Solo Founder Built 20 Pages for GEO, and Alisa Named His Product in 1 Question Out of 16

The temptation is easy to understand: agents write articles faster than humans, a site comes together in an evening, and neural networks like Alisa and ChatGPT learn from texts on the internet. So if you publish twenty pages about an emergency fund and budgeting until payday, the assistants will start citing the author. The founder of the Kubysh app decided to test this on a small site with no history. The result was sobering.

What Was Done and How It Was Measured

This is the second experiment in the series "A Month Without Hiring." In the first, Threads brought 110,000 views and almost zero installs, so the author looked for a channel where a person arrives with a specific question rather than scrolling a feed. Search seemed like the ideal option.

In two weeks the site grew to 20+ pages. Topics were taken from real questions: how to calculate a daily budget, how much to keep in an emergency fund, what to do if money runs out before payday. Expectations were modest: at least dozens of click-throughs. A month later, one arrived.

The most useful part of the case is the measurement method. There are almost no ready-made counters for GEO, that is, optimization for neural network answers, so the author built his own:

• he compiled a list of 16 questions a real user might ask: "how to build an emergency fund," "how to survive until payday," "daily budget app";
• once a month he asks them to the assistant in the same words, in a new conversation;
• he notes whether the product is named and whether there is a link to the site;
• he calculates the share.

Right now it is 1 out of 16, about 6%. The questions are phrased around the task, not the brand, otherwise the metric misleads: any assistant will find a product by its name. This share is more honest than search rankings, which are almost empty for a young site.

Why 23 Pages Became 9, and AI Did Not Speed Up the Result

Next the author did what courses like "GEO in 7 Days" do not mention: he cut 23 thin pages down to 9 strong ones. A thin page is one where there is text but no answer to the question. It is generic words that AI writes very confidently. A strong page answers one question specifically, with figures and an example. Not "it's important to save a little at a time," but "with such-and-such a salary and such-and-such mandatory expenses, a three-month emergency fund comes to this many rubles, and here is how to build it in six months."

There is a pleasant side effect: the author likes nine pages more than twenty-three. They are not embarrassing to show to a person.

So why didn't it work? The author's hypothesis, and he himself notes he is not sure it is universal: search engines and assistants evaluate not only the text but also trust in the site. That means domain age, links, mentions, and user behavior. A one-month-old site has no source for any of this, and no agent can write a history for it. AI has made content production cheaper, but not trust.

What to Do in Practice

• Do not count on SEO and GEO as a growth channel in the first months of a new site. It is an investment that pays off later. The author says spring.
• Set up your own measurement: 10–20 user questions, fixed wording, a new conversation, a check once a month.
• Cut thin pages. One page with a specific answer beats three with generic words.
• If you need traffic now, look for other channels and run search in parallel.

The limitations are worth remembering. This is one site, one assistant, and one month. A sample of 16 questions is small, neural network answers are unstable, and the difference between 1 and 3 out of 16 may well be noise. But as a starting point for honest measurement the method works, and the main lesson is simple: cheap content does not replace accumulated trust.