An AI Runs This Website. Here Is Exactly What It Earns.
£1,212.50 in July 2026. Every source, every cost, everything that failed. Updated monthly, past months never rewritten.
£1,212.50
July 2026
£2,890.50
Tracked since June 2026
1 to 3 hrs
My time per week
~£11
Running costs / month
How this started
It started with a credit card. When I signed up for Capital on Tap I did what everyone does and went hunting for a referral code, and the AI I asked found one on some stranger's website in seconds. That stuck with me. When my own card arrived I thought: it would not take long to put my code on my own website. So I did.
Then the post started ranking, and referral commission started landing in my account. That felt slightly unreal, so I found a referral code for every tool I genuinely use and recommend, wrote those up too, and eventually handed the whole publishing operation to an AI. I run a marketing agency by day. This site is what happens when I point the same systems at my own corner of the internet.
The numbers
July 2026 by source
Every tracked month
| July 2026 | Payouts | Amount |
|---|---|---|
| Octopus Energy referrals | 6x | £300 |
| Wise referrals | 10x | £240 |
| Tide referrals | 2x | £200 |
| Capital on Tap referrals | 2x | £150 |
| Monzo referrals | 15x | £85 |
| Book sales | 18x | £77 |
| TopCashback | 3x | £75 |
| Hostinger referrals | 1x | £65.50 |
| Audiobook sales | n/a | £20 |
| Total | £1,212.50 |
Down 28% on June. Capital on Tap fell from 12 referrals to 2 after its main post lost rankings in the May Google core update. Tide paid out for the first time and Wise more than doubled.
| June 2026 | Payouts | Amount |
|---|---|---|
| Capital on Tap referrals | 12x | £900 |
| Octopus Energy referrals | 9x | £450 |
| Book sales | 19x | £76 |
| Monzo referrals | 15x | £70 |
| Wise referrals | 4x | £54 |
| TopCashback | 2x | £50 |
| Audiobook sales | 12x | £48 |
| Hostinger referrals | 1x | £30 |
| Total | £1,678 |
First properly tracked month. Capital on Tap referrals landed from sign-ups earlier in the spring, when the post still ranked around position 4.
Costs, and honest profit
Revenue reports without costs are marketing. Here is what this site costs to run each month:
- Claude subscription (this site's share): £10 (roughly 5% of a claude max plan. the rest of the subscription runs my agency and other projects)
- Vercel hosting: £0 (free tier)
- Domain: £1 (about £11 a year)
Call it roughly £11 a month all-in, so effectively everything above is profit before tax. The real cost is the 1 to 3 hours a week of my time, and the months of unpaid work before any of it earned a penny.
How the machine works
On weekday mornings at around 2am, a scheduled Claude routine wakes up, picks a target keyword from a weekly rotation, researches what already ranks, and queues a content brief. At around 5am a second run reads that brief, writes the full post, commits it to GitHub, waits for Vercel to deploy, then loads the live page and checks its own work. On Fridays it audits existing pages and emails me a weekly report instead.
The routine follows hard rules I wrote for it: UK English, no fabricated claims, and it may only recommend products I personally use. An approved affiliate programme is not a reason to recommend something, and it never adds money links to the mental health writing on this site. My 1 to 3 hours a week go on reviewing its briefs, making the strategy calls, and fixing what it flags.
What has not worked
This section is the part every other report leaves out, and it is the part I would want to read.
- Affiliate networks earned me £0. I applied to over 30 programmes on AWIN. Months later most are still pending and the approved ones have paid nothing. Every pound this site has made came from direct referral programmes for products I already used.
- Google took away my best earner. The Capital on Tap post ranked around position 4 and drove 12 referrals in June. A core update knocked it to page 2, two recovery attempts have not brought it back, and July fell to 2 referrals. Because referrals lag clicks by weeks, August may look worse before it looks better. I am telling you that now rather than explaining it away later.
- AI Overviews eat clicks. My Wise page gets over 3,400 impressions a month on its main search term and almost nobody clicks, because Google answers the question above the results. The pages that still earn are the ones matching questions AI answers badly.
- The AI died for seven weeks and I did not notice. The publishing routine silently vanished in June. A separate small automation kept making cosmetic updates, which masked it. Nobody noticed until a deep audit in late July. It is rebuilt now, with better self-checks. Automation fails quietly; assume yours is failing until it proves otherwise.
Methodology
- Numbers come from my personal income tracker and count payouts that landed in the month, not clicks or pending commissions.
- Tracking started properly in June 2026. The site earned before then, but I was not recording it systematically, so this page does not guess at lifetime figures.
- Book and audiobook royalties are included because they are part of this site. My agency's client revenue is not, and never will be.
- Costs are itemised above; the Claude figure is an honest apportionment of a subscription shared with my other work.
- Past months are append-only. When a new month is added, old ones stay exactly as published.
Frequently asked questions
- Is this website really run by an AI?
- The publishing is, yes. A scheduled Claude routine researches a target keyword in the early hours, writes the post, commits it to GitHub, waits for the deploy and checks the live page itself, twice a day on weekdays. I spend 1 to 3 hours a week steering: reviewing what it did, making strategy calls, and fixing things it flags. Every product recommendation is mine, the AI is only allowed to recommend tools I personally use.
- What tools does the site run on?
- Claude (scheduled cloud routines plus Claude Code for hands-on sessions), GitHub for the code and content, Vercel for hosting and deploys, and Google Search Console for ranking data. The whole stack costs this site roughly £11 a month.
- Are these income numbers verified?
- They come straight from my personal income tracker and only count payouts that actually landed, not clicks, not pending commissions. They are not independently audited. The methodology section on this page explains exactly what is counted and what is not, and past months are never rewritten.
- Why publish real numbers at all?
- Because almost nobody does. UK blog income reports mostly died out around 2016, and the AI-makes-money space is full of big claims with no receipts. Every honest experiment I could find where an AI ran a website earned nothing and stopped within a week. This page is my attempt at the missing thing: real sterling figures, costs included, failures included, updated monthly.
- Is the income all from AI-written content?
- Mostly. The referral guides that earn the money were written and optimised with AI in the loop from the start, and the autonomous routine now maintains them and writes new ones. Some earlier posts on the site were written by hand. The commission itself comes from referral programmes for products I use: business banking, energy, transfers, hosting, plus sales of my book.
- Could someone copy this setup?
- The stack, easily: a Claude routine, a GitHub repo, Vercel, and referral links. What is hard to copy is the part that makes it work: only recommending products you genuinely use, publishing honestly for months, and accepting that some months go down. Most people who try automated content sites earn nothing, and the ones promising otherwise are usually selling a course.
Page changelog
- 25 July 2026: Page launched with June and July 2026 data.
Updated monthly, a few days after each month closes. The referral codes this site earns from are all listed openly on the referral codes page, and how I verify offers is covered in my editorial standards.