Some businesses sell something a visitor understands in five seconds: a haircut, a boiler service, a table for four. Others sell something that takes a site visit, a diagram and twenty minutes of patient talking before the customer knows what they are buying. Heat pump installation. Commercial lease reviews. Managed IT with three tiers and a list of exclusions.
For the second kind, the website usually fails in one of two ways. Either it is so simplified that it says nothing ("tailored solutions for your needs") or it is written by the engineer and reads like a specification. Visitors leave either way, confused and slightly embarrassed to ask.
A well-configured explain services chatbot closes that gap by doing what your best salesperson does on the phone: start simple, check what the person actually wants to know, and go deeper only when asked.
The layered answer, and why it works
A good explainer does not start with everything. Picture a heat pump installer's website and a visitor asking the obvious question.
Visitor: Would a heat pump work in my house?
Assistant: In most homes, yes, as long as the house holds heat reasonably well. The main things that matter are insulation, the size of your radiators and whether there's outdoor space for the unit. Do you know roughly when your house was built, or whether the loft and walls are insulated?
Visitor: 1930s semi, loft is done, walls I'm not sure.
Assistant: That's a common type for heat pumps. Many 1930s semis have cavity walls, and whether they've been filled makes a real difference to running costs. Some radiators may also need upgrading so they can heat rooms at lower water temperatures. A survey checks both. Would you like to request one?
Notice what did not happen. No explanation of refrigerant cycles, no efficiency ratings, no wall of caveats. The first layer was one plain sentence. The second layer was specific to what the visitor said. The detail was available but not forced.
Rewrite the source content before the assistant reads it
The assistant repeats the tone and vocabulary of what you feed it. If your service page says "we deliver a holistic assessment of thermal performance", that is roughly what visitors will hear back.
A quick before and after, taken from the kind of copy that sits on many technical service pages:
| Original page copy | Rewritten for explanation |
|---|---|
| Comprehensive thermal performance assessment | We check how well your house keeps heat in, room by room |
| SLA-backed tier 2 remote remediation | If something breaks, we fix it remotely within the time in your contract |
| Schedule of condition prepared pre-commencement | Before work starts, we photograph and record the state of your property |
| Subject to site-specific constraints | It depends on your house. Here is what usually changes the answer |
You do not have to rewrite the whole website. Pick the ten questions people ask most and write plain answers as question and answer pairs. The AI FAQ generator can give you a starting list to edit. Then read how to train your chatbot on your website content for how the crawl and your hand-written answers work together.
Writing explanations that land
Here is the process that tends to produce answers a visitor can actually use.
- Write the one-sentence version first. If you cannot explain what the service does for the customer in one sentence, the assistant will not manage it either.
- Add the conditions separately. "It depends" is honest, but only useful when followed by what it depends on. List the two or four factors that change the answer.
- Give a typical example. A 1930s semi, a ten-person office, a shop with one till. Examples let people place themselves.
- Say what the service does not include. Exclusions answered upfront save angry emails later.
- Define the unavoidable terms once. If customers must know what a "survey" or "retainer" means, put a plain definition in your content and let the assistant use it. The glossary pattern works well here.
- Test with someone outside the trade. Ask a friend to question the assistant. Every time they frown, rewrite.
Signs a person needs to take over
Explaining is not the same as advising, and a complicated service is exactly where the line blurs. The assistant can explain how something generally works. It should not decide what is right for this particular customer when that depends on facts only a professional can assess.
Useful triggers for a handover:
- The visitor describes their specific situation in detail and asks what they should do. General explanation has run out; judgement is needed.
- The answer has legal, financial, safety or health consequences. The assistant gives general information from your content and makes clear it is not advice, then routes to a qualified person.
- The visitor has asked the same question twice in different words. The explanation is not landing. A person on a call will do better.
- They are comparing detailed options such as two contract tiers with different exclusions and want a recommendation.
When nobody is on duty, the handover becomes a captured enquiry with the conversation attached, so whoever calls back already knows what was explained and where the confusion was. That context is worth a lot for a service that normally takes three emails to scope. The article on when a chatbot should hand over to a human goes further on setting those rules.
What to review after launch
Complicated services generate unusual questions, so the unanswered question log matters more than usual. Read it weekly for the first month. You will find three kinds of gap: facts you never published, terms visitors use that you do not (they say "air source", you say "ASHP"), and questions that reveal the service page itself is confusing.
Fix the content, not only the assistant. A clearer page helps everyone, including the visitors who never open the chat.
Start with one question
Pick the single question your team explains most often on the phone, the one that takes ten minutes and a pen. Write the one-sentence answer, the factors it depends on, one typical example and the point where you would say "we need to look at your situation". Add that to the assistant as a question and answer pair and test it against five different phrasings.
When that one explanation works, do the next one. Most complicated services come down to perhaps a dozen of these, and the features you need to handle the rest (lead capture, tickets, handover) are already there to catch whatever the explanation cannot.