Adding an AI chatbot to your website

Start by writing down the questions

Before choosing a chatbot, spend a week noting down the questions that reach your sales phone and inbox. If they look like "what are your opening hours" or "how long does shipping take", you're in luck: every ready-made tool can answer those. If they look like "which model cuts 12 mm sheet" or "is there an English data sheet for this product", the chatbot has to actually know your products. That list decides which route you take.

When a ready-made tool is enough

Tools such as Jotform, Elfsight or Botpress add a chat window to your site quickly. They read your pages and answer frequent questions. For a company selling a single service, a clinic that works by appointment or a shop with a handful of products, that is often enough, and the monthly fee is low.

The limit is this: these tools know what is written on your site. They only know what sits in your data sheets, dimension tables and price lists if you upload those separately and keep them in order. If your documents are spread across hundreds of files, keeping that order becomes a project in itself. Checking which document an answer came from, and what the bot does when it doesn't know, is also left to you.

The real question: is the answer right

A chatbot is worth what its answers are worth. One that quotes the wrong dimension, recommends a model that doesn't exist or makes up a price costs more than having no chatbot at all. A customer who gets one wrong answer doesn't ask a second time.

The chatbot we build answers only from your documents. Every answer is tied to the document it rests on. When a question has no answer in those documents, it says it doesn't know and passes the question to you. It does not fill the gap with a guess.

We run the same approach where a wrong answer is most expensive: law. Geolex answers real-estate law questions from 7,500+ legal records, 46 statutes and 172 court rulings, citing the relevant article, and it is live. Our copyright advisor works across 2,600 articles and rulings, catches a fabricated case number, and has been used in a real case. The engine that would run on your catalog is the same one.

What a company with a product catalog needs

In a company with a catalog, buyers don't search for a product by the name you gave it. They don't know the model code; they know the thickness they need to cut. The site shows them a list, they get lost in it and pick up the phone. In that situation, this is what the chatbot has to do:

• Know your products. Its knowledge is your catalog, data sheets, dimension tables and frequent questions. When a visitor gives a dimension, it lists the models that fit and explains why each one fits.
• Speak the visitor's language. Turkish, English and any other language you add. For an exporter, no question is lost to a time difference any more.
• Not make things up. If price, stock and delivery time aren't connected to your systems, it doesn't state them. It passes the question to you along with the visitor's contact details. Some of those conversations come straight back as sales leads.
• Show you what gets asked. Which product people ask about, which questions go unanswered, which size is requested but missing from your catalog. Customers won't tell you what your catalog lacks; this log will.

Can it run on WhatsApp too

Yes. The same knowledge can run in a window in the corner of your site, on its own page and on your WhatsApp line. Update one data sheet and all three update. On the WhatsApp side, your business line has to be connected to Meta's WhatsApp Business Platform, which goes through Meta's own approval steps.

If you build it with us

We run the build for you. What we need from you is your documents and the questions your customers really ask. Throughout, you can count on the following:

1. Gathering the documents is our job. Catalog, data sheets, dimension tables, FAQs, past correspondence if there is any. It's fine if they're scattered; you don't need to tidy them first.
2. No sentence without a source. Every answer records which document it came from. If you question an answer, we open its source together and look.
3. It doesn't go live until it passes the test. The questions your customers really ask, and the answers you give them, become a test list. The chatbot isn't put on your site until it passes that list.
4. We read the first week's conversations together. Gaps get closed in that round. After that, you can see the log of what gets asked yourself.
5. Your data stays yours. Your documents stay on your own server and are never sent to train an outside model. We decide together, during setup, who gets access to what.
6. One place to update. Change a data sheet and the site window and the WhatsApp line both update at once.

What determines the price

What a chatbot costs depends less on how many products you have than on the state of your information. The main things that shape the quote:

• How scattered the information is. One tidy catalog and hundreds of PDF data sheets, spreadsheets and email archives are not the same job.
• Number of languages. Each language means passing the test list in that language too.
• Number of channels. The site alone, or the site plus WhatsApp.
• Live data connections. If price and stock are to be read live from your software, connecting to that software is a separate piece of work.
• Upkeep after launch. How often your catalog changes, and who will do the updates.

That's why, instead of a fixed price list, we work out the scope with you by looking at your documents.

What a chatbot should not do

• It doesn't close the sale. It informs, leads to the right product and brings you in. It doesn't take orders.
• It doesn't quote prices it isn't connected to. Giving no price beats giving a wrong one.
• It can't know what isn't in your catalog. A chatbot is only as good as its knowledge base; a missing data sheet means a missing answer.