Site search for B2B: from a customer's question to a useful result
When does a B2B site need search, and how can you check it? Index scope, relevance, no-result states and tests based on real customer questions.
Article language: English

In this article
A customer types the name of a service and finds silence, or a list of unrelated articles. A search box alone does not make information findable. You need rules for what to include, a sensible order for results, and a way forward when nothing matches. Start with the questions customers actually ask. A magnifying glass is a symbol of search, not proof that an answer can be found.
First decide which tasks need search and which would benefit from a clearer menu. Then establish what may appear in results and what counts as a useful answer. This exercise helps you specify and assess the work without buying more technology than you need.
When does search do more than a better menu?
Describe the tasks that navigation makes difficult. A customer may know a model number, compare a dozen documents or return for a specific manual. Those are different problems from a menu item labelled “Solutions” that says too little. On a small site, a clearer label and better information architecture may be enough. The guide to B2B information architecture (in Polish) explains how to organise that structure.
Collect questions from sales, support and existing contact channels. Distinguish a search for a known item from exploration. “X12 manual” needs a precise match; “how to reduce downtime” might lead to a service, a guide and a conversation. In both cases the visitor should understand why a result appeared.
Base the decision on how often these tasks arise, the size of the collection and the capacity to maintain the index: the organised set of material the search tool examines. Assign an owner before choosing a tool. Without one, a search box may simply reveal disorder faster. Plan who is responsible for keeping the documents current.
What belongs in results, and what must stay out?
List the material types: services, products, articles, manuals and downloads. For each, record its source, language, publication status and owner. Drafts, test copies and documents requiring permission should stay out of the public index. An obscure filename is not access control.
Agree what happens when material is withdrawn. An old price list should not remain searchable merely because its file is still on the server. Record when a publication change reaches the index and how to check that the new result is visible. Updates may happen during a site build or through a separate process.
Consider PDFs too. If indexed, identify the format and link to the correct version. When the text cannot be extracted, a manually maintained description may be more useful than another algorithm. The publisher remains responsible for keeping the material current in either case.
How do you test relevance?
Choose a set of real queries covering important tasks. There is no universal number that guarantees quality. For each query, record the expected destination, acceptable alternatives and a misleading result. Include questions for which the collection has no correct answer.
Use product names, everyday terms, abbreviations, accents and common misspellings. Before adding synonyms, check their meaning in the industry: words that sound similar to a site author can describe different services to a customer. A subject owner should approve the mapping. Include local spelling conventions, including Polish diacritics when Polish material is in scope.
Do not optimise only for the test. If a title needs rewriting so a person can understand a result, improve the destination page too. Keep the query set after launch and rerun it when content, the engine or ranking rules change. The guide to writing a service page (in Polish) can help with that revision.
How should a result explain what it opens?
A result card needs a clear title, a short excerpt and the type of material. A technical catalogue may also need a model or document version. A date helps when it changes the usefulness of the answer; not every page needs every field. Choose according to the decision the visitor must make. If categories can be filtered, test whether the remaining scope stays clear after each change. Check filters and result counts alongside the search results. See also the article on filters and result counts (in Polish).
Set an order for services and guides. A specific service query should make its scope easy to reach; an educational question may justify a manual first. Paid placement or manual promotion needs a clear editorial rule so it does not quietly undermine relevance.
Inspect long titles and excerpts on a phone: truncation can hide the difference between product versions. Name links after the material instead of repeating “Learn more”. The destination should deliver what the result promised. The B2B service-page layout helps connect that promise to the destination. For related guidance, see the article on B2B service-page layout (in Polish).
No results, loading or failure?
These are three different states and need different messages. Saying “No results” during an outage suggests the company has nothing to offer. Asking someone to retry an accurate but overly narrow query is not much help either.

Keep the query visible on an empty list. Suggest another term, removal of a filter or a relevant section. Contact can be a further route, but should not be the only response to every empty search. Show active filters and make them easy to clear.
Test an empty submission, a very long query and rapid successive searches. An older response should not replace a newer one. On a slow connection the visitor must retain control of the field. Assess page speed separately: a heavy search feature can affect even those who do not use it. See the Polish Signal article “page speed (in Polish)” for related guidance.
Keyboard and screen-reader checks
Give the field a name and the button a clear purpose. A magnifying-glass icon alone does not explain whether the search covers the whole site, a catalogue or one section. Distinguish the labels if there are two search areas. The W3C search-landmark example explains this distinction: https://www.w3.org/WAI/ARIA/apg/patterns/landmarks/examples/search.html
Complete the journey with a keyboard: open the field, enter and submit a query, read its status, choose a result and return. Keep focus visible after the list changes and do not move it unexpectedly while someone types.
A suggestion list adds obligations. If it cannot be operated correctly, start with a simple form and a results page. A successful check helps assess basic website accessibility (in Polish), but does not prove the whole site conforms to an accessibility standard. If suggestions select a specific item, design a clear distinction between typed text, a suggestion and a committed record (in Polish).
A hypothetical acceptance exercise
Imagine a company sells equipment and services older models. A customer enters the code of a discontinued product. The team expects the current service manual for that model first and information about its successor second. The successor's product page cannot replace the document needed for a repair.
A similarly numbered document appears instead. The catalogue owner finds that search ignores a hyphen and groups model names too broadly. The team narrows that rule while keeping typo tolerance in descriptions, then reruns the whole query set to check for regressions.
It records the decision and adds the troublesome query to recurring tests. In this example acceptance requires the correct document at its real address, a clear version indication and a working return to results. These are conditions of the example, not a universal relevance threshold.
What should you measure after launch?
If measurement is appropriately implemented, examine categories of queries, empty results and next steps. Do not collect full queries containing personal data or case numbers without a need. Define access and retention; aggregated problem categories may be enough to improve the vocabulary.
Many empty results may indicate missing content, different customer terminology or a faulty index. Review examples and separate those causes before changing the algorithm. Frequent clicks do not prove a problem was solved: the visitor may have returned to search again.
Set a review rhythm and name a decision-maker. A renamed service, a new product range or withdrawn documentation should trigger an additional check. Search quality depends on maintaining the collection as well as the field itself.
Agreeing scope and ongoing cost
Before seeking a quote, describe three possible approaches: titles and descriptions for a small catalogue; full text plus a synonym dictionary; or several sources with ranking rules and separate administration. Choose based on important test questions that the simpler option cannot handle, not the number of features.
Ask what happens when an index or query limit is reached: missing results, delay or extra cost? Establish how rules and dictionaries can be exported so the company can continue after changing suppliers. Separate the cost of launching the search from the ongoing cost of maintaining its content. Even a simple mechanism needs decisions about which documents remain current and what to call a new product.
For each ongoing task, name an owner, access and a checking point. If a supplier offers a problem report, agree its contents using a concrete example rather than accepting the vague promise of “analytics”. A sensible first stage is the smallest option that passes the important agreed scenarios. Make further expansion dependent on the gaps you have identified and your ability to maintain it. Record why it was chosen. A missing service description may call for editing rather than another technical layer.
Missing answers can reveal a gap in an FAQ based on customer questions (in Polish), but first check whether the material is absent or merely hard to find.
Acceptance checklist
- The index has an owner and excludes drafts and private material.
- Test queries include expected destinations, misspellings and questions with no answer.
- Results distinguish services, articles and documents and show the needed version.
- Loading, errors and no matches have different messages.
- The form works on a phone and by keyboard, including the return from a result.
- Changes and withdrawals have been checked at the real public address.
Include this exercise in the acceptance of the whole site. The website acceptance checklist (in Polish) can help structure that wider review. To design search alongside a catalogue and offer, explore Aeneia Design.







