How do AI assistants read a website?

AI systems mostly work from the text and machine-readable signals of a page: the title tag, meta description, headings, visible text, structured data (schema.org JSON-LD) and files like llms.txt and robots.txt. Design, images and animations largely disappear — what's left is the raw text and markup.

Why doesn't an AI recommend my business?

Usually not because it rejected you — because it couldn't verify you. If a site never states clearly what the business is, where it operates and what it offers in machine-readable form, an assistant asked for a recommendation has nothing solid to cite, and answers with businesses it can verify instead.

llms.txt is an emerging standard: a plain-text file at the root of a website that gives AI systems a clean, curated summary of what the site and business are. It's cheap to add and almost no small-business site has one yet.

How can I check my own site's AI visibility?

Run a free scan at kovelstudio.com/scan — it checks structured data, metadata, llms.txt, robots.txt and mobile speed, then shows you verbatim what an AI understood from your homepage and what it answers when asked for a recommendation in your category and area.

Findings · July 2026

What AI assistants actually see on Australian business websites.

We built a scanner that reads business websites the way an AI system does — same inputs, same blind spots. Building and testing it taught us more about AI visibility than any article we'd read. Here's what actually happens when an AI reads a small-business site.

1. The design disappears. The text is everything.

Feed a beautiful homepage to an AI system and what it actually processes is the raw text: the title tag, the headings, the paragraphs, and any structured data. The hero video, the animations, the carefully chosen photography — gone. A site can cost $15,000 and read like three vague sentences to a machine.

The most common pattern we see in testing: a salon or trade site whose homepage heading is a slogan — "Love Is In The Hair" is a real example from a Sydney salon we scanned while building the tool. Charming for humans. To a machine, it says nothing about what the business is, where it is, or what it sells. If the title tag is just the brand name too, the AI is left guessing from scraps.

2. Structured data is the difference between "verified" and "vague".

Schema.org JSON-LD is a block of machine-readable facts — business type, location, services, opening hours — that sits invisibly in the page. Large franchise sites almost always have it (their SEO teams put it there). The typical independent small-business site doesn't. That's the visibility gap in one sentence: the machine can quote facts about one business and only guess about the other.

3. Almost nobody has llms.txt yet.

llms.txt is an emerging standard: a plain text file at your site root that hands AI systems a clean summary of who you are and what you offer. It takes an hour to write. In all our scanning of Australian small-business sites while building this tool, we're yet to find one in the wild. Early-mover advantage rarely comes this cheap.

4. The recommendation test is the honest one.

The hardest-hitting check in our scanner asks an AI, cold: "Can you recommend a good [your category] in [your area]?" — then shows the answer verbatim. When a business isn't in that answer, it's usually not rejection. It's absence: the assistant names businesses it can verify, and stays generic about the rest. Machine-readable facts are how you become citable.

5. What actually moves the needle

From most to least impact for a typical small-business site:

A title tag and main heading that state the facts — what you are, where, for whom.
Schema.org JSON-LD for the business, its services and FAQ.
A plain-language summary and FAQ on the page — the facts an assistant needs to describe you.
llms.txt — curated summary for AI crawlers.
Mobile speed — slow pages get crawled less and abandoned more.

None of this is exotic. It's mostly missing because nobody told small-business owners it exists.

Check your own site

The scanner is free and takes about 30 seconds. It runs these exact checks on your site and shows you, verbatim, what an AI understood — and what it answers when asked for a recommendation in your category and area.

Run the free scan