Six ways to build smarter on-site AI search
Silicon Labs’ “Ask AI” on-site search
AI search tools on corporate websites are shifting from concept to reality. A handful of early adopters are replacing traditional keyword search tools with a ChatGPT-style experience that feels faster, smarter and more personalised.
At a recent Bowen Craggs Club event, we spoke with Melanie Mrochen, Senior Product Manager for Digital Experience at Siemens and Lauren Handel, IoT Product Manager at Silicon Labs, about how they are using AI to transform on-site search on their corporate websites – what worked, what didn’t and how to build user trust in AI tools.
Here are six takeaways from the discussion.
Put users first
“For many customers, finding the right information on our website quickly is still a challenge,” said Siemens’ Melanie Mrochen. “Traditional keyword search was just not enough anymore.”
Both companies stressed that AI should be deployed to solve real user pain points - not simply because it is trendy. A quicker, more efficient search experience is a great starting point, but it needs to be tailored to stakeholder needs and surface the information they will find the most relevant.
Involve stakeholders from day one
“You can’t build an AI in a silo,” Melanie explained.
Siemens brought legal, IT, data privacy and compliance colleagues into the project early. This ensured risks were spotted and addressed, while also helping to secure buy-in.
Guardrails are non-negotiable
Silicon Labs partnered with Kapa.ai because, as Lauren put it, “It does not guess. If it does not know, it lets the person know – and that was huge for us.”
Siemens took a multi-agent approach with deny lists and strict content barriers, built on AWS infrastructure in partnership with Groundfog. In both cases, the focus was on transparency and control to avoid misleading answers.
Siemen’s products and services-focused AI on-site search
Content quality drives performance
AI search is only as good as the content behind it. “We discovered people were finding outdated press articles,” Melanie recalled, prompting Siemens to exclude older content.
Make sure you are reformatting complex tables or images into more AI-friendly formats, and prioritising the most relevant, recent content, so the system can parse it effectively.
Human oversight keeps standards high
Silicon Labs relies on engineers to review answers that users have flagged as unhelpful and retrain the model. As Lauren put it: “There always needs to be human oversight when we’re adding AI tools.”
Siemens runs automated multilingual tests to spot errors.
This keeps AI tools accountable and reliable.
Adoption requires visibility
Simply adding an “AskAI” button wasn’t enough at Silicon Labs.
“We had to market it and make it very obvious,” Lauren said. Once users tried it, feedback was overwhelmingly positive: “Where was AskAI all along? I love this.”
Revolutionising your on-site search will take time and investment, but if your broader teams are aligned and onboard, you will be one step closer.
If you are interested in exploring the on-site AI searches currently available, take a look at: