Editing technical documentation for early-stage AI startups
Small AI teams ship features faster than their docs. A weekly retainer to keep their docs clean turns into steady freelance income.
How it works
Founders know their docs are bad but hiring a full-time writer is premature. A part-time editor with taste — someone who can read a PR and turn it into a changelog — is a rare fit. Retainers of $1-3k/month per client are common.
Step by step
- 1Shortlist 15-20 seed-to-Series-A AI startups whose docs are visibly rough. Use their public GitHub, changelog and 'Docs' subdomain — grade each on typos, missing examples, and broken code blocks.
- 2Pick the 3 with the best product but worst docs. Fork the repo, read the last 20 merged PRs, and understand the product enough to write about it accurately.
- 3For each, submit 1-2 high-quality documentation PRs for free. Fix a broken example, rewrite a confusing page, or add a missing quickstart. Keep the PR small — one clear improvement each.
- 4In the PR description, include a two-sentence note: 'Happy to keep helping with docs as you ship. Reply here or DM if useful.' No pitch deck, no rates.
- 5When one founder replies with interest, propose a paid trial: 4 hours a week for 4 weeks at $150/hr, invoiced weekly. Deliver a Loom recap every Friday.
- 6After the trial, convert into a rolling monthly retainer at $2-3k. Define scope in one paragraph: 'Weekly changelog, quickstart maintenance, up to 2 new pages per month.'
- 7Repeat until you have 3-5 retainers. Stagger start dates so onboarding doesn't overlap.
- 8Build a private notes library of common AI-doc patterns (evals, RAG, tool-calling) so each new client onboards in one week, not four.
- 9Once you're booked, raise rates on the next new client by 20% and phase out your lowest-margin one.
Discussion
The free PR trick is underrated. It's the modern cover letter.
Related methods
- Translating public-domain books and selling them on Kindle
An idea I keep coming back to: use modern translation tools + human editing to bring forgotten public-domain books to English readers.
- Going deep on a single bug bounty program
Instead of chasing every program on HackerOne, I picked one large target and learned it end-to-end.
- Voice-over work for B2B explainer videos
A small home booth and a Fiverr Pro profile paid off my mortgage over four years.
- Tutoring high-school math over Zoom, evenings only
Six regular students, three evenings a week, $70/hour. Pays a mortgage on top of a day job.