AI & chatbots

Should your WordPress site have an AI chatbot?

A chatbot can answer repeat questions and capture leads at 2am, or it can annoy buyers and leak data. Here is how to tell which one you would get.

Engineering delivery session
On this page
  1. TL;DR
  2. When it earns its keep
  3. Three ways to add one
  4. What a useful one does
  5. Before you switch it on
  6. Common questions
  7. How we would set it up
AI chatbot interface answering a customer question on a WordPress website
TL;DR
  • A chatbot earns its keep when you get repeat questions with clear answers, your team loses hours to first-line replies, or buyers arrive outside office hours. It does not when your traffic is low or your questions need real judgement.
  • On WordPress, the cheapest path is a hosted widget. The safest for data is an EU-hosted assistant grounded in your own content. The most flexible is a custom build on your existing stack.
  • GDPR is the deciding constraint for German and EU sites: where the model runs, what it logs, and whether you have a data processing agreement all matter before you switch it on.
  • The point is not deflection. A good chatbot captures the lead and hands a qualified conversation to a human before the buyer gives up.
  • See how we approach AI and automation, and what an AI automation project costs before you scope one.
When it earns its keep

A chatbot pays back on repeat questions, not on every visitor.

The test is simple. Look at your support inbox and your contact form for the last month. If 40% or more of the messages are variations of the same handful of questions (pricing, availability, where to start, does it integrate with X), a chatbot can answer most of them instantly and free your team for the conversations that need a person. If every message is different and needs real judgement, a chatbot will frustrate people and you are better off making your contact path faster.

Volume matters too. Below a few hundred relevant visitors a month, the setup and monitoring cost rarely pays back, and a clear contact page does the job. Above that, the after-hours capture alone often justifies it: a buyer who lands at 9pm and gets a useful answer plus a booked call is a lead you would otherwise lose by morning.

Be honest about intent. A chatbot is a strong fit for support and lead qualification. It is a weak fit as a replacement for a sales call, a substitute for clear pages, or a gimmick on a site that has no real volume to handle.

The point is not deflection. A good chatbot captures the lead and hands a qualified conversation to a human before the buyer gives up.
Three ways to add one

Hosted widget, EU-hosted assistant, or custom build.

Most German SMEs we work with land on the middle option: an EU-hosted assistant grounded in their own pages, with a clean handoff to a human and a booking link.
ApproachBest forData residencyRough cost
Hosted widget (plugin or SaaS)Small sites that want answers live this weekOften US-based; check the processor and transfersLow monthly fee, fast setup
EU-hosted assistant grounded in your contentGerman and EU sites with privacy obligationsEU region, your content as the knowledge baseSetup project plus monthly run cost
Custom build on your stackTeams that need it wired into a CRM or booking systemYou decide; can be fully EU or self-hostedHigher build cost, lowest long-run lock-in
What a useful one does

Four jobs a WordPress chatbot should actually do.

Knowledge base of website pages feeding an AI assistant
01

Answer from your own content, not the open internet

A chatbot grounded in your pages, docs, and FAQs answers in your words and stays accurate. One that improvises from a general model will invent pricing, promise features you do not offer, and contradict your site. Grounding the assistant in your own content (retrieval over your pages) is the difference between a helpful tool and a liability.

Lead capture form completed through a chat conversation
02

Capture and qualify the lead

The goal of a support chat is not to end the conversation. It is to capture an email or a booked call and pass a qualified lead to your team with context: what the visitor asked, what they need, and how urgent it is. A chatbot that answers a question and then lets the buyer leave without a next step has done half a job.

Chat conversation handing off from an AI assistant to a human team member
03

Hand off to a human cleanly

Every assistant should know its limits and say so. When a question needs judgement, pricing negotiation, or empathy, the bot should offer a real handoff: a booked call, a routed email, or a live agent if you staff one. A confident wrong answer costs you more than an honest "let me get a person for that".

Dashboard summarising the most common questions asked in chat
04

Report what people actually ask

The transcripts are a gift. The questions your chatbot fields are your real FAQ, your missing pages, and your weakest copy, in the visitor's own words. A chatbot that logs and summarises common questions (within your privacy rules) tells you what to fix on the site itself.

At a glance
40%
A rough threshold: when this share of inbound messages are repeat questions, a chatbot starts to pay back.
24/7
After-hours capture is often the clearest win: a useful answer plus a booked call while your team sleeps.
1
One handoff path that works beats ten clever answers. Always give the buyer a route to a human.
Before you switch it on

The GDPR and setup checklist for a German or EU site.

Confirm where the model runs and where transcripts are stored. For EU sites, prefer an EU region or a self-hosted model.
Sign a data processing agreement with every processor in the chain, and list them in your privacy notice.
Decide what the chatbot logs, for how long, and who can read it. Do not retain more than you need.
Gate the chat behind consent if it sets non-essential cookies or sends data to a third party.
Ground the assistant in your own content so it cannot invent pricing or features.
Write the handoff rules: which questions route to a human, and how fast you respond.
Test the failure cases: rude input, off-topic questions, and requests the bot must refuse.
Common questions

What teams ask before adding a chatbot to WordPress.

Is an AI chatbot GDPR-compliant on a German website?

It can be, but compliance depends on the setup, not the technology. The deciding factors are where the model runs, what the chatbot logs, how long transcripts are kept, and whether you have a data processing agreement with each processor. For a German or EU site, an EU-hosted assistant grounded in your own content, with a clear privacy notice and consent for any non-essential data, is the practical path. A US-hosted widget that logs everything by default is the risk.

Will a chatbot replace my contact form?

It should not replace it, it should sit alongside it. Some buyers want to type a question and get an instant answer; others want to fill a form and move on. Keep both. The chatbot handles repeat questions and captures leads after hours; the form stays as the simple, no-friction path for people who already know what they want.

How do I stop the chatbot from making things up?

Ground it in your own content and constrain it. An assistant that answers only from your pages, docs, and FAQs (retrieval over your content) stays accurate. Add explicit rules: do not quote prices it cannot verify, do not promise features, and hand off to a human when unsure. A general model with no grounding will improvise, and that is where wrong pricing and false promises come from.

What does it cost to run a WordPress chatbot?

A hosted widget is a low monthly fee and fast to set up. An EU-hosted assistant grounded in your content is a setup project plus a monthly run cost that scales with usage. A custom build wired into your CRM costs more upfront but avoids per-seat lock-in. The ongoing cost is usually model usage plus monitoring; budget for someone to read transcripts and improve the answers, because an unmonitored bot drifts.

How do I measure whether it works?

Track three things: the share of conversations that end with a captured lead or booked call, the share resolved without a human, and the questions it fails. Resolution rate alone is misleading; a bot that ends conversations without capturing anything is losing you leads quietly. The transcripts also tell you which pages to fix, which is value beyond the chat itself.

Can I add one without slowing the site down?

Yes, if you load it carefully. Chat widgets are a common cause of poor Core Web Vitals because they pull in heavy scripts on every page. Load the widget lazily, after the page is interactive, and only on pages where it earns its place. See our Core Web Vitals guide for how third-party scripts affect INP.

How we would set it up

EU-hosted, grounded in your content, with a real handoff.

For most German and EU SMEs, we start with an EU-hosted assistant grounded in the site's own pages and FAQs, gated behind consent where needed, with transcripts kept only as long as they are useful. The chatbot answers the repeat questions, captures the lead, and books a call or routes an email when a person is needed. The handoff is the part most setups get wrong, so we design it first: which questions go to a human, and how fast you reply.

Then we watch it. The first month of transcripts shows what people actually ask, which pages are missing, and where the answers need tightening. See how we approach AI and automation for systems that earn their run cost, and the AI automation project cost guide before you scope one.

Concrete solution

Bring the operational risk.You get a clear diagnosis and a concrete next step.

or book a 15-minute call

We are the right fit if you want a team that pushes back when it matters. See outcomes and metrics

Reviewing first?

Company evidenceon the site.

Engagements with commercial outcomes on Work. Team bios and operating model on About. Nothing to download. Review it before you commit to a call. Open to review. Commit when ready.