Every webshop hits the same wall. The same handful of questions arrive every day — where is my order, does this run small, can I still return it after three weeks — and answering them by hand eats the hours you meant to spend on the shop itself. An AI chatbot for your website is the obvious fix. It is also the easiest thing to buy badly.
This is a buyer's guide rather than a pitch. What follows is what actually matters when you compare chatbot software for your website, what the marketing pages leave out, and the questions worth asking before you paste anything into your theme.
What a website chatbot actually does
A website chatbot sits in the corner of your pages as a chat widget. A visitor opens it, types a question in plain language, and gets an answer without waiting for anyone. That is the whole promise, and on the surface every product delivers it.
The difference is where the answer comes from. Older chatbots run on rules you build by hand: if the message contains the word "return", show the returns paragraph. Modern AI chatbots read the question, search your own content for something relevant, and write a reply based on what they find. The first kind breaks the moment a customer phrases things unexpectedly. The second kind copes with phrasing it has never seen, because it is not matching keywords — it is using an AI model to understand what was asked.
That flexibility cuts both ways. A bot that can write any sentence can also write a wrong one. Which is why the most useful thing to look for is not how clever the AI sounds in a demo, but what it does when it has nothing to go on.
Scripted bots, AI chatbots and AI agents
Three terms get used interchangeably in sales copy, and they are not the same thing.
- Scripted bots — decision trees with buttons. Cheap, predictable, and irritating the moment a visitor wants something off-menu.
- AI chatbots — they read free text and answer from a knowledge base you supply. This is what most webshops need.
- An AI agent — goes further and takes actions: looking up an order, checking stock, starting a return. Genuinely useful, but only when it is wired into the systems that hold that data.
Most shops that think they need an AI agent actually need a good answering bot plus order lookup. Start there. The rest is an upgrade path, not a starting requirement.
Six things to check before you install one
1. Does it answer from your own content?
A generic AI chat trained on the open internet knows nothing about your delivery cut-off or your 60-day returns window. It will still answer, confidently, and it will be wrong. What you want is an assistant grounded in a knowledge base you control: your shipping policy, your returns terms, your product pages, your FAQ. If the vendor cannot explain in one sentence where answers come from, that is your answer.
Feeding that knowledge base well is a skill in itself — worth reading our guide on how to train a chatbot on your own data before you upload anything.
2. What happens when it does not know?
This is the question that separates products. A bot with no answer has two options: guess, or stop and fetch a person. Guessing is how customers end up quoting a refund policy you never wrote. Stopping is boring, honest, and far cheaper.
Ask to see the failure case in the demo. Ask what the visitor sees, whether the conversation stays open, and whether the human colleague who picks it up gets the transcript. We wrote a whole piece on what a clean chatbot human handover looks like, because it is the part everyone skips and then regrets.
3. How much work is setup?
Some platforms are a chatbot builder in the full sense: a canvas, nodes, branching conditions, a workflow you maintain forever. That is powerful if you have someone to maintain it. If you do not, it becomes a stale flowchart that nobody dares touch.
The lighter alternative is to skip flow-building entirely — paste one script tag, upload your policies, done. For a shop of two people that difference is weeks.
4. Where does the conversation live?
Website chat is where most webshop questions start, because the customer is already looking at the product. Some vendors also cover WhatsApp, Facebook Messenger and Instagram from the same inbox. That is worth paying for if your customers already message you there — and worth nothing if they do not. Do not buy channels you will never open.
If you do run several, check that they share one knowledge base and one conversation history. A customer who asks on WhatsApp in the morning and on your site in the afternoon should not have to repeat themselves. Vendors describe this as moving seamlessly between channels; in practice it means every customer interaction lands in the same place, with the same context attached.
5. What does it actually cost to run?
Watch for three separate numbers: the platform fee, the per-conversation charge, and the AI usage cost. Plenty of tools advertise a free plan or a 14-day free trial and then meter conversations tightly enough that a busy month costs real money. Veyra takes a different route: the plan is free, and you connect your own OpenAI key, so the AI usage is billed to you at cost by OpenAI rather than marked up by us. If that model interests you, our piece on what a free chatbot for your website really includes goes through the fine print.
6. Does it fit how you already work?
If your team lives in a CRM or a helpdesk, check that conversations land somewhere they will be seen. If your team lives in Slack or Teams, a notification there beats another dashboard nobody logs into. The best AI chatbot for you is the one whose alerts arrive where you already look.
How does an AI chatbot on your website work?
Under the surface, most of them work the same way. Your documents are split into chunks and indexed. When a visitor asks something, the system retrieves the chunks most likely to be relevant and hands them to a language model along with the question. The model writes an answer using only that material. Nothing is memorised into the model itself, which is why updating your returns policy updates the bot's answers straight away.
That design has a useful property: because the answer is built from retrieved text, the system can tell when it retrieved nothing convincing. A well-built assistant uses that signal to say "I am not certain, let me get a colleague" instead of improvising. Artificial intelligence that knows the edge of its own knowledge is worth more in customer support than the sort that is merely fluent.
Two practical consequences follow. First, the quality of your answers is mostly the quality of your documents — a vague shipping page produces a vague reply, no matter which AI model sits behind it. Second, you can change behaviour without retraining anything: edit the page, and the next visitor gets the new answer in real time.
Questions buyers ask
Do I need a chatbot builder, or just a bot?
Only build flows if you have a process that genuinely branches — a warranty claim that needs three pieces of information before it can go anywhere, say. For everything else, a retrieval-based AI bot answering from documents is less work to set up and much less work to keep current. Flow builders reward teams with someone whose job includes maintaining them.
Can I build an AI chatbot without code?
Yes. For a website chatbot the only technical step is pasting one line of HTML into your theme, which every major platform supports through a header-scripts field. Shopify, WooCommerce, Lightspeed and plain HTML sites all take the same snippet. Our setup docs walk through each one.
How long does it take to set up?
Installing the chat widget takes minutes. Getting good answers takes longer — realistically an afternoon of gathering your shipping terms, returns policy and top twenty customer questions into clean documents, then a week of reading transcripts and patching gaps. Anyone promising a finished bot in five minutes is describing the install, not the result.
Will it replace my support team?
No, and you should be sceptical of anyone who says otherwise. It removes the repetitive layer: tracking, delivery times, stock, returns windows. What remains — complaints, damaged parcels, judgement calls — is exactly the work where a human reply protects the customer experience. If you want the fuller argument, our pillar guide to the customer service chatbot covers what to hand over and what to keep.
Is a chatbot better than live chat?
They solve different problems. A bot answers instantly at three in the morning; a person handles the conversation that decides whether someone orders again. Most shops end up running both, with the bot in front. The trade-offs are laid out in chatbot vs live chat.
What are the strongest use cases?
Order status, delivery windows, returns and exchanges, size and compatibility questions, and stock. Those five cover the bulk of a webshop inbox and they are all answerable from documents you already own. Product recommendations and upsell flows are further down the list — interesting, but they are not what is burning your afternoons.
How many support tickets will it actually deflect?
Nobody honest can tell you before you start, because it depends entirely on your mix of questions and the quality of your documentation. Run it for a fortnight, count the conversations that ended without a human, and you will have a real number instead of a vendor's average.
What does it do outside opening hours?
This is where website chatbots earn most of their keep. Orders are placed at eleven at night and questions arrive with them. A bot that can answer delivery and returns questions overnight turns what would have been support tickets into finished conversations, and the ones it cannot finish are queued with the full transcript for the morning. Nothing is lost; it is just answered later by the right person.
A sensible order to roll it out
Install the widget first and let it sit in front of customer questions with handover switched on. Read the first hundred conversations properly — they will tell you which answers are missing far more accurately than any planning session. Fill those gaps, then widen the scope.
Resist the urge to automate everything in week one. A narrow bot that is right is worth more than a broad one that is plausible, and customers forgive "let me get a colleague" far more readily than they forgive a confident wrong answer. The rest of what an assistant can take off your plate is covered in our overview of customer service automation.
If you want to see the moving parts before committing, the features overview lists what is in the free plan, including the handover behaviour and where notifications land.