AI Guides › Playbooks
By Nigel Guy · 7 min read
The usual AI side-income plan is volume: point a model at a niche, publish forty videos, pages or ebooks a week, and wait. It feels like progress because output is cheap and the dashboard fills up. It is also the exact pattern platforms now write rules against, and when two people use identical tools, the one who publishes more generic output is not the one who gets paid.
The rule: sell a specific result to a named person, use AI to deliver it faster, and keep a human check between the model and the customer.
Honest means three things. You expect the first month to earn nothing or close to it. You treat AI as a production tool, not as the product. And you can describe your offer without the word "passive".
I have no verified figures for what typical AI side-incomes earn, and nobody selling a course on it can supply a trustworthy average either, so none appear here. What you can verify is how platforms treat mass-produced AI output:
Policies change, so check each page before you build on it (links below). The durable point: the thing that gets demoted is interchangeable output at scale. The thing that survives is work with a person's judgement visibly in it.
A ledger is one table you fill in at each step. Five rows, in order. Don't move on until a row is filled.
| Step | Question the row must answer | Evidence that counts |
|---|---|---|
| 1. Buyer | Who exactly pays, and what do they pay for today? | A named type of person you can message this week |
| 2. Result | What finished thing do they get, and by when? | One sentence with a deliverable and a deadline |
| 3. Human check | What do you personally verify before it leaves? | A written checklist, not a feeling |
| 4. Proof | Which three jobs, even free, show the result? | Before and after samples, with permission |
| 5. Price | What is the fixed fee and what is excluded? | A number in £ and a scope line |
Pick a buyer you can reach without a platform algorithm: local businesses, a former industry, a community you belong to. Write the result as something finished: "twelve product descriptions in the shop's tone, returned in two working days", not "AI content services".
This is the row that separates you from anyone else with the same tool. Models invent facts, names and prices with confidence. Your checklist should cover facts, tone, anything legal or financial, and whether the client's confidential data was pasted somewhere it should not have been. Use this prompt to draft your own checklist:
You are an experienced editor who reviews AI-assisted work before it goes to a paying client.
Context: I deliver [SERVICE, e.g. product descriptions] to [CLIENT_TYPE]. A typical job is [TYPICAL_JOB].
Goal: write a review checklist of no more than 10 items that I run on every deliverable before sending it. A good checklist catches invented facts, wrong tone, privacy leaks and anything the client would be embarrassed by.
Steps:
1. Ask me up to 5 questions if you need more detail about the service, the client's tone, or the risks. Do not guess.
2. Produce the checklist as a table: check, how to do it in under a minute, what a fail looks like.
3. Mark the 3 items most likely to fail for this kind of work.
Constraints: no generic advice such as "proofread"; each item must be specific to [SERVICE]. Do not invent legal requirements; where a rule might apply, say "verify with an official source".
Before answering, check that every item is something a person can actually do and tick.
Fill in the square brackets with your service, client type and a typical job.
Do three jobs for people you know, with their consent to show the result. Keep the brief, your draft process and the final output. Do not use fake testimonials or stock case studies; if you have no results yet, say you are new and show the samples.
Quote per job, not per hour, because AI makes your hours shorter and you do not want your fee to shrink with them. State what is excluded, such as extra revisions. Raise the price when you are fully booked, not before.
Asha works in a dental practice and spots that local practices have patchy websites and patient reminder emails. She fills in the ledger.
A second person has the same assistant and instead posts daily "dental tips" to a new channel. Her output is larger, generic, and sits exactly where a platform's reviewers are told to look. Asha's output is smaller and has a named buyer, a check and a reason to be paid. Neither is guaranteed income; only one is a business you can improve.
Judge by inputs you control for the first eight weeks: conversations started, quotes sent, jobs delivered. Then by whether any client came back or referred someone. Reach for views and follower counts last; they were never the money.
A tracking prompt, if you want help reviewing your own numbers:
You are a sober business coach for a part-time freelancer.
Here are my last [NUMBER] weeks of data: conversations started [CONVERSATIONS], quotes sent [QUOTES], jobs delivered [JOBS], repeat or referral clients [REPEATS], hours spent [HOURS], fees in £ [FEES].
Goal: tell me which step of the funnel is weakest and one change to test next week.
Constraints: if the numbers are too few to conclude anything, say so rather than inventing a trend. Do not promise income. Ask for any missing figure instead of guessing.
Before answering, check that your advice names a single change I can test in seven days.