AI and the sky

How AI Agents Read a Website

An AI agent is a program built around a language model that can search, open and read web pages, then summarise what it found. It does not see a site the way a person does; it reads the text and structure. Clear headings, short direct answers, structured data and cited sources help it understand a page correctly.

When you ask an AI assistant about the night sky, it sometimes answers purely from what it learned during training. At other times it searches the web and reads specific pages. This article explains how the second case works, and what you can and cannot expect from it.

What is an AI agent?

A language model is a program that predicts text based on a huge number of examples. On its own, it does not open websites. An AI agent is a system built around such a model and given tools: run a search, load a page, read a file, compare several sources and write an answer. It works in steps: set a goal, act, look at the result, decide what comes next.

It helps to know the limits, too. An agent can misread a page, miss an important section or trust a weak source. That is why checking facts matters (see AI hallucinations and What AI can and cannot do).

Crawling, structured data and clear text

Crawling

Crawling means automatically opening pages and following the links between them. Search engines have done this for a long time to build an index. Some AI systems use such a ready-made index; others load a page at the moment you ask. A site can give guidance to bots through a robots.txt file, where the owner describes which parts are open to automated reading. Following it is an agreed convention on the web, not a technical barrier.

One practical detail: not every automated reader runs JavaScript, the scripting language of web pages. Main content is safest when it is present in the page’s HTML itself rather than appearing only after scripts run.

Structured data

Structured data is an extra layer that tells a machine what a page is: an article, a question with an answer, an event, an organisation. The most widely used shared vocabulary is schema.org, often written in JSON-LD (JSON for Linked Data). That way the title, author, date and questions do not have to be guessed from the page layout.

Structured data helps with understanding, but it does not guarantee that a page will be shown or cited. Search engines publish their own guidelines on which types they support and how.

The llms.txt file

llms.txt is a proposal for a simple text file in a site’s root that briefly introduces the content and points to the most important pages. It is not an established standard, and there is no general rule that all AI systems read it. We treat it as a possible convenience, not a visibility switch. More in llms.txt and structured data.

Clear writing

The strongest factor remains the plainest one: well-written text. For both people and machines, these things help:

• a title that says what the page is about;
• a direct answer in the first sentences;
• subheadings that divide the topic logically;
• terms defined at first use;
• dates and clearly separated claims.

Why sources and distinctions matter

An agent often gathers snippets from many places and merges them into one answer. If a snippet does not make clear what is scientific fact and what is cultural tradition, mixing them up is easy. In a subject like the sky this matters a great deal (see The difference between astronomy and astrology).

Astronomy is a science, and its claims can be tied to sources such as NASA, ESA or the International Astronomical Union (IAU). Astrology is a tradition and is described as one: “in the astrological tradition…”. When a text draws this line explicitly, both readers and agents are less likely to present an interpretation as established fact. If you are curious how such systems are used for horoscopes, read How an AI horoscope works.

Sources have a second job: they let the reader verify. A good assistant answer points to where the information came from, and you can open that page and read it yourself.

How Astronativa arranges its encyclopedia for people and machines

The encyclopedia is built so that one page works for a human reader and an automated one alike:

• A short answer first. Each article opens with two or three sentences that answer the main question directly.
• Clear structure. Subheadings, short paragraphs and explained terms.
• Separation by section. Astronomical facts and astrological readings are kept apart and labelled, so they do not blur together.
• Cited sources. Articles link to scientific and encyclopedic pages that anyone can check.
• Internal links. Related topics are connected, so a reader, human or machine, can follow the logic of a subject.
• Two languages. Bulgarian and English are separate, natural texts, not mechanical translations.

None of this guarantees that a particular assistant will cite Astronativa. It simply lowers the chance of misunderstanding and makes checking easier.

Reading AI answers sensibly

If an assistant tells you something about the sky, ask where the information comes from, open the cited source and notice whether the answer separates fact from tradition. That way you use AI as a helper for orientation, not as the final authority.

Positions in Astronativa are astronomically exact. The interpretation is for inspiration. · Glossary · Terms · Privacy