AI Guides › Playbooks
By Nigel Guy · 8 min read
Most side hustle advice runs on vibes: someone lists ten "AI income ideas", none of them dated, none of them sourced, and you spend a month building a portfolio for a lane nobody is buying. It feels productive because you are learning tools. It fails because you never checked whether anyone pays for the output, or whether the price is already falling.
The rule: pick a lane only from dated, sourced demand data, then run a weekend test with a pass mark you write down before you start. If it misses the mark, you drop it.
These are the three most useful public datasets on AI-assisted freelance demand at time of writing. All three come from marketplaces that have a commercial interest in looking busy, so treat them as signals, not guarantees.
| Finding | Source and date | What it measures | The catch |
|---|---|---|---|
| Upwork skills that explicitly reference AI grew 109% year on year. Fastest growers: AI video generation and editing (+329%), AI integration (+178%), AI data annotation and labelling (+154%), AI image generation and editing (+95%), AI chatbot development (+71%) | Upwork In-Demand Skills 2026, published 4 February 2026 | Freelancer earnings on completed jobs, 2025 against 2024, US-originated demand only | US clients, not UK. Percentage growth from a small base can look dramatic |
| Fiverr searches for Claude Code specialists up 938%; n8n AI automation +125%; AI UGC video ads +265%; AI-related services in Video & Animation +278% against Programming & Tech +94% | Fiverr Business Trends Index 2026, published 9 June 2026 | Global search volume, November 2025–April 2026 against May–October 2025 | Searches are interest, not purchases. Six-month windows swing |
| Freelancers doing AI work earn 34% more per hour, but generative AI and creative production saw contract starts up 90% while per-contract earnings fell 13%. Complex AI-augmented professional work saw earnings up 45% | Upwork Future Workforce Index 2026, published 14 July 2026 | Survey of 2,400 US skilled workers plus Upwork platform data | US-focused. "Complex" is Upwork's own LLM-based classification |
Read together, the third finding is the one that matters. Volume is rising fastest in simple AI production work, and that is exactly where pay per job is falling. The premium sits where you bring judgement or a domain you already know, with AI as the tool.
Map the data onto three lanes, then pick the one that sits on top of something you already do.
| Lane | Signal behind it | Who it suits | Price pressure |
|---|---|---|---|
| AI video editing and short ads for small businesses | Upwork +329% (earnings), Fiverr AI UGC ads +265% (searches) | People who already edit video or run social accounts | High: lots of new sellers; per-contract earnings in creative AI work fell 13% |
| Workflow automation (n8n, AI integration, simple agents) | Upwork AI integration +178%, Fiverr n8n +125%, Claude Code +938% | People comfortable with spreadsheets, APIs or light code | Lower for now, but the client must trust you with their systems |
| Your existing profession, AI-augmented (bookkeeping, HR, marketing, research) | Upwork: AI-augmented professional services volume +72%, earnings +22% | Anyone with a trade or specialism | Lowest: you compete on domain judgement, not tool access |
The third lane is the least exciting and, on the Upwork data, the best paid. Skip it only if you have no specialism to bring.
Before you touch a tool, fill in four lines:
Write the platform fees into the maths. At time of writing, Fiverr credits sellers with 80% of each completed order, and Upwork's freelancer service fee ranges from 0% to 15% per contract. Check both pages before you price.
Search Upwork jobs and Fiverr gigs for your lane. Copy 15–30 listings into a document: title, budget or price, what the buyer actually asked for. Then paste them into Claude or ChatGPT with this prompt.
You are a freelance market analyst helping someone decide whether to offer a service as a side business in the UK.
Context: I am considering this lane: [LANE_DESCRIPTION]. My relevant experience is: [YOUR_EXPERIENCE]. Below are [NUMBER] job posts and competitor listings I copied from freelance marketplaces on [DATE].
Goal: tell me whether there is paying demand I could realistically win, based only on the listings provided.
Steps:
1. Group the listings by what the buyer actually needs, not by the words they used.
2. For each group, give the count, the price or budget range, and any repeated complaints or requirements.
3. Identify where buyers ask for judgement, domain knowledge or accountability, and where they only want cheap output.
4. Compare the price range with my target of £[TARGET_HOURLY_RATE] an hour after a fee of [PLATFORM_FEE]%.
Output: a short table of groups, then three bullet points: strongest opening, weakest opening, and one fixed-scope offer I could test.
Constraints: use only the listings below. Do not invent market sizes, growth rates or prices. If a listing has no price, say so. If my experience or target rate is missing, ask me before answering.
Before you answer, check that every number you give appears in or is calculated from the listings.
Listings:
[PASTE_LISTINGS]
Fill in your lane, experience, date, target hourly rate, the platform fee and the pasted listings.
Do not build a portfolio site. Do one of these: send three tailored proposals to live job posts, publish one fixed-price gig, or message three people you already know who fit the buyer description. Use this prompt to draft the offer.
You are an experienced freelancer who writes plain, specific proposals that win work without overselling.
Context: I offer [SERVICE] to [BUYER_TYPE]. My proof is: [REAL_EXAMPLES_OR_RELEVANT_EXPERIENCE]. The job post or person I am writing to: [PASTE_JOB_POST_OR_DESCRIBE_CONTACT].
Goal: a proposal under 150 words that shows I understood their problem and offers one fixed-scope first job.
Steps:
1. Restate their problem in one sentence using their details.
2. Propose a small first deliverable with a clear scope, turnaround and price of £[PRICE].
3. Say which parts use AI tools and which parts I check by hand.
4. End with one easy question they can answer yes or no.
Constraints: British English. No claims of results, clients or credentials I have not listed above. No hype words. If the proof section is empty, ask me for it rather than inventing any.
Before you answer, reread the draft and remove any sentence that would be untrue if a client asked me to prove it.
Fill in your service, buyer type, genuine proof, the post or contact, and your price.
On Sunday night, score the card against the pass mark you wrote on Friday. Hit it: run a second weekend with paid work as the goal. Missed it: drop the lane and test the next one. Don't move the pass mark.
Priya is a part-time bookkeeper in Leeds. Her card: "AI-assisted month-end tidy-up for UK sole traders, £120 fixed." Pass mark: 20 matching posts in 30 days and a typical budget that clears £25 an hour after a 15% fee. Her scan finds 14 matching posts, mostly US clients wanting US tax software. She misses the mark, so she drops the marketplace route and instead sends the proposal to three local sole traders she already knows. One says yes. Her next card tests the local route, not the marketplace one.