Help center Core configuration Knowledge base Configure preferred terminology for AI and search Updated Thursday, November 5, 2026 · Marc Staine AI Terms describes the wording, aliases and definitions your organization uses. Focused term lists help supported search and AI features recognize customer language and use consistent preferred terminology. Applies to: Supportbench Core. Location: Settings → Knowledge Base → AI Terms. Access and available controls depend on your role and account capabilities. Before you begin You need Administer AI Terms. Availability and editing admission are checked for the account and role.Choose a preferred expression and real alternatives people use. Shared terminology is separate from an article’s Search keywords or tag audience.Reopen the saved result and test a representative consuming workflow. A successful save alone does not verify every reader, assignment or rendered result. Configure and verify Open Settings → Knowledge Base → AI Terms. On Word lists, choose New Word List or open an existing list.Enter list Name/Description and review Divisions and Active. Empty Division selection means all applicable Divisions. Choose Save Word List and reopen it.Open Terms and choose New Term. Enter Use this wording, Other wording and the relevant short-form, broader-topic or definition values described in the field reference.Review the meaning, then choose Save Term. A proposed term is not stored merely because it appears in an AI suggestion.Use Test phrase with realistic customer wording after saving. The test reads saved terms; it does not test unsaved form edits.If using AI Assist, inspect each suggestion and deliberately save the useful terms. Review the supported import inputs before importing a larger vocabulary.Retest the intended search/AI workflow and record what was observed. A glossary entry alone does not prove every search surface or external AI index will retrieve an article. Settings reference Settings and actionsSettingWhat it controlsHow to chooseWord list / New listSelects or creates a terminology collection.Group terms by product or support domain.Name / DescriptionList identity, up to 150/500 characters.Explain the list’s purpose and ownership.DivisionsLimits the list to selected divisions; empty means all divisions.Avoid overlapping lists with conflicting definitions.ActiveMakes a saved list available to supported consumers.Activate reviewed terminology.Preferred termCanonical wording, up to 250 characters.Use the exact wording your team wants in answers.Also known asAliases separated by comma, semicolon, pipe or line break.Include actual customer phrases and common alternative names.Short form / Broader topicAcronym and grouping context, each up to 250 characters.Use context to distinguish ambiguous abbreviations.DefinitionConcise explanation, up to 2,000 characters.State distinguishing meaning instead of marketing claims.Term search / New / Edit / Delete selectedMaintains the list’s saved terms.Review selected rows before deleting.Test phraseTests against saved terminology.Save changes first and test a realistic customer phrase.Import file name / Spreadsheet contentPaste CSV/TSV content with a header; preview and import mapped rows.Use the template and review mappings/errors before import. Limits are 2 MiB and 1,000 data rows.AI Assist source text / SuggestProposes terms from supplied source text. Suggestions need review and save.Supply approved product terminology and check every proposed definition. Behavior to knowAn unavailable-storage warning means the feature is not ready in that environment. Do not interpret an empty list as proof terminology is configured.AI suggestions are drafts; they are not saved automatically. Recommended use Prefer terms grounded in customer and agent language, including useful product aliases.Avoid ambiguous mappings that collapse distinct products or error conditions into one term.Keep a vocabulary owner and review actual retrieval examples after changes. Worked example: distinguish two product tasks that customers describe similarly A team supports a Gateway product and a Controller product. Customers call both tasks “export settings,” but the procedures are different. Choose a product-specific Word List and the intended Divisions instead of applying ambiguous terminology everywhere.Create a preferred term such as Gateway configuration export and add real customer phrases such as download gateway settings. Write a definition explaining the product and task. Create a distinct Controller configuration export term when it describes a different task; do not assign the same ambiguous alias to both.Save the Word List and each Term. Use Test phrase with the saved gateway wording, then the controller wording. Inspect the matching result before adding more aliases; testing does not use unsaved suggestions.Test the actual supported search/AI workflow with the intended reader language and access. Confirm that it finds the correct complete article and not merely the right term. Test a short follow-up separately when the customer conversation depends on prior context.If the wrong article remains prominent, investigate its content, audience and the particular search surface. Do not add unrelated synonyms or duplicate articles to force a match. AI Terms, article Search keywords and Tags serve different purposes. Troubleshooting ProblemWhat to checkTest phrase ignores a new entrySave the Word List/Term first; testing uses saved terms.A suggestion disappearedAI Assist suggestions require review and an explicit Save Term action.The list applies more broadly than intendedAn empty Divisions selection means all applicable Divisions.A term does not fix a particular search resultCheck the actual search surface, article content/audience and supported matching behavior; terminology is only one input.I cannot edit the vocabularyReview AI Terms editing access and account availability. Related configuration Knowledge Base → Organize knowledge-base topics and accessKnowledge Base → Configure article templates and AI authoring guidanceKnowledge Base → Manage knowledge-base tags and audiencesKnowledge Base → Configure shared article styles and calloutsUse Supportbench AI in daily B2B support: choose the right tool and verify the resultMake Supportbench knowledge useful and discoverable in search and AI answers Test search relevance before changing terminology Use Test article search to investigate a missing or poorly ranked article. This is a read-only test of the Knowledge Base articles available to your current agent account. You need access to AI Terms settings and to the Knowledge Base. Settings access alone does not grant permission to read an article. The article search preview uses all saved terms applicable to the search. Test phrase can test the selected word list. Unsaved edits are not part of either test. Open Settings → Knowledge Base → AI Terms. Enter the customer's exact wording in the existing test input. Use 2–400 characters.Use Test phrase when you only want to check which saved terms recognize the wording. A recognized term does not, by itself, prove an article matched.Expand Test article search. Leave Keyword priority at Current ranking and select Test article search.Review up to 20 ranked results and the total matches. Record the expected article ID and position. Applied search concepts show accepted equivalents and any spelling alternatives. Each result identifies actual title, Search keyword, indexed body, equivalent wording, complete-phrase or intent evidence. These are explanations of matching evidence, not confidence percentages or proof that the article answers the question.Choose Emphasize keywords — preview only and run the same query again. Compare the expected article and unrelated results with the current order. This increases authored keyword evidence while retaining concept coverage, complete-phrase and intent priorities. It ranks the whole admitted population before selecting the first 20, so an article outside the original first page can appear.Repeat with short phrases and complete questions. Changing the input or mode clears the previous preview; an older response cannot overwrite the current test. This control does not save an account policy, edit keywords or call an AI provider. For example, save drop-in with the aliases drop in, dropin and dropins. Test “how can i offer drop in classes”, “drop in classes”, “drop in” and “dropin”. The configured compound and plural forms should preserve the drop-in concept. A fuller question retains meaningful actions such as offer or schedule, numbers and negative intent, so its relative order can differ appropriately. Do not map offer, enroll and schedule to one synonym merely to force identical rankings. A deliberately authored Search keyword is a short phrase describing an actual use of an article. An exact complete phrase is stronger evidence than incidental shared words. Add a genuine phrase such as “on the list for more than one class” only if that article answers it; “on the list” alone is ambiguous. Edit keywords through the article's normal authorized authoring workflow, then repeat the tests after search indexing catches up. This preview does not publish content or bypass the existing keyword-edit permission. If keyword emphasis is unavailable, the test shows current ranking and says so. Supported plain English queries use typed relevance; exact article IDs, advanced syntax and unsupported languages retain their existing search path. Missing keyword storage or indexing also disables the keyword comparison. Do not interpret no matches as evidence that an article does not exist: check access, status, the exact wording and indexing separately. A failed lookup is reported as an error, not a successful empty list. The preview is for your agent access. It does not impersonate a Help Center visitor. Compare public results using the same portal, language, topic and visitor permissions. Suggestions and AI replies also require usable, eligible source content, so fewer results can be correct. Keep a test record with query, article ID, mode, actor/portal scope and observed order before requesting an account-wide tuning change. Was this article helpful? Yes No Thank you for your feedback.