I Automated My Business, Then Audited It. The $12.99 Sale Never Happened, and the 71% Growth Was a Bot Farm.

By Andrii Klymenko · Updated September 05, 2026

Six weeks ago I wrote that my automated business had made $54 in 111 days, and that the useful part of the story was not the number but how quietly things had broken. I meant it as a closing thought. Last week I sat down and actually audited the thing, line by line, the way I would audit a system at work, and found out I had been too generous with myself.

The failures I described back then were the ones I had already found. The audit was about the ones still running.

I'm an avionics engineer by training. On an aircraft, the instrument that lies to you is more dangerous than the instrument that dies, because a dead gauge announces itself and a lying gauge lets you fly confidently into the ground. I know this. I say it to people. And then I spent five months steering a business by a dashboard where almost every number was wrong.

Here is what the audit found, in the order that hurt.

The sale that never happened

My store's analytics said one paid order in 2026, $12.99. The same screen said zero products sold. I had noticed that contradiction weeks earlier and filed it under “reporting is weird sometimes,” which is a sentence I would never accept from anyone else.

So I opened the order. The billing email was qa-test@creatifystore.com. It had been created through the REST API. No payment method, no transaction ID, no download permissions issued to anyone. And no line items at all — the order was empty. It was a test I ran myself on the 4th of July and left sitting in “completed” status, where the reporting layer dutifully counted its total as revenue and its contents as nothing, because there were no contents.

The store's entire reported income for the year was a ghost I created and forgot about. Meanwhile the two orders that did have products attached in the reporting table turned out to be abandoned carts — nobody paid for those either. Every dashboard I had glanced at for two months was showing me a precise inverse of reality: revenue where there was none, and nothing where the products were.

The month my best article disappeared

One article of mine outperforms everything else I have written by a factor of fifteen. In August its impressions in Google search dropped to zero and stayed there. I assumed the algorithm had moved on. Things get old, and I had other things to look at.

The audit checked what the server was actually returning for that address, without following redirects, and there it was: a permanent redirect sending every visitor somewhere else. One line, in a small plugin I had written myself, dated the 10th of August. Anyone who found the page in search was bounced off it before a single word loaded, and Google, seeing a permanent redirect, removed the page from its index.

I wrote that line. I have no memory of writing it, which is somehow worse than if I did.

The page is technically clean now — it returns 200, points at itself, sits in the sitemap with a fresh date. Three weeks after the redirect came out, it still has zero impressions. Removing yourself from an index takes one line and one afternoon. Getting back in takes as long as it takes, and nobody tells you how long that is.

The seventy-one percent that was a bot farm

Traffic was up 71% over the previous period. Search impressions were up 120%. I remember feeling good about that for most of a day, and I remember catching myself wondering what I had done right.

Then the audit asked a question I had not asked: up on which pages?

Home, 113 views. My account, 110. Cart, 76. Contact, 57. Refund policy, 38. The best-performing actual article that week: six.

Nobody reads a refund policy thirty-eight times. I pulled the raw server logs and counted 6,381 requests over three days from 193 different addresses, every one of them a Tor exit node, all reporting the same desktop browser, all working through the account and checkout pages in the same order. That was 11.8% of everything the site served that week.

They were driving real headless browsers, executing the JavaScript, which is exactly why my analytics counted them as human beings. It also finally explained a number I had been quietly proud of, that 87% of my traffic was direct. Direct traffic that high is not loyalty. It is software arriving without a referrer.

Their user agent string began with a literal quote character, which no browser on earth sends. One rule in the web server config. The next day: ten such requests, ten refusals.

With the noise gone, the real number surfaced. Views of actual articles and products, before the bots showed up and after they left, run between thirteen and twenty-two a day. Flat. The 71% growth had been a vulnerability scanner looking for a way into my checkout, and I had been reading it as an audience.

Four copies of the same article in four days

In the last week of August my writing engine published the same subject on four consecutive days. Same three competing products, same structure, same questions in a slightly different order, four separate addresses.

There is a duplicate check. It is not lazy code — it compares each new headline against every published headline using fuzzy matching and blocks anything above a threshold. It ran all four times and reported success.

The bug was in the order of operations. The engine picks a topic, writes a working headline, checks that headline for duplicates, generates the article, and only then does a later step rewrite the headline into its final published form. So the check had been comparing a title that never gets published against titles that were. On the second day the working headline was character for character identical to the first day's, and it still went through, because by then day one was live under a different name.

None of the four is indexed. Not one impression between them. Google looked at four nearly identical pages competing for the same search and decided the correct answer was none of the above.

The character that cost thirteen times more

Posting to X costs about a cent and a half. Posting to X with a link in it costs twenty cents. That change came in back in April and I did not know about it.

One of my scheduled jobs had been posting a product link every day as a reply to its own post, specifically to get around the algorithm's dislike of links in the main body. Clever, I thought at the time. Also six dollars a month for one tweet a day, inside a bill I had not opened in a while.

When I finally opened it: $64.20 in thirty days. Most of that was not posting at all but reading — five separate scripts were each independently polling for mentions every fifteen to thirty minutes, and every page of mentions bills per post returned and again per author profile returned.

What did $64.20 buy? Sixteen visits to my site in ninety days. Google sent 200 in the same period. Facebook sent thirty.

The link lives in my profile now, where it costs nothing. The polling went from 187 runs a day to fourteen, with the same posting volume and the same replies, because most of those runs were re-reading the same mentions the previous run had already read.

The actual number

Four places where money could arrive. Etsy: four orders, CA$173.36 across June, July and August, which is about US$126 and represents almost everything I have ever earned. Gumroad: 739 product views in the last thirty days and zero orders, with a lifetime total of ninety-nine cents from eleven downloads of a free item. My own store: nothing. A subscription product with live payment links and three plans: no subscribers.

One hundred and twenty-seven dollars. Five months, four storefronts, 185 published articles, 106 scheduled jobs, and one hundred and twenty-seven dollars.

Running it costs somewhere north of $150 a month.

What I got wrong, precisely

The obvious conclusion is that automation does not work, and that is not what happened. The machine did what I built it to do, every single day, without complaint. The articles are not bad. The publishing never missed.

What I did was automate production and leave measurement to be done by hand, and then not do it by hand.

Every failure above has the same shape. Something broke without raising its voice, a dashboard kept reporting a comfortable number, and the comfortable number was the only thing I ever checked. The redirect did not throw an error. The bots did not look like bots on a traffic graph. The empty order looked like a sale. The duplicate check ran, logged success, and blocked nothing at all.

Automation multiplies whatever is already there. Point it at a funnel that works and you get more customers. Point it at a broken instrument and you get a very efficient, very consistent, entirely convincing lie, delivered to your inbox every morning at eight with a green check mark next to it.

What changed after

I stopped adding things. There are ten products in the store and I have not built an eleventh, which for me is genuinely the hardest part of this.

The ghost order is cancelled, so the store now reports zero revenue, which is both correct and considerably more useful than $12.99. The bots are blocked and my traffic numbers are real for the first time since spring. The duplicate check now runs against the headline that actually gets published. The X spending has a ceiling and an hourly watchdog that reports reading and posting separately with a projection to the month, because a bill you only read once a month is not a bill you control.

And there is one rule I follow now that is worth more than all of the above. A green check mark with no evidence behind it is a lie you are telling yourself. If a script reports that it published something, it now has to fetch the thing back and prove it exists. If a dashboard reports revenue, I open the order and look for line items. If traffic is up, I look at which pages went up, not by how much.

That rule cost me every comfortable number I had. What is left is $127 and a set of instruments I can finally believe, which after five months of flying blind feels like the better half of the trade.

—

---

Written by Andrii Klymenko — Researcher & Writer. Exploring the human side of science. One micro-shift at a time.

☕ Want to support my work or explore more?
→ andriiklymenko.carrd.co
Or tip $1 instantly via PayPal:
paypal.me/AndriiKlymenko400/1

📚 Related Articles

• Premium Automated Boutique: Why Most Fail
• Automated Niche Business Ideas: Passive Income With AI
• Automated SEO Blog Engine: Technical Guide
• Solopreneur AI Business: Build Revenue With No-Code in 2026