Short answerWhen a customer asks an AI assistant to find, compare or contact businesses, the assistant reads your website in their place. It understands facts stated as text on clear pages, structured data that matches the page, consistent details across the web and forms with plain labels. It struggles with facts hidden in images, PDFs or pop-ups, and some page settings shut it out: on its page about AI features Google names nosnippet and noindex as the controls that limit what Search shows, and Bing says noarchive keeps a page out of Copilot answers. A website that assistants understand gets described correctly, and one they cannot read gets skipped or misquoted.
A facilities manager in Business Bay asks an assistant to find three cleaning companies that cover offices in DIFC, work weekends and send a quotation within a day. The assistant searches, opens a dozen websites, reads what each one says and comes back with three names, a line on each and the next step. The manager never sees the other nine websites. They were opened, read and dropped.
In that search, the first visitor to every one of those websites was software working for a customer. It did not look at the design. It looked for facts, and it moved on when it could not find them.
What an assistant reads well
- Facts written as text. Services, areas, prices, hours, licences and contact details in sentences and lists on the page.
- One clear page per question. A page for each main service or area, opening with a short answer, is easier to quote than one long page about everything.
- Structured data that matches the page. Google says the markup must describe what visitors can see, and Bing says structured data can support clearer grounding of its answers, without a guarantee.
- The same details everywhere. When your website, Google profile and listings agree, an assistant can state your facts. When they differ, it hedges.
- Forms with plain labels. Assistants that act for a user fill in forms. “Number of offices” and “Preferred start date” are clear; three unlabelled boxes are not.
What an assistant reads badly
- Text inside images, such as a price list designed as a graphic or a menu photographed on a table.
- Facts only in PDFs or downloads, such as a company profile that is the only place your services are listed.
- Content behind pop-ups, sliders or tabs that need clicks before the facts appear.
- Vague copy. “Solutions for every need” gives an assistant nothing to repeat.
The settings that quietly shut assistants out
Some page and server settings were added years ago for other reasons and now decide whether AI answers can use your pages. Check them before anything else.
- robots.txt rules. OpenAI says OAI-SearchBot must be allowed for ChatGPT search, and blocking Bingbot keeps you out of Bing and Copilot.
- noarchive and nocache. Microsoft says noarchive prevents a page from being used in Copilot answers, and nocache limits Copilot to the address, title and snippet.
- nosnippet and max-snippet. On its page about AI features, Google names these as the controls that limit what Search shows from your pages.
- noindex. A page that is not indexed cannot be a source for Google's AI answers, Bing or Copilot.
- Bot protection at your host. OpenAI says the host or content delivery network must let its crawler through, and a strict firewall may not.
About llms.txt
llms.txt is a proposal from 2024 for a plain-text summary file that points language models to a site's key pages. Google's Lighthouse calls it an emerging convention and treats it as optional. We found no statement from Google, OpenAI or Microsoft that their search or AI answers use it. We publish one on this site because it costs little, and we do not count on it.
Why this matters more in Dubai
Customers here search in English and Arabic, often on behalf of a company, and compare several providers before they send one message. An assistant that can read your services, areas, licence and prices in either language can put you on the short list. One that finds a slogan and a contact form cannot. Our guide on why ChatGPT may not recommend your business lists the common reasons in detail, and a website knowledge catalog keeps the facts identical wherever they appear.
Check your website this week
- Read your robots.txt and make sure Googlebot, Bingbot and OAI-SearchBot are allowed.
- Search your page code for noarchive, nocache and nosnippet, and remove them where you did not mean them.
- Find one fact that exists only in an image or PDF and put it on a page as text.
- Make the first sentence of each service page say what you do, for whom and where.
- Give every form field a plain label.
Making a website easy for assistants to read and use is what an agent-ready website from MM² starts with.
Questions
What does it mean for a website to be readable by AI?
That an AI assistant can find and understand your business facts: services, areas, prices, hours and contact details as text on clear pages, structured data that matches the page, consistent details across the web and forms with plain labels.
Can page settings stop AI assistants from using my website?
Yes. On its page about AI features, Google names nosnippet, max-snippet and noindex as the controls that limit what Search shows from your pages. Microsoft says noarchive keeps a page out of Copilot answers and nocache limits Copilot to the address, title and snippet. robots.txt rules and host firewalls can block crawlers entirely.
Do I need an llms.txt file?
It is optional. llms.txt is a proposal, and we found no official statement from Google, OpenAI or Microsoft that their search or AI answers use it. It is cheap to add, but the facts on your pages matter far more.
Is a website readable by AI also better for customers?
Usually yes. Clear facts, one page per question, plain labels and content that does not hide behind pop-ups help people on a phone as much as they help an assistant.
Will an AI-readable website get my business recommended?
No one can promise a recommendation. A readable website removes the reasons an assistant has to skip or misquote you, and you can test the effect by asking assistants the questions your customers ask.