getanekI built a bot that audits marketing while I sleep
What the audit machine reads, how it scores, and the leaks it finds every single time
Most nights a machine reads a company’s ads, website and competitors, scores them, and writes a report. In the morning I read it with coffee. I did not write a word of it.
This is the first system I am taking apart in public. It is the one I built first, because it answers the question I got asked most in five years of Dubai real estate: “Why is my marketing not working?”
The honest answer was always the same. Nobody was looking. Not at the ads, not at the page the ads land on, not at what the competitor down the road was running. Looking takes hours. So I built something that looks for me.
What it reads
The machine starts with two inputs: a website URL and, if there is one, a Meta page. From there it collects on its own.
- The website. It crawls up to 500 pages. Titles, headings, load speed, broken links, missing schema, mobile layout, what the page asks the visitor to do.
- Live ads. It pulls the company’s current ads from the Meta Ad Library. Creative, copy, how long each one has been running, how many variants exist.
- Competitors. It pulls the same for the competitors. An ad that has been running for 90 days with six variants is an ad that is making money. That is the only signal you need.
- Search. What the site ranks for, what it should rank for, and whether AI search engines can even read it.
No logins. No access to anyone’s ad account. Everything it reads is public. That matters, because it means it can run on any business before a single conversation happens.
How it scores
Collecting is the easy part. The value is in the judgment.
The machine runs about 50 checks. They are split across Meta ads, Google ads, website conversion, follow-up speed and AI visibility. Each check has a weight. The weights are not academic. They come from seven years of watching the same leaks drain the same budgets.
Some of the checks:
- Is there a phone number or WhatsApp button above the fold on the page the ads land on?
- Does the landing page ask for one action, or five?
- Has any ad been refreshed in the last 30 days, or is the account running on fumes?
- Is the site tracking conversions at all, or is the ad platform guessing?
- Does the site have structured data so Google and AI assistants know what the business is?
- How many of the competitor’s winning ads have a format this business has never tried?
Every failed check gets an estimated cost. Not a precise number, a ranked one. The report sorts by money leaked, biggest first. The top of the report is always the thing to fix today.
What it finds
I have run this on brokerages, developers, a supply company, a beauty studio, a snack brand. Different industries, same report.
One recent run: a company spending a serious monthly budget on Meta ads. Good creative. Decent targeting. The landing page had no phone number above the fold and a contact form with nine fields. Every lead that came in was answered the next business day. The ads were not the problem. The ads were fine. Everything after the click was broken.
Another: a site with 40 pages and not one of them told Google what city the business was in. Ranking for nothing, spending on ads to make up for it.
Another: three competitors running the same video format for four months straight. The business I audited had never run a video ad. The market had already tested it for them. They just were not watching.
This is the pattern. The money is rarely lost in the ad. It is lost in the page, the form, the follow-up, and the things nobody checked because checking was boring.
Why I built it instead of doing it by hand
I am dyslexic. Reading 50 pages of a website and cross-referencing them against an ad library is exactly the kind of work my brain refuses to do. So I described what a good audit looks like to an AI, step by step, and had it build the machine with me.
That is the whole method, and it is the method I will keep showing here. I talk. The machine types. The output is a system that does the tedious part forever.
The stack, for the people who want it: Claude for the reasoning, Playwright for the crawling, the Meta Ad Library for the competitor data, and a set of skill files that encode what to check and why. The report comes out as markdown, then as a slide deck. The whole run costs less than a coffee.
What it is not
It is not a free audit offer. I work with a brokerage in Dubai and I am not going to audit its neighbours. The machine is a showcase of what a system can do when it runs without you, and it is one of several I run.
Next ones I will take apart here: the WhatsApp outreach engine that writes and sends personalized messages from a business’s own number, the warehouse of 1.6 million Dubai property listings sitting in a database in my living room, and the 8am watchdog that reads every ad account and flags what is bleeding before I am awake.
If you want the full breakdown of how the audit machine is wired, with the check list, it is on the system page. Name and email, then it opens.
← All posts