DEV Community: Sarah Hänschen

When AI Shops for Customers: Is Your Product Data Ready?

Sarah Hänschen — Tue, 20 Oct 2026 12:20:29 +0000

A lightweight hiking shoe, waterproof, suitable for longer hikes, and under 180 euros: someone searching for a product this way is describing a specific need. An AI assistant can compare suitable offers and help with the selection. To do so, it must be able to identify which products actually meet the requirements.

For retailers, this brings the quality of their product information into sharper focus. An attractive image sparks interest. Compelling copy communicates benefits. Whether a product is actually suitable, however, often depends on specific details: material, dimensions, intended use, available variants, or delivery time.

The more tasks AI takes on in preparing a purchase, the more important one simple question becomes: Are your products described in a way that allows a reliable recommendation to be made?

AI Is Becoming a Relevant Gateway to the Product Range

AI-powered product search has long been a topic for retailers. In July 2026, the purchase conversion rate of visitors who reached US retail websites through AI services was 60 percent higher than that of visitors from other channels. This is shown by recent Adobe data.1

For retailers, this is a reason to examine this route to their own product range more closely. What information do potential customers find when they research with the help of AI? Can they understand the product benefits? And is the information still accurate when a recommendation turns into concrete purchase interest?

It is worth looking beyond visibility alone. A product can appear in an answer and still be unsuitable—for example, because a particular size is unavailable or an accessory does not fit the existing device. The quality of the information on which the selection is based is therefore also crucial.

From Product Recommendations to Shopping Instructions

The willingness to give AI more responsibility already exists: 29 percent of online shoppers would like to delegate purchasing decisions or purchases to AI within the next five years.2

This makes the question of how well a customer’s request can be matched with the available offering increasingly important. One possible instruction might be: “Find a compatible replacement battery for my device that can be delivered by Friday at the latest.”

To support this request effectively, several details must align: the exact device model, confirmed compatibility, available stock, and reliable delivery information. A general description such as “high-performance battery for many devices” is of little help here.

Preparing for AI-powered shopping processes therefore begins with a task retailers can address directly: providing complete, unambiguous, and up-to-date information that is essential to purchasing decisions.

Which Product Information Is Particularly Important Now

Clearly Identifiable Products and Variants

Model names, item numbers, and variants must be clearly linked. If a jacket is available in blue, that does not automatically apply to the same jacket in black. Price, size, and delivery time should also be clearly assigned to the respective variant.

Attributes That Answer Customer Questions

Which details are relevant depends on the product range. For a desk, width, depth, and height adjustability may matter. For a monitor, connections and resolution are crucial. For spare parts, it must be clear which models they are suitable for.

Questions directed to your customer service team are a useful starting point: what do buyers regularly want to know before making a decision? This is precisely the information that should be reflected in your product data.

Current Prices and Availability

A recommendation must still be suitable at the time of purchase. Therefore, check how quickly changes to prices, stock levels, and delivery times reach the respective channels. Responsibilities and update processes are just as important as the initial data maintenance.

Consistent Information

The product page, catalog export, and connected data sources should communicate the same essential information. If dimensions differ between two places, the selection becomes unnecessarily uncertain. Such inconsistencies should be systematically recorded and corrected at their source.

Clear Descriptions and Specific Attributes Belong Together

Good product data must fulfill two tasks: help people make decisions and provide unambiguous information for further processing.

For the hiking shoe, a clear description can explain which types of hikes it is intended for. Specific attributes should also be maintained separately. This makes it possible, for example, to distinguish whether “lightweight” is a general marketing claim or can be put into context through a specified weight.

Limitations also belong in a helpful description. What uses is the product not intended for? What additional accessories are required? What conditions must be met? Answering such questions early makes an informed selection easier.

Trust Requires Decisions That Can Be Understood

As automation increases, clear rules become more important. Seventy-six percent of consumers want clear guidelines on when an AI assistant takes action.3

For the design of an AI-powered shopping process, this means recommendations should be understandable, price changes should be visible, and purchasing decisions should receive clear approval. If important information is missing, that uncertainty should remain apparent.

Particularly when it comes to compatibility or safety-related product attributes, a plausible-sounding assumption must not replace confirmed information. In such cases, a helpful follow-up question can be more valuable than a premature recommendation.

How to Prepare Your Product Range Step by Step

Start with a manageable product category and collect typical customer questions. Then check whether they can be answered using the existing product information.

Prioritize the gaps you identify according to their importance for the purchasing decision. Missing compatibility information or contradictory dimensions usually deserve attention sooner than additional marketing copy. Also establish who will add the information and keep it up to date in the future.

Then test the same questions again: can suitable products be found? Are the variants correct? Can recommendations be justified using specific attributes?

This creates a verifiable foundation for better product search—in your own store and in preparation for AI-powered shopping processes.

How effectively can your product range be explored using real customer questions? Talk to Quasiris about your product search and the next practical steps.

Find out which AI features support your product search, better understand customer needs, and make suitable offers discoverable in our overview of the AI Features of the Quasiris Search Cloud.

💡 How Good Is Your Search Really?

Irrelevant results, high zero-results rates, and labor-intensive manual ranking rules often indicate that a search solution is not yet making full use of its potential.

Our free Search Review provides an initial assessment of the quality of your existing search and potential opportunities for optimization.

↓ Request Your Free Search Review Now ↓

Find out more now!

Transparency notice: This image was generated using artificial intelligence and is for illustrative purposes.*

Sources

1 Adobe Digital Insights. (2026). AI traffic trends report: August 2026. Adobe. https://business.adobe.com/assets/pdfs/resources/sdk/ai-traffic-trends-report-august-2026/ai-traffic-trends-report-2026-08-19.pdf

2 DHL eCommerce. (2026). 2026 e-commerce trends report. https://www.dhl.com/content/dam/dhl/local/global/dhl-ecommerce/documents/pdf/g0-ec-trends-report-2026.pdf

3 Capgemini Research Institute. (2026). What matters to today’s consumer 2026. Capgemini. https://www.capgemini.com/insights/research-library/what-matters-to-todays-consumer-2026/

Inside AI Search #4: Question Answering

Sarah Hänschen — Tue, 20 Oct 2026 12:07:51 +0000

In the “Inside AI Search” series, Quasiris shows what lies behind the individual features of modern AI-powered search. This article focuses on Question Answering. The idea is simple: users should not have to find the right document first and then search for the information themselves. They ask a question and receive a specific answer.

Technically, however, there is considerably more to it than a chat window.

A reliable Question Answering solution must understand what was asked, where the relevant information is located, and whether the available sources can provide a reliable basis for an answer at all.

What Is Question Answering?

Question Answering refers to systems that directly answer questions in natural language.

For example, a user asks:

“How long is the notice period for contract type A?”

A traditional search returns several contracts, policies, or FAQ entries.

A Question Answering system instead attempts to first find the relevant information and then provide a direct answer based on it.

The underlying source should remain available.

This distinguishes professional Question Answering from a general chatbot that answers exclusively from its model knowledge.

The Real Work Happens Before the Answer

A large language model can formulate an answer very convincingly. Whether it is correct, however, depends heavily on the information available to it.

A Question Answering process therefore usually consists of several components:

1. Understand the Question

What does the user actually want to know?

2. Search for Relevant Information

Which documents, databases, or knowledge sources contain the answer?

3. Evaluate the Results

Which of the retrieved content is particularly relevant to the question?

4. Assemble the Context

Which passages does the model need?

5. Generate the Answer

The information is expressed clearly.

6. Provide the Source

The user can trace the underlying information.

Retrieval quality is therefore an essential part of answer quality.

An Answer May Require Several Search Steps

Simple questions can often be answered using a single document.

Enterprise questions, however, are often more complex.

For example:

“Which server configuration does Project X use, and will the current version of our software run on it?”

A document about Project X might contain only a server ID.

Only a second source contains the server’s technical specifications. A third may contain the system requirements for the current software version.

Modern Question Answering systems must recognize such chains of information.

In 2026, Google Research describes precisely this challenge in Enterprise RAG. Traditional single-stage retrieval systems reach their limits when answering questions that require information from multiple sources or several successive search steps. Agentic retrieval methods break down such questions and iteratively search for additional context.1

Enterprise Documents Are More Complex Than Plain Text

Knowledge is not found only in continuous text.

Reports contain tables. Presentations contain charts. Manuals combine images, headings, and technical values. PDFs have layout structures that are relevant to interpretation.

This makes Enterprise Question Answering demanding.

An Enterprise RAG method presented at ACL 2026 therefore explicitly takes the structure of different document types into account. On a heterogeneous enterprise dataset and two public benchmarks, the method achieved improvements of up to 32 percentage points over visual comparison systems, particularly for report-like document structures.2

The practical implication is:

Documents must not only be stored. Their structure must be made usable for search.

Direct Answers Need Boundaries

A good Question Answering system must not only know when it can answer.

It should also recognize when the available information is insufficient.

Take the question:

“When will Product X be available again?”

If no source contains a delivery date, several responses are possible.

A poor response would be:

“Product X is expected to be available next week.”

A better response would be:

“There is currently no confirmed delivery date recorded in the available data.”

This ability to exercise restraint is a central component of trustworthy Question Answering systems.

Recent analyses of search-augmented LLMs even show that more retrieval does not automatically lead to better answers. Irrelevant or noisy search results can reduce quality and encourage additional hallucinations.3

More data is therefore not automatically better.

What matters is the right data.

What Agentic Retrieval Changes

Traditional RAG often follows the pattern:

Question, search, context, answer.

For more complex tasks, this can be too rigid.

In 2026, Microsoft Research introduced AgenticRAG, an approach in which the model combines existing enterprise search infrastructure with tools for searching, opening, navigating, and summarizing. In the benchmarks examined, the approach achieved, among other results, 92 percent answer correctness on FinanceBench, placing it just two percentage points behind a scenario with direct access to the correct supporting evidence.

This shows where Question Answering is heading.

Search no longer provides context only once. A system can actively determine:

I am still missing a piece of information. I need to search again.

Where Question Answering Is Particularly Useful

Direct answers are particularly suitable for information-intensive processes.

Internal Knowledge Search

Questions about policies, processes, and internal documentation can be answered directly.

Technical Support

Employees can search manuals and troubleshooting documentation using natural-language questions.

Customer Service

Service agents receive relevant information from multiple knowledge sources in a single interface.

Product Advice

Product characteristics and differences between products can be explained in a targeted way.

Self-Service

Customers can retrieve information without needing to know the structure of a portal.

Question Answering Does Not Replace Document Search

Direct answers are convenient. However, there are situations in which users still need the complete document.

For contracts, technical regulations, or legally relevant content, for example, the exact wording can be crucial.

A good search experience should therefore combine both:

A direct answer when a specific piece of information is requested.

The original source when details or context need to be checked.

Additional results when multiple perspectives or documents are relevant.

Question Answering thus extends search rather than replacing it entirely.

How Can Quality Be Assessed?

An answer that sounds good is not necessarily a good answer.

An evaluation should consider several levels:

Is the answer factually correct within the subject area?
Do the sources actually contain the statement?
Were the most relevant sources used?
Were important limitations included?
Can the system recognize unanswerable questions?
Do permissions work correctly?
How up to date is the information used?

Particularly in an enterprise environment, evaluation should be carried out using real questions from the relevant department.

A general benchmark score cannot replace this assessment.

Conclusion: The Best Answer Begins with the Right Search

Question Answering makes enterprise knowledge more accessible. Instead of searching documents manually, users can ask their actual question and work directly with the relevant information.

Quality, however, does not emerge only when a language model formulates the answer. It begins with retrieval, the data sources, and the decision about which content may be used for an answer.

Implement Question Answering with Quasiris

The Quasiris Search Cloud supports Question Answering and connects natural-language questions with search and retrieval across connected knowledge sources.

Would you like to provide direct answers from your documentation, wikis, or product data? Talk to our search experts about your use case.

We show you which AI features can also create added value for you in our overview of the AI Features of the Quasiris Search Cloud.

💡 How Good Is Your Search Really?

Irrelevant results, high zero-results rates, and labor-intensive manual ranking rules often indicate that a search solution is not yet making full use of its potential.

Our free Search Review provides an initial assessment of the quality of your existing search and potential opportunities for optimization.

↓ Request Your Free Search Review Now ↓

Find out more now!

Transparency notice: This image was generated using artificial intelligence and is for illustrative purposes.
Sources

1 Google Research. (2026). Unlocking dependable responses with Gemini Enterprise Agent Platforms’ Agentic RAG. https://research.google/blog/unlocking-dependable-responses-with-gemini-enterprise-agent-platforms-agentic-rag/

2 Association for Computational Linguistics. (2026). Enterprise RAG for heterogeneous document understanding. In Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics: Industry Track. https://aclanthology.org/2026.acl-industry.134/

3 Microsoft Research. (2026). AgenticRAG: Agentic retrieval for enterprise knowledge bases. https://www.microsoft.com/en-us/research/publication/agenticrag-agentic-retrieval-for-enterprise-knowledge-bases/

Inside AI Search #3: Smart Snippets

Sarah Hänschen — Tue, 13 Oct 2026 11:56:38 +0000

In the “Inside AI Search” series, Quasiris shows what lies behind the individual features of modern AI-powered search. This article looks at a feature that may seem small at first glance but can significantly change how people interact with search results: Smart Snippets.

A traditional search displays a title, a link, and an excerpt from the retrieved document. Smart Snippets go one step further. They attempt to make the exact part of the content that matches the current question visible, or to generate a short, understandable answer from it.

This reduces one of the biggest points of friction in search: finding a relevant document does not yet mean finding the relevant information.

What Are Smart Snippets?

A snippet is a preview of a search result.

In a traditional document search, it might look like this:

Travel Expense Policy 2026
“… business trips must, in accordance with the applicable policy, before departure …”

For the question:

“What is the hotel allowance for Berlin?”

this excerpt is of little help.

A Smart Snippet can specifically highlight the passage that relates to the question or summarize the information concisely.

The result might, for example, read:

Hotel costs in Berlin are covered up to the defined maximum amount. Details can be found in Section 4.2 of the Travel Expense Policy.

This gives the user guidance without losing sight of the source.

Why Snippets Are Becoming More Important

Search results are increasingly not just scanned but evaluated directly on the results page.

An eye-tracking study published by Microsoft Research in 2026 examined how people interact with search pages containing generative summaries. The generated content received significantly more attention than traditional results, even though users continued to interact with traditional links.1

This has an important implication for enterprise search and product search:

The content visible directly in the results list strongly influences whether a user perceives the information as relevant.

A poor excerpt can make a good result appear unusable.

A good snippet, on the other hand, can explain why the result is relevant before the document is even opened.

Smart Snippets Are Not Just Any AI Summaries

A generated summary of an entire document and a Smart Snippet serve different purposes.

A summary answers:

What is this document about?

A good Smart Snippet, on the other hand, answers:

Why is this document relevant to my current query?

This is a crucial difference.

The same product page may require a different excerpt depending on the query.

For:

“How long does the battery last?”

the snippet should highlight battery life.

For:

“Is the device compatible with Windows 11?”

the same product page is relevant, but different content should be displayed.

Smart Snippets are therefore query-dependent.

The Source Must Remain Visible

Generative AI can turn long documents into short, easy-to-understand texts. However, this very capability creates a new responsibility.

Users must be able to distinguish between:

The original content
An extracted excerpt
An AI-generated summary

The more important the information, the more relevant this transparency becomes.

In 2026, for example, Google more closely integrated its generative search interfaces with direct links, source previews, and further content. The aim is to continue giving users access to the underlying websites alongside the summarized information.2

The same principle applies to enterprise search.

A Smart Snippet should shorten the path to the source, not hide the source.

Grounding Determines Reliability

When Smart Snippets are generated, they should be based on the content actually retrieved.

This so-called grounding connects the generated output to a verifiable source.

Current systems rely on precisely this approach. Google, for example, describes grounding as a way to base model responses on retrieved information and connect sources directly to the answer.3

In enterprise systems, this source may be a product database, a wiki, technical documentation, or an internal search index rather than the public web.

The underlying principle remains the same:

First find relevant information, then generate the presentation from it.

Not Every Result Needs Generated Text

Smart Snippets do not necessarily have to be generated by a large language model.

With structured data, direct extraction can often be better.

For example, if a user searches for:

“Delivery time Product X”

and the delivery time is available as a structured attribute, the system should display this information directly.

Generative methods become particularly useful when:

Relevant information is spread across several sentences
Long documents are being searched
Content needs to be summarized clearly
Different wording is used
A natural-language question needs to be answered

The appropriate method therefore depends on the data type.

Smart Snippets in Practice

There are many possible areas of application.

Knowledge Management:

A policy is found, and the passage relevant to the question is displayed directly.

Technical Support:

A manual provides the relevant solution step directly in the search result.

E-commerce:

Product attributes that match the current query are highlighted.

Customer Service:

Service agents immediately see the relevant passage from a knowledge article.

Document Search:

Long PDFs can be assessed without having to open every document in full.

Smart Snippets can therefore save time, particularly when working with large collections of information.

What Should Be Measured During Implementation

Success cannot be determined solely by whether users click on results less often.

Fewer clicks can be positive because the information is already visible. However, they can also mean that the snippet is not convincing.

Useful questions therefore include:

Does the snippet provide the information being sought?
Is the source clearly identifiable?
Do the snippet and the original content agree?
Do users reach the information they need more quickly?
Are important limitations presented correctly?
Does the system recognize when a reliable short answer is not possible?

The last point is particularly important.

A system that does not have a sufficient basis for an answer should not attempt to fill the gap creatively.

Conclusion: Smart Snippets Shorten the Path from Result to Information

Search quality does not end with ranking.

A result must also make it clear why it is relevant. Smart Snippets create precisely this connection between the results list and the actual content.

Used correctly, they can save users time, provide guidance, and preserve access to the original source.

The Quasiris Search Cloud supports Smart Snippets to provide relevant information matching the query directly within the search experience.

Would you like to explore how Smart Snippets could work with your product data or documents? Discuss your use case with our search experts.

We show you which AI features can also create added value for you in our overview of the AI Features of the Quasiris Search Cloud.

Transparency notice: This image was generated using artificial intelligence and is for illustrative purposes.

Sources

1 Allawati, S., McKay, D., Sanderson, M., Thomas, P., & Trippas, J. R. (2026). An eye tracking study: Are AI overviews changing search behavior? Microsoft Research.
https://www.microsoft.com/en-us/research/publication/an-eye-tracking-study-are-ai-overviews-changing-search-behavior/

2 Budaraju, H. (2026). 5 new ways to explore the web with generative AI in Search. Google.
https://blog.google/products-and-platforms/products/search/explore-web-generative-ai-search/

3 Google. (2026). Grounding with Google Search. Google AI for Developers.
https://ai.google.dev/gemini-api/docs/google-search

Inside AI Search #2: Predictive Search

Sarah Hänschen — Mon, 21 Sep 2026 12:18:07 +0000

In our “Inside AI Search” series, we look at features that make search not only more intelligent but also genuinely more helpful in everyday use. Predictive Search comes into play particularly early: before a search query has even been fully entered.

Many people know this principle as autocomplete or typeahead. However, modern Predictive Search goes beyond completing popular search terms.

It attempts to anticipate what a user is likely looking for and which next step would be helpful.

What Is Predictive Search?

Predictive Search generates suggestions as you type.

If someone starts with:

“Hiking…”

suggestions like these might appear:

Hiking boots for men
Waterproof hiking boots
Hiking backpack 30 liters
Hiking jacket for women

The real added value does not lie in saving a few keystrokes.

Good suggestions help users formulate a better query. They introduce terms from the product range, correct spelling mistakes, and show possible search directions.

Extensive UX analyses by Baymard show precisely this. Autocomplete not only helps users type faster but, above all, helps them find suitable search queries and the right search scope.1

From Autocomplete to Predictive Search

A simple autocomplete system can rank suggestions solely by frequency.

Frequently searched terms appear near the top.

This works, but it only takes part of the context into account.

Modern Predictive Search can also process signals such as:

The text already entered
Current trends
Language
Category
Available products or documents
Previous search steps
Location
Current session

User context, where its use is permitted

For example, in addition to common search queries, Google’s autocomplete also takes language, location, current trends, and, under certain conditions, previous searches into account.2

This turns basic text completion into a ranking problem.

The system must decide which few suggestions are most helpful at that moment.

Context Becomes More Important Than Popularity Alone

Imagine two users who both enter “Java.”

One person is browsing developer documentation. The other is searching a travel platform.

For the first, suggestions about Java versions or frameworks might make sense. For the second, trips to the Indonesian island may be relevant.

Predictive Search should therefore take the context of the application into account.

Recent work on autocomplete in dialogue systems confirms this effect. An EACL publication from 2026 shows that incorporating the preceding conversation context can significantly improve the quality of completions. At the same time, the study reveals something interesting: more complex models are not automatically always better. For previously known input patterns, traditional statistical methods could be more efficient and, in some cases, perform better, while neural models showed advantages for unfamiliar queries.3

This is an important insight for practical use.

Predictive Search does not necessarily require a large language model for every input. Different methods can be combined depending on the situation and latency requirements.

Milliseconds Matter

Search suggestions need to appear while the user is typing.

A response after several seconds might be technically correct, but in practice it would hardly qualify as Predictive Search anymore.

Latency is therefore an essential part of the feature.

A method presented at ACL 2026 for a large mobile shopping application even attempts to predict whether the current input is already complete. If the search intent is recognized, the search can be prepared earlier. In the deployment described, the method reduced response time by around 640 milliseconds on average.4

This shows how closely search experience and search technology are connected.

Generating a relevant suggestion is not enough. It has to appear at the right moment.

Predictive Search Can Suggest More Than Search Terms

Depending on the use case, different types of suggestions can be useful.

Query Suggestions

Complete search queries are suggested.

Product Suggestions

Specific products appear while the user is still typing.

Inside AI Search #1: Natural Language Understanding

Sarah Hänschen — Thu, 17 Sep 2026 13:45:02 +0000

In der Reihe „Inside AI Search“ zeigt Quasiris, was hinter den einzelnen Funktionen moderner KI gestützter Suche steckt. Den Anfang macht Natural Language Understanding, kurz NLU. Die Technologie sorgt dafür, dass eine Suchmaschine nicht nur einzelne Begriffe verarbeitet, sondern erkennt, was ein Nutzer mit seiner Anfrage erreichen möchte.

Das wird immer wichtiger. Menschen gewöhnen sich daran, Fragen so einzugeben, wie sie sie auch einer anderen Person stellen würden. Google berichtet beispielsweise, dass sein AI Mode inzwischen mehr als eine Milliarde monatliche Nutzer erreicht und sich die Zahl der dort gestellten Anfragen seit dem Start in jedem Quartal mehr als verdoppelt hat. Gleichzeitig werden Suchanfragen komplexer und stärker in natürlicher Sprache formuliert.1

Was bedeutet das für Suchfunktionen in Unternehmen, Shops und digitalen Anwendungen?
Was ist Natural Language Understanding?

Natural Language Understanding bezeichnet Verfahren, mit denen Systeme die Bedeutung menschlicher Sprache erfassen.

Bei einer klassischen Suche steht häufig der Abgleich von Begriffen im Mittelpunkt. Gibt ein Nutzer „Notebook 16 GB günstig“ ein, lassen sich diese Begriffe relativ direkt mit Produktdaten vergleichen.

Komplexer wird eine Anfrage wie:

„Ich brauche einen leichten Laptop für Geschäftsreisen, der einen Arbeitstag ohne Laden durchhält.“

Hier muss die Suche mehrere Informationen erkennen:

Gesucht wird ein Laptop.
Mobilität ist wichtig.
Ein geringes Gewicht ist wahrscheinlich relevant.
Die Akkulaufzeit spielt eine zentrale Rolle.
Der Nutzungskontext sind Geschäftsreisen.

NLU hilft dabei, solche Zusammenhänge aus der Sprache herauszuarbeiten.

Das Ziel besteht nicht darin, jedes Wort isoliert zu interpretieren. Entscheidend ist die Bedeutung der Anfrage als Ganzes.
NLU ist mehr als Synonymerkennung

Natural Language Understanding wird häufig mit semantischer Suche gleichgesetzt. Beide Technologien hängen zusammen, erfüllen aber unterschiedliche Aufgaben.

Semantische Suche konzentriert sich darauf, Inhalte zu finden, die inhaltlich zu einer Anfrage passen.

NLU setzt früher an. Es hilft dabei zu verstehen, was in der Anfrage überhaupt ausgedrückt wird.

Dazu können unter anderem gehören:

Entitäten:
Produkte, Personen, Orte, Unternehmen oder andere eindeutig erkennbare Objekte.

Eigenschaften:
Größe, Farbe, Zeitraum, Preis oder technische Merkmale.

Beziehungen:
Beispielsweise „kompatibel mit“, „geeignet für“ oder „Alternative zu“.

Absichten:
Möchte jemand etwas kaufen, vergleichen, erklären lassen oder ein Problem lösen?

Kontext:
Welche Informationen aus der aktuellen Anfrage oder einem vorherigen Dialog sind relevant?

Erst auf dieser Grundlage kann eine Suchplattform entscheiden, welche Datenquellen, Filter und Rankingverfahren sinnvoll sind.
Warum natürliche Sprache für Enterprise Search relevant wird

Auch innerhalb von Unternehmen kennen Mitarbeitende häufig nicht die genaue Bezeichnung eines Dokuments.

Sie suchen beispielsweise nicht nach:

„HR Policy 4.3 Mobile Work International“

sondern fragen:

„Wie lange darf ich aus dem Ausland arbeiten?“

Für die Person ist die Frage eindeutig. Für eine klassische Keyword Search können jedoch entscheidende Begriffe fehlen.

Das Problem wird größer, wenn Unternehmenswissen über verschiedene Fachabteilungen hinweg entstanden ist. Ein Fachbereich spricht von „Mobilarbeit“, ein anderer von „Remote Work“ und ein dritter von „ortsunabhängigem Arbeiten“.

Natural Language Understanding hilft, die Sprache des Nutzers von der internen Bezeichnung der Information zu entkoppeln.

Wie relevant solche Interfaces inzwischen auch für strukturierte Unternehmensdaten werden, zeigt eine 2026 veröffentlichte Untersuchung zu Natural Language Search in Unternehmensdatenbanken. Dort konnten Nutzer komplexe Fragen in natürlicher Sprache stellen, statt Datenbankstrukturen oder Abfragesprachen kennen zu müssen. Die Autoren identifizierten dabei insbesondere die Vollständigkeit der zugrunde liegenden Daten als begrenzenden Faktor.2

Die technische Qualität der Sprachverarbeitung allein reicht also nicht aus. Das System muss auch auf die richtigen Informationen zugreifen können.
Was passiert mit einer natürlichen Suchanfrage?

Eine NLU gestützte Suche kann eine Anfrage in mehrere Verarbeitungsschritte zerlegen.

Nehmen wir die Frage:

„Welche Ersatzteile brauche ich für die Wartung von Modell X nach 5.000 Betriebsstunden?“

Das System kann daraus erkennen:
Bestandteil Erkannte Information
Suchziel benötigte Ersatzteile
Produkt Modell X
Kontext Wartung
Bedingung 5.000 Betriebsstunden
erwartetes Ergebnis passende Teile oder Wartungsdokumentation

Anschließend kann die Suche gezielt relevante Datenquellen und Dokumente berücksichtigen.

Bei noch komplexeren Fragen kann auch eine automatische Umformulierung der Anfrage sinnvoll sein. Aktuelle Verfahren zeigen, dass generative Modelle aus natürlich formulierten Fragen explizite Suchspezifikationen erzeugen können, die anschließend von klassischen Retrievalsystemen verarbeitet werden. Dadurch lassen sich moderne Sprachverarbeitung und skalierbare Suchtechnologien miteinander verbinden.3

Wo Natural Language Understanding besonders hilfreich ist

NLU schafft vor allem dort Mehrwert, wo Nutzer ihre Anfrage nicht zuverlässig in wenige Keywords übersetzen können.

Typische Einsatzgebiete sind:

Produktsuchen mit beratungsintensiven Sortimenten
interne Wissenssuche
technische Dokumentationen
Kundenservice und Self Service
Chatbots
mehrstufige Suchdialoge
Suche in Fachanwendungen

Ein Ersatzteilkatalog mit Artikelnummern benötigt beispielsweise weiterhin eine präzise lexikalische Suche. Ein Serviceportal, in dem Nutzer Fehler mit eigenen Worten beschreiben, profitiert dagegen stark von Sprachverständnis.

Beides kann innerhalb derselben Suchanwendung erforderlich sein.
NLU macht schlechte Daten nicht automatisch gut

Natural Language Understanding löst nicht jedes Suchproblem.

Wenn Produktinformationen fehlen, Dokumente veraltet sind oder Metadaten widersprüchlich gepflegt wurden, kann das System aus einer gut verstandenen Frage trotzdem keine zuverlässige Antwort ableiten.

Ebenso sollte nicht jede Interpretation automatisch akzeptiert werden. Besonders bei mehrdeutigen Anfragen kann eine Rückfrage sinnvoller sein als eine falsche Annahme.

Aus:

„Apple Adapter“

könnten beispielsweise mehrere Absichten entstehen.

Meint der Nutzer einen Adapter für ein Apple Gerät? Einen bestimmten Apple Adapter? Einen Adapter für ein bestimmtes MacBook Modell?

Gute NLU erkennt daher nicht nur Bedeutung, sondern auch Unsicherheit.
Fazit: Gute Suche beginnt mit dem Verständnis der Anfrage

Natural Language Understanding verändert die Schnittstelle zwischen Nutzer und Suchmaschine.

Menschen müssen weniger darüber nachdenken, welche Begriffe ein System erwartet. Stattdessen können sie zunehmend beschreiben, was sie tatsächlich benötigen.

Für Unternehmen bedeutet das jedoch nicht, Keyword Search abzuschaffen. Der größte Nutzen entsteht, wenn Sprachverständnis mit bestehenden Suchverfahren, strukturierten Daten und einem passenden Ranking kombiniert wird.
Natural Language Understanding mit Quasiris testen

Die Quasiris Search Cloud unterstützt Natural Language Understanding für komplexe Suchanfragen und lässt sich mit weiteren Suchverfahren und bestehenden Datenquellen kombinieren.

Sprechen Sie mit unseren Search Experten darüber, wie natürlich formulierte Anfragen in Ihrem konkreten Anwendungsfall verarbeitet werden können.

Welche AI Features dabei den größten Mehrwert schaffen können, zeigen wir Ihnen zudem in unserem Überblick zu den AI Features der Quasiris Search Cloud.

Quellen

1 Reid, E. (2026). Eine neue Ära der KI-Suche. Google. https://blog.google/intl/de-de/produkte/suchen-entdecken/google-suche-io-2026

2 Jain, S., Tripathi, S., Qiao, S., & Jindal, A. (2026). Tursio database search: How far are we from ChatGPT? arXiv. https://doi.org/10.48550/arXiv.2603.18835

3 Satouf, A., Zong, Y., Amadou Boubacar, H., Piantanida, P., & Piwowarski, B. (2026). QueStER: Query specification for generative keyword-based retrieval. In Findings of the Association for Computational Linguistics: EACL 2026 (pp. 5957–5968). Association for Computational Linguistics. https://doi.org/10.18653/v1/2026.findings-eacl.312