Why neural networks answer about a product in general terms: analysis and checklist
Analysis of a Habr article from October 29, 2026: how neural networks read a website, why text from a browser may not reach the robot, and what to check in page code.
On October 29, 2026, Habr published an article about why neural networks describe a product in general terms, although the website describes it in detail. A typical picture: the description takes three screens, while ChatGPT, GigaChat, Alisa AI and Perplexity answer in general terms or confuse the product with a competitor. The reason is that the model does not read the whole site, but several found fragments.
What a neural network sees instead of the whole site
A neural network (a model that assembles an answer from found pieces of text) does not work like a reader. It receives several fragments found by search or a robot. If these fragments do not contain direct text about what this product is and for whom it is, details on other pages will not help. This is the main reason for general answers.
Example: the website has three screens about capabilities, integrations and scenarios. But the found fragment includes the menu, general promises and reviews. The model answers in general terms because it does not have a short definition of the product. It is not obliged to read the site to the end.
Therefore, you should look not at the completeness of the description, but at which pieces of the site can actually reach the model. If they do not contain the answer to what this is and for whom, the rest of the text does not save it.
Why part of the text may not reach AI
Vercel and MERJ found out: robots of ChatGPT and Claude download page scripts but do not execute them. This means that text that the site draws in the browser does not reach them. This is client-side rendering (when content is assembled by a script already in the browser).
Example: the product description is loaded after the page opens. A person sees it. The robot may see an empty block. This does not mean the site is bad. It means the robot does not execute the script and receives not what the visitor sees.
Therefore, the exact text about the product must be in the page code, and not only appear on the screen after loading. If the product definition exists only in a dynamic block, it may not reach the model.
ChatGPT and Alisa AI have different routes to the site
ChatGPT has its own robots. According to OpenAI help, search is handled by OAI-SearchBot (the robot through which sites get into search results). GPTBot (the robot for collecting training data) is not connected with search. The settings in robots.txt (a file with rules for robots) are independent for them. There is also ChatGPT-User: it opens a page when the user has asked a question, but it does not crawl the web itself and does not affect whether the site gets into search.
Alisa AI does not crawl sites itself. It splits the question into subquestions, sends them to Yandex Search and uses it. For a site, this means that its representation in Yandex Search is more important than an attempt to please Alisa directly.
For GigaChat and Perplexity, the original source has no analysis of the mechanics — there is no data. Therefore, one should not transfer conclusions about OAI-SearchBot or Yandex Search to them without verification.
Where the exact text should be located
You need to look not for a beautiful description, but for fragments that can actually reach the model. They must contain direct text: what this product is, for whom it is and how it differs. And this text must be in the page code.
Example: if the product definition exists only on the last tab or in a block that appears after a click, the robot may not receive it. If a short definition is in the HTML next to the name, the model has a chance to quote it instead of inventing something general.
You should check what the robot sees, not what a person sees in the browser. These are different things, and the discrepancy is exactly what produces general answers.
What to do about this
- Write a short paragraph in the page code: what this product is, for whom it is and how it differs.
- Check that this paragraph is in the HTML, and is not only drawn by a script after loading.
- Look at which fragments of the site are found by search and robots. If they do not contain a product definition, rewrite these fragments.
- Separate the robot settings in robots.txt: OAI-SearchBot, GPTBot and ChatGPT-User solve different tasks, their rules are independent.
- For Alisa AI, check the site's representation in Yandex Search: it relies on it and does not crawl the site itself.