STRG.answer · Search and chat in one
Visitors ask in full sentences, your full-text search demands exact words. STRG.answer answers the same query twice over: a sourced answer from your content, and the matching results next to it. Your team reads along on every conversation and can take it over.
The query
A query does not run one path after another; it runs three at once. Only what all three bring back goes to the language model as context.
A question in full sentences
Exact matching for product numbers, proper names and abbreviations, where every character counts.
Typos and name variants that exact matching would never catch.
What is meant rather than typed, even when no word of the question appears in the text.
A sourced answer from your own content The language model formulates exclusively from the passages found and names the source. If the index turns up nothing solid, the system says so.
The problem
Someone used to asking an AI a complete question no longer types two keywords into your site either. They write “Which variant is suitable for outdoor use in frost?”. Full-text search finds nothing, because none of those words appear that way in the datasheet. The visitor does not conclude that your search is bad. They conclude that you do not carry the product. The same precondition now also decides whether your content appears in generated answers beyond your website: structurally indexed and semantically understood.
How it works
STRG.answer does not replace your CMS and migrates nothing. We read your content through the existing interface and build a semantic index alongside it.
Connected to your CMS, PIM or shop system, indexed section by section and kept up to date. The index knows not only documents but how they relate: which variant meets which specification, which article belongs to which product.
Full-text, phonetic and semantic in a single pass. Technical terms and your internal nomenclature are learned along the way instead of maintained by hand.
The language model formulates exclusively from the passages found and names the source. If the index turns up nothing solid, the system says so instead of guessing.
Alongside it, searches run against further content, products and services. Where a concrete offering fits, it is linked directly.
Chat functions
On your website sit visitors who do not know your range and do not speak your nomenclature. Behind it several agents work in parallel: one searches, one formulates, one checks which offering fits. Each can be configured and switched off individually.
The fourth question still knows what the first was about. Visitors narrow down inside the conversation instead of starting over with new keywords.
While the answer forms, the result list fills up next to it. Every statement carries its source and is one click away in the original.
Where a question points to a product, a programme or a service, the answer names it and links straight there.
The system classifies the intent and tone of the question and steers the length and register of the answer with it. The same classification shows sales how far along a visitor is. Details in the FAQ under mood detection.
Your team can take over any running conversation. From that moment a person answers, in the same window, with no channel switch for the visitor.
The backend
The system stands on a public website and talks to people who are not customers yet. That makes it a sales channel, and a sales channel needs a console.
In the backoffice you steer the channel: which agents run, with which language model, and what stays switched off. Every conversation can be retraced down to the passages the answer rests on. The rest is analysis, and for a sales team that is often worth more than the search itself.
Use cases
Twenty years of articles are an asset as long as they can be opened up. Readers get the context that keeps them on the site with every article, and the archive becomes reach again.
What became of the pilot regions after the 2019 care reform?
A customer rarely knows the name of the product they need, but they know their use case. Semantic search translates use case into variant, including the specifications that only appear in the datasheet.
Which seal withstands 180 °C and oil, and fits DN 50?
Insurance, further education, B2B services: the prospect knows their situation, not your catalogue of offerings. The conversation turns it into a matching offering, and where it gets concrete, your team takes over.
I am an engineer, 39, and want to move into leadership. Which programme fits?
Proof
The semantic analysis algorithms come out of our STRG.behave research project (2017), optimised continuously since and extended with LLM mechanics in 2023. So not a prototype that goes under load for the first time with your project. How far that carries is clearest in the archive of the weekly newspaper Die Furche: the complete holdings since 1945, semantically located rather than sorted by keyword.
Integration & operations
STRG.answer is connected to your existing system, not developed into it. Proven with Storyblok, TYPO3, WordPress, Adobe Experience Manager and STRG.CMS. In the frontend there is a search and chat component that adopts your design, or you address the API directly.
The connection
Your content stays where it is. Beside it we build an index that is kept up to date.
CMS, PIM or shop system, read through the existing interface or an export.
Indexed section by section, stored semantically, updated on every change. Operated in the EU.
A search and chat component in your design, or your own interface straight on the API.
Research, fast content discoverability and context now decide reach.
FAQ
Read about semantic search in STRG.Magazine
Next step
We take a delimited part of your content, index it and show you what the system answers with it and where your content has gaps. Not a pitch deck, but your data in a running search.
