Zum Inhalt springen
← All posts

AI Chatbot for Your Website: What It Can Do — and What It Cannot

A RAG chatbot on your own data instead of „just a GPT widget": real use cases, GDPR requirements, realistic costs, and a checklist for whether your content is ready.

An ai chatbot for website project rarely fails on the technology – it usually fails on wrong expectations. Some people believe a widget will answer every customer question from tomorrow; others dismiss the whole thing as a toy bot that only talks nonsense. Neither is true. After several chatbot integrations for companies and shops in Vienna and across the DACH region, I can tell you where a modern AI chatbot delivers real value, where its limits are, what GDPR-compliant actually means, and what costs you should realistically expect. No hype, with concrete numbers.

RAG chatbot vs. „just a GPT widget”

The most important distinction first, because it decides everything else. A plain GPT widget hangs a general language model on your site. It sounds competent but knows nothing about your business – not your prices, not your opening hours, not your return policy. When a customer asks, the model guesses. And a bot that confidently states falsehoods is worse than no bot at all.

A RAG chatbot works differently. RAG stands for „retrieval-augmented generation”: before the model answers, it searches your own content – website texts, FAQ, product data, manuals – and phrases the answer from exactly those sources. The result is a bot that knows your company, answers with your real information, and ideally even links to the right page. The practical difference: the GPT widget is installed in an hour and deleted in the second week because it confuses customers. The RAG bot needs preparation, but delivers answers you can stand behind. How such systems fit sensibly into a business, I cover in my practical guide to using AI in your business.

Simplified, a RAG bot answers in three steps: first, your content is prepared once and made searchable. Second, for every user question the system pulls out the relevant passages. Third, the language model phrases an answer from exactly those passages – and only those. The decisive effect: the model no longer invents freely but works on the short leash of your facts. That is precisely what separates a bot you can trust from one that guesses at the first price question.

What an AI chatbot is genuinely good for

A chatbot replaces neither good sales nor real consulting. But there are three jobs where it reliably pays off.

FAQ relief: catching the same questions over and over

In almost every business, the same ten to twenty questions make up the bulk of enquiries: delivery times, returns, opening hours, prices, compatibility. A well-fed bot answers these instantly and around the clock – and your team gains time for the cases that genuinely need a human. This is the use case with the fastest, most clearly measurable value.

The often-overlooked bonus: the bot shows you in black and white what your customers actually ask. After a few weeks the conversation logs reveal the patterns – which information is missing from the website, which step in the ordering process confuses people, which product keeps triggering the same query. That feedback is worth real money regardless of whether the bot answers the question itself: it tells you exactly where your website needs to improve.

Lead qualification: pre-sorting the right enquiries

Instead of a form everyone fills in the same way, a chatbot can clarify the decisive points in conversation: which project, which budget, which timeframe? What reaches you is not a bare email address but a pre-qualified enquiry – and uninterested visitors tie up no time. For shops this can extend to product advice; how that looks in practice is on my AI for online shops page.

24/7 availability without a night shift

Most enquiries come in the evening and at the weekend – exactly when nobody is at their desk. A chatbot catches those moments: it answers what it can, and for the rest cleanly captures the contact details so nobody gets lost by Monday. Not a replaced employee, but a door that never fully closes.

The value is also measurable, and that is exactly what you should demand before starting such a project. Sensible metrics are the number of conversations fully resolved by the bot, the share of cases cleanly handed to a human, and the number of contacts captured outside office hours. A bot whose effect does not show up in numbers is a gut-feeling project – and those rarely end well.

Where the limits are

Just as important as the strengths are the honest limits – ignore them and you build in disappointment:

  • No real consulting on complex cases. As soon as it involves individual quotes, negotiation or delicate customer situations, a human should take the wheel. The bot should recognise this and hand over cleanly, not keep talking artificially.
  • Only as good as its data. What is not in your content, the bot cannot know. Outdated or contradictory information produces outdated or contradictory answers – garbage in, garbage out.
  • No fix for bad processes. If your delivery times are unclear, the chatbot does not make them clearer – it just exposes the problem faster.
  • Residual risk of wrong answers. Even a RAG bot can miss. That is why clear boundaries, a visible note that this is an AI, and an easy path to a human all belong in the setup.

GDPR: where it gets serious in the EU

For operators in the DACH region, this is where the project becomes viable or not. A chatbot processes user input – often personal data. These four points must be right:

  • Where does the data sit? If the widget sends every input uncontrolled to a US server, that is a data-protection problem. Prefer providers with EU hosting, or models run inside the EU.
  • Consent (opt-in). The chatbot must not record secretly. Users need to know they are talking to an AI and be able to consent to processing – cleanly integrated into your cookie and consent setup.
  • Data minimisation. Collect only what you need. No chat history needs to be stored forever; define deletion periods.
  • Data processing agreement. If you use an external AI service, you need a data processing agreement (DPA) with the provider. Without it, the legal basis is missing.

That sounds more involved than it is – with the right architecture, GDPR compliance is built in from the start, not bolted on afterwards. How I set up such integrations in a privacy-friendly way is on my AI integration page.

What do setup and running cost?

Realistic guidance instead of blanket promises – the figures refer to the DACH region and a cleanly implemented, GDPR-compliant project.

Variant One-off setup Running / month Suitable for
Simple FAQ bot low four figures approx. 30–80 € Small site, clearly defined question set
RAG bot on your own data mid four figures approx. 80–250 € SME with lots of content, lead qualification
Shop integration with product advice upper four figures+ from approx. 200 € Online shop, product data, stock connection

The running item has two parts: the language model usage cost (depending on conversation volume) and the upkeep – because your content changes and the bot must grow with it. A chatbot is not „install once, never touch again”, but a system that wants maintenance. That is exactly why I warn against the ten-euro widgets: you save on setup and pay in wrong answers that cost customers.

Checklist: is your content ready?

A RAG bot is only as good as what you feed it. Before the project, check:

  1. Are your core questions documented? A maintained FAQ with the 20 most common questions is half the battle.
  2. Is the information current and free of contradictions? If two pages show different prices, the bot will find both.
  3. Is product and service data structured? Clear text is searched more reliably than a wall of PDFs.
  4. Is there a defined handover path? Decide when the bot hands over to a human – and to whom.
  5. Is the legal frame sorted? Consent, privacy policy and DPA belong before go-live, not after.

Conclusion

An AI chatbot is neither a cure-all nor a toy. As a RAG system on your own data, it reliably relieves recurring questions, pre-qualifies enquiries and stays available around the clock – provided your content is maintained and the GDPR basis is right. The plain GPT widget, by contrast, confuses more than it helps. What matters is not whether you add a chatbot, but whether you feed it real information and honestly draw the line on what it can do. That is when a buzzword turns into a tool that measurably takes work off your team.

Wondering whether an AI chatbot is worth it for your website? Send me a quick message via the contact form. I will look at your content and tell you honestly whether – and in what form – a bot pays off for you.

Häufige Fragen

What is the difference between a RAG chatbot and a GPT widget?

A GPT widget hangs a general language model on your site – it does not know your business and guesses on company-specific questions. A RAG chatbot searches your own content (website, FAQ, product data) before every answer and phrases the reply from those sources. The result: a bot that delivers your real information instead of confidently stating falsehoods.

Is an AI chatbot GDPR-compliant?

It can be, if four points are right: the data should be processed in the EU (EU hosting), users must know they are talking to an AI and consent to processing (opt-in), data minimisation applies with defined deletion periods, and you need a data processing agreement (DPA) with the AI provider. With the right architecture this is built in from the start.

What does an AI chatbot for a website cost?

A simple FAQ bot starts in the low four figures one-off plus around 30–80 euros a month. A RAG bot on your own data is a mid four-figure one-off and 80–250 euros a month. A shop integration with product advice is correspondingly higher. The running item covers language model usage and upkeep, because the bot must grow with your content.

Will a chatbot replace my customer service staff?

No. A chatbot catches the same standard questions and is available around the clock, but for complex cases, individual quotes or delicate situations a human should take the wheel. The best setup lets the bot recognise when to hand over to a staff member – it relieves your team but does not replace it.

Free checklist: 10 points before your website goes live

Practical tips from real projects — straight to your inbox, no spam. Unsubscribe any time.

    Alex
    Alex · Buntweb

    Web developer and IT service provider from Vienna. For over ten years I have been building and maintaining websites and online shops — focused on clean technology, honest advice and solutions that work in everyday business.

    Ask a question