A five-star rating on Google used to be enough to win the call. Now a customer just asks ChatGPT or Perplexity who does heating repair nearby, and the assistant names two or three companies, not ten. If your business isn't one of those two or three, the five stars never even enter the conversation.
A Custom GPT from the GPT Store can't be embedded on your website, OpenAI never built that path. Here's why the iframe workarounds fail, what a real chatbot widget does differently, and how to move your GPT's content into one that actually loads on your site.
This shift is already large. Recent local-search surveys put AI-assistant use for finding a nearby business at around 45 percent of consumers, up from roughly 6 percent a year earlier. Set against that, independent testing across AI assistants found ChatGPT actively recommends barely more than 1 percent of the local businesses it gets asked about. Handwerk companies, SHK and electrical especially, sit right in that gap: heavily searched, rarely named.
The fix isn't a bigger ad budget, and it isn't chasing more reviews, though reviews still help. It's making the specific facts an AI system needs (your services, your price ranges, your service area, your emergency policy) available in a form it can lift cleanly and trust. That's a different job from ranking on Google, and most Handwerk websites haven't done it yet.
AI search visibility, often shortened to GEO (generative engine optimization) or AEO (answer engine optimization), measures how likely an AI assistant is to name your business, quote your prices, or link to your site when someone asks it a trade-specific question. A page can rank first on Google and still never get quoted inside an AI Overview or a ChatGPT answer, because the assistant isn't hunting for the best-ranked page. It's hunting for the clearest, most self-contained answer to the exact question it was asked.
Customers no longer ask AI assistants generic questions. They ask hyper-specific ones: what a Heizungsthermostat replacement costs in Munich, which electrician in Augsburg does same-day Notdienst calls, whether a roofer covers a specific Landkreis. Most Handwerk homepages answer none of that. They open with a line like "Ihr zuverlässiger Partner für Sanitär, Heizung und Klima seit 1995," which is true and completely unextractable. There's no price, no service area, no named service the model can point to.
Three more technical gaps compound the problem. Schema markup, when it exists at all, is usually the generic LocalBusiness type instead of a trade-specific one like Plumber or Electrician, so search engines can't confidently categorize the business. Pricing, if it's published at all, sits in a PDF Preisliste that crawlers rarely parse well. And a surprising number of older WordPress security plugins block all bots by default, including the AI crawlers a business actually wants to let in.
None of this requires a redesign. It requires five specific, checkable changes:
Not directly. A chat widget's conversation happens inside a shadow DOM after a visitor clicks it open, so an AI crawler indexing your homepage never sees those exchanges. No search engine gives you credit for an answer a bot gave one visitor in a private session.
Where it helps is upstream. Training a RAG chatbot like AmueAI on your service pages and price sheets forces you to write exactly what an answer engine wants: a specific, source-grounded answer to "what does a Heizungswartung cost" or "do you do emergency call-outs on weekends," broken into small chunks instead of buried inside a paragraph about your company history. Once that content exists for the chatbot to retrieve, publishing the same chunks as a visible FAQ section costs almost nothing. Now two different systems, your chatbot and the AI crawlers reading your site, are working from the same accurate source.
The two problems share a root cause. Nobody has ever made your website answer the specific question a customer, or an AI model, actually has. Building a chatbot on your real documents surfaces that gap fast, because the bot fails on exactly the questions your site never answered either.
Before changing anything, find out where you actually stand:
Pick your three most common calls this month, probably a maintenance-cost question, an emergency-callout question, and a service-area question, and check exactly how your site answers each one in plain text. If the honest answer is "it doesn't, but our team knows it," that's the fix, and it's the same fix whether you're writing for a customer, a chatbot, or ChatGPT. If you're already running a trained chatbot on your price sheets, you've done most of the underlying work already; the source documents just need a public FAQ page pointing at them too.
Pricing pages advertise $20 to $150 a month, but per-resolution fees, EU VAT rules, and currency conversion change what actually lands on the invoice. Here's a real monthly cost breakdown for a small business, not another vendor comparison chart.
