To train an AI agent on your website, the platform reads your pages, breaks the text into searchable pieces, and stores them so the agent can find the right passage when a customer asks a question. It then writes the answer from that passage instead of relying on guesswork.
Why train an AI agent on your website?
Your website already holds most of what customers ask about: services, prices, policies, opening hours, and contact details. When you train an AI agent on your website, you give it exactly the information you want customers to see, in your own words, so the answers match what you publish.
How an AI agent learns from your website, step by step
Step 1: Crawl your pages
You enter your website URL. The platform visits your working pages and collects their text. OrhanAI starts this in the background as soon as you create the draft, so you can see progress while it runs.
For more information about how website crawlers work, see Google's documentation on robots.txt.
Step 2: Clean and split the content
Menus, footers, and repeated elements add noise. The text is cleaned and divided into smaller passages, so each piece covers one idea, such as a return policy or a service description.
Step 3: Turn passages into embeddings
Each passage is converted into a numeric representation of its meaning, called an embedding. This allows the agent to match a customer's question to the right passage by meaning, not only by shared words. A customer who asks about getting their money back can still reach your refund policy.
Learn more about how text embeddings work.
Step 4: Retrieve the right passage at question time
When a visitor asks something, the system searches the stored passages and picks the most relevant ones.
This process is also explained in Google's guide to retrieval-augmented generation (RAG).
Step 5: Write an answer from approved sources
The agent composes a reply using those passages. This is the idea behind source-grounded AI: the answer is tied to your approved content, and source references stay available for review. If no relevant passage exists, the agent uses fallback wording instead of inventing an answer.
For a technical overview, see OpenAI's guide to retrieval and semantic search.
What you should do after the first training
1. Review the draft. Confirm what the agent understood about your business.
2. Test real questions. Use the Playground to try the questions customers actually ask, including awkward ones.
3. Check the sources. Make sure the answers come from the right pages.
4. Fill the gaps. Add missing information to your website or as an extra training source.
5. Publish. Add the embed script to your approved domain.
How to make your website easier for an AI agent to learn from
Write clear pages with one topic each, for example one page per service.
Publish an FAQ page that answers common questions in plain sentences.
Keep prices, hours, and policies up to date, and state them in text rather than only inside images.
Put your contact details, location, and booking information where they are easy to find.
Use descriptive headings, because they help the system split and match content correctly.
What an AI agent cannot learn from your website
A website-trained agent only knows what is on the pages it can read. It will not learn from content behind a login, from information that exists only in your team's heads, or from pages that block crawlers. Add those as separate training sources, or publish the information on your site.
Keeping your AI agent up to date
Websites change. When you update prices, add services, or revise policies, run the crawl again so the agent learns the new content. Review the list of unanswered questions regularly, because each one points to something your content does not explain yet.
Frequently asked questions
How long does it take to train an AI agent on a website?
Orhan AI builds a first draft in minutes. The larger the site, the longer the crawl takes, and you should leave time to review and test the answers before you publish.
Do I need to write training data by hand?
No. A no-code AI agent learns from the pages you already have, and you can add extra sources when needed.
Will the AI agent make up answers that are not on my site?
A well-built agent answers from approved sources and uses a controlled fallback when it lacks the information. Testing before launch confirms this for your own content.
What happens when I change my website?
Re-crawl the site so the agent picks up the changes.
Article by

