I asked ChatGPT which AI assistant a small Russian-speaking business should use. It replied: "I can be your AI assistant for a small business." Then came a list of what it would help with: sales scripts, content plans, unit costs. No other service was mentioned.

Claude answered almost the same way: "I'm ready to help you as an assistant for a small business." Gemini named Komanda.ai. The model sitting on top of Yandex search results listed seven services: DOWEDO, Komanda.ai, Aximum, DAAP, SberBusiness's Business Assistant, Kaiten AI and Solo.

We were not in any of those answers. That is expected, and I would not write an article about a single question. But we ask 37 of them, and we ask them regularly.

## How we measure this

Since June 16 we have been running a monitor, 17 runs so far. A script sends 37 questions that business owners typically ask to five systems: ChatGPT, Claude, Gemini, Perplexity and a "Yandex search results plus YandexGPT" pipeline. The last one works roughly like Yandex's Neuro, but it is our own build. We do not query Alice directly.

The questions vary. Some are general ("what tasks in a company can you hand to an AI"), some are about specific systems ("AI for MoySklad", "an AI agent for Bitrix24"), some are about marketplaces and the laws that apply to sellers.

Each question goes to each system up to five times: answers drift, and one answer tells you nothing. For systems that search the web we record the source links and the first 600 characters of the answer. We do not store the full text, so a service named at the end of a long answer may have been cut off.

The September 21 run produced 810 answers. All numbers below come from it.

## What people mean by "a ChatGPT alternative"

In Yandex Wordstat, "chatgpt alternatives" gets 941 searches a month and "chat gpt alternative" (a different spelling in Russian) another 540. The long tail is mostly consumer: "free", "Chinese", "uncensored". Business questions are rarer. "chatgpt for business" gets 214 searches a month, "chatgpt for business with closing documents" gets 13.

Thirteen searches look like nothing. But that is the only phrasing that clearly has a company's accounting department behind it.

In Yandex results for "ChatGPT alternative", the first page is all listicles: T—Zh, vc.ru, thecode.media, Habr, Skillbox, pikabu. No product pages in the top ten. Someone looking for a ChatGPT replacement lands in a list, and the author of that list makes the choice for them.

## Where the models get their answers

A model that searches the web shows its sources. We counted which sites appear in answers most often.

| Site | Answers out of 810 that link to it |
|---|---|
| vc.ru | 258 |
| sostav.ru | 129 |
| habr.com | 111 |
| samreshuuu.ru | 97 |
| moysklad.ru | 90 |
| klerk.ru | 78 |

vc.ru leads by a wide margin. The advice "use such-and-such service" often turns out to be a retelling of someone's article on vc.

Our site is fourth on this list, and at first I was pleased. Then I looked at which answers it appears in.

## Where we are cited and where we are not

We grouped the questions by topic and looked at the share of answers that link to our site.

| Question topic | Share of answers linking to us |
|---|---|
| AI for MoySklad | 50% |
| Selling on Ozon | 44% |
| Laws for marketplace sellers | 30% |
| Replying to reviews | 19% |
| amoCRM | 2% |
| "Which AI assistant should I choose" | 1% |
| Bitrix24, sales analytics, taxes, tenders | 0% |

The models take our reference texts: how to connect MoySklad, what a seller should check, which laws apply. The product is named in 2% of answers. In questions where a person is actually choosing a tool, we are absent.

We are read as a reference. Others get recommended.

Over September the share of answers linking to us moved from 11% to 13%. That is within the noise.

## Why ChatGPT recommends no one

In this run ChatGPT gave 185 answers and never linked to our site. Perplexity linked to us in 36 answers out of 95, the Yandex pipeline in 50 out of 185, Claude in 11 out of 160, Gemini in 9 out of 185.

I do not have a proven explanation, only an observation. When a question sounds like a request ("I need an assistant for my business"), ChatGPT and Claude answer as an assistant and offer themselves. When it sounds like a search ("how to connect AI to 1C and a CRM without a developer"), both search and recommend a specific service, in this run Albato. The Yandex pipeline named NextBot and 1C:Naparnik for the same 1C question.

## What this means when you choose

If you pick a tool based on a model's answer, you pick based on who has been written about more on vc.ru and Habr. That is not a bad filter, it just measures fame. Whether the service fits your company is not visible from it.

I would ask any candidate a few boring questions:

- can you pay by invoice as a company and get closing documents;
- where is the data processed, and is it used to train models;
- can the service log into your 1C, CRM or marketplace account itself, or will you be exporting files by hand;
- what do you get at the end: text you still have to format, or a finished file.

We have a separate [checklist on data and Russia's 152-FZ](/resources/kak-vybrat-ii-agenta-bezopasnost-dannyh-152-fz), and a piece on [choosing an AI agent for 1C](/resources/ii-agent-dlya-1s).

We build this kind of tool ourselves, [Samreshuuu](/compare/chatgpt), which is why we track these answers.

## Sources

- Samreshuuu AI answer monitor: 37 questions, 5 systems, up to 5 repeats per question, run of 21.09.2026, 810 answers.
- Yandex Wordstat, Russia, data as of 23.09.2026.
- Yandex search results for Russia, snapshot of 23.09.2026.
- Our articles: [AI agent for 1C](/resources/ii-agent-dlya-1s), [152-FZ checklist](/resources/kak-vybrat-ii-agenta-bezopasnost-dannyh-152-fz), [AI hallucinations](/resources/gallyucinacii-ii-pochemu-neyroset-otvechaet-po-raznomu)