Three uncomfortable truths
- "AI + blog" is the most crowded strategy on earth. Search engines now explicitly penalise masses of pages generated at scale, whatever tool made them. A site of 300 AI-written articles about productivity will not get approved for ads, and if it does, it will not rank.
- South African traffic is cheap. Ad rates for local readers run at a fraction of US or UK rates. 20,000 visits a month from South Africa might earn R2,000–R3,000 in advertising, not R30,000.
- Affiliate programmes for South African retailers are thin. Some large marketplaces run small-commission programmes, others have no local affiliate scheme at all. Budget for direct deals with niche stores rather than assuming a big commission.
None of that means don't do it. It means the winning version is narrow, useful and hard to copy — not broad and cheap to generate.
Picking a niche that can win
Score any idea on four things, out of five each:
| Test | What it means | Fail condition |
|---|---|---|
| Buying intent | People search it just before spending money | Only curiosity searches ("what is…") |
| First-hand data | You can produce prices, photos, tests nobody else has | You would be summarising other sites |
| Advertiser value | Advertisers pay well for this reader | Hobby audience with no purchase |
| Copy resistance | A competitor cannot reproduce it in a weekend | It is a list of publicly available links |
A niche like "Korean groceries in South Africa, with prices we checked" scores 4/4: buying intent, first-hand data, a food-and-retail advertiser market, and copying it requires actually going to shops. A niche like "AI tools for productivity" scores 1/4 and is where most people start.
What to actually build
Four layers, in this order:
- A data spine. 50–200 products or places, each with a real attribute set: price, size, where, date checked. This is the thing nobody else has. It takes a week and it is the moat.
- Ten to fifteen real pages. Not 100. Each one answers a specific question, has the data embedded, and says when it was checked. Quality is the approval path.
- One tool. A calculator, filter or comparison table that does something the reader could not do with a static page. Tools earn links, and links are still how a new site gets found.
- The boring pages. About, how we check prices, contact, privacy, disclosure. Reviewers look for these; readers rarely do.
Where AI genuinely helps: turning your data into structured tables, drafting the first version of a guide you then rewrite, and generating the schema markup. Where it hurts: writing pages that have no data behind them.
How it earns, in order
- Affiliate (month 1+). A niche store or two with a real programme, plus direct deals where no network exists. Keep links labelled and few — a page that is mostly affiliate links gets refused by ad networks and ignored by readers.
- Display ads (month 3–6). Realistically 15–25 substantial pages before applying. Expect a rejection the first time and fix the thinnest pages rather than resubmitting blind.
- Direct sponsorships (month 6+). A shop pays for a clearly marked "featured" slot once you have traffic in one city. This is usually more money than ads at small scale.
- Digital product (month 6+). A R150–R250 guide built from the data you already collected, sold to the subset who want it all in one file.
The maths, honestly
| Month | Visits/mo | Ad revenue (low-rate country) | Affiliate + sponsor |
|---|---|---|---|
| 1–3 | 30–300 | R0 | R0–R200 |
| 4–6 | 500–2,000 | R40–R250 | R200–R800 |
| 7–12 | 2,000–8,000 | R250–R1,200 | R800–R3,000 |
| 12–24 | 8,000–30,000 | R1,000–R4,000 | R2,000–R8,000 |
If you want this to be a business rather than a hobby, point the same site at a higher-value market: a niche with US/UK/AU readers and clear buying intent can earn five to ten times per visit. That usually means writing in English for a global reader, while keeping a local section as your first-hand data advantage.
What breaks it
- Publishing speed. Two or three real pages a week beats twenty generated ones. Networks and search engines both notice the difference.
- Thin pages. Tag pages, paginated archives and empty category pages should be kept out of search. Low-value pages are the most common reason a site fails review.
- Copying. Summarising competitor articles without adding data is the fastest way to be filtered out of search results.
- One traffic source. Search updates happen. Build an email list or a following on one platform from month one so you are not renting all your traffic.
First 90 days
- Week 1–2: Choose a niche that passes the four tests. Buy the domain. Build the data spine with 50 items.
- Week 3–4: Publish the data page and five guides that use it. Every page gets a date.
- Week 5–6: Add the tool (calculator or filter), then three more guides. Apply to one affiliate programme.
- Week 7–10: Two pages a week, one outreach email a day to a small blog or forum that would genuinely link to the data page.
- Week 11–13: Reach 15 substantial pages. Apply for ads. Fix what the reviewer would call thin.
Recommended AI stack for this playbook
You do not need all of these. Start with the first two and add the rest when a client pays for them.
- HostingerCheap hosting for a one-page offer or a client micrositeCheck price →
- CloudflareDomains, DNS and free SSL for client micrositesCheck price →
- NotionOnboarding docs, SOPs and delivery checklistsCheck price →
- CanvaTemplates, brand kits and client-ready decksCheck price →
Affiliate disclosure: these are affiliate links. If you subscribe we may earn a commission at no cost to you, and it does not change our recommendation. Vendor prices change — check the vendor page.
A niche site pays in year one, not month one. If you need cash now, run the receptionist or automation playbook and build the site with the profits.
Check the maths with your own numbers
Load this playbook's model into the calculator and change the price, hours and churn to match your market.
Other playbooks
The WhatsApp AI receptionist agency: R3,500 a month per shop
Every small shop in the country already runs its business on WhatsApp. Almost none of them answer it properly. Here is the offer, the price, the tool stack and the maths.
Service businessAI short-form video for local businesses: R2,900 a month for 12 posts
Local businesses know they should post. They will pay someone else to do it. The AI part is the edit, the captions and the scripting — not the trust.
Productized serviceThe R6,500 one-time AI setup for a small shop
A fixed-scope, fixed-price package you can deliver in two days: Google presence, FAQ bot, review replies and a simple content engine.