Use a VPS for the cron-based content runners; run paid-traffic flows on a real laptop.
How is this different from Make.com / Zapier / N8N automations?
Those tools chain pre-built actions on a fixed graph. The operator is a planning agent that picks its own actions per step. Better for tasks where the order isn't known up front — exactly the case in affiliate testing.
Can I use a different LLM (OpenAI, Gemini, local)?
The orchestrator currently calls Anthropic's SDK directly. Swapping to OpenAI is a ~30-line change in orchestrator/agent.py — both APIs have nearly-identical tool-use schemas. Local models (Llama via Ollama) work but are noticeably worse at long, multi-tool planning chains today.
Will Facebook detect that a bot is running my account?
Probably not from the browser behavior alone — the human-motion layer + stealth patches handle the obvious tells. The detection risk is mostly from your actual marketing behavior: spammy creative, suspicious landing pages, policy-violating ad copy. Those would get a human banned too.
Why headed Chrome instead of headless?
Headless Chrome leaks ~10× the bot signal of headed. The visibility cost (a Chrome window opens on your desktop) is worth the trust gain. If you genuinely need server-side runs, see the --headless flag in browser/playwright_human.py but expect more rejections.
How do I add my own tools to the operator?
See the architecture page. One decorated Python function. ~10 lines.
Facebook & ads
I just got banned. What do I do?
Read Module 10 in full. The TL;DR: don't panic-resubmit, request review once, switch to your backup admin profile to keep BM access, rotate to a fresh ad account or agency account, swap in a new virtual card.
Do I need a real business / LLC to do this?
Not to start. Facebook accepts personal use of ad accounts. Networks (Clickbank, Digistore24) pay individuals via PayPal or wire. At ~$2k/mo in income, register an LLC for liability + tax reasons. Talk to an accountant in your jurisdiction.
What's a "lookalike audience" and how do I make one?
See Module 07 — Audiences. Short version: you give Facebook a list of people you know are good customers (a "seed"); FB finds 1–10% of the country who match. The seed can be hashed emails from your CRM, or it can be your pixel's automatic visitors after a few hundred have visited.
Can I use the hashed audience files I've seen shared in the affiliate community?
Technically yes (Facebook accepts the file format), but you're handling personal data of third parties without consent. Under GDPR/CCPA that's a real legal issue. Better path: build your own audiences from your pixel data once you have lander traffic.
The course
Is the course included with the download?
The course is the website you're on. All 13 modules are free-to-read. The operator download includes a copy of the playbook in Markdown so it works offline too.
Will you add video versions of the modules?
Maybe later. Written-first is intentional — searchable, skim-able, easy to update.
Can I translate the course / operator into my language?
Yes. MIT license. Fork it and translate. Send a link and I'll feature it here.
How do I report a bug / suggest an improvement?
Money & payouts
When and how do Clickbank / Digistore24 pay?
Clickbank: weekly or bi-weekly to your bank account / Payoneer / check. Threshold $100 default (adjustable). 4–6 week initial holdback on new accounts. Digistore24: similar, monthly to bank or PayPal, faster for European accounts.
Do I need to pay taxes on this?
Yes. It's self-employment income. Track every dollar in (commissions) and every dollar out (ad spend, hosting, software). Net is taxable. Talk to an accountant in your country.
Can the operator refund / chargeback fraud cost me money?
Clickbank's chargeback rate is around 1–3%. Refunded sales are clawed back from your commission. If your refund rate exceeds ~10% on an offer, Clickbank will warn or restrict your account — usually means the offer has a quality problem, switch.
Didn't find your question? Reach out →