A glossary controls terms whose translation should be consistent or intentionally unchanged: product names, feature labels, industry language, regulated phrases, and words with a specific brand preference. Each entry should include the source term, approved target value, language pair, context, ownership, and notes about capitalization or forbidden alternatives.

Do not add every common word. Oversized glossaries create conflicts and make review harder. Prioritize terms that affect meaning, navigation, trust, or search intent, and retire entries that are no longer used.

Translation memory reuses approved segments when the source and policy still match. A safe identity includes the site or tenant, language pair, normalized source, content hash, and relevant translation policy. Reusing a segment globally by source text alone can leak one customer's wording into another customer's site or apply the wrong meaning in a different context.

Track provider, model, review state, creation time, and last use. When glossary rules or protected-term policies change, include a policy version so stale values are not silently reused.

Conflicts appear when the same source has multiple approved translations, when a glossary changes, or when source content is edited slightly. Show reviewers the alternatives, context, usage locations, and recency. Define whether a decision updates one occurrence, a page, a language, or the shared memory.

Keep an audit trail and avoid destructive bulk replacement without a preview. A terminology correction can improve consistency, but it can also change product names inside URLs, code, or customer-authored content if scope is too broad.

Review new terms before major launches, sample high-use memory entries, and monitor reviewer overrides. Repeated corrections often reveal an unclear source sentence, missing context, or a glossary rule that is too general. Fixing the source can be more valuable than adding another exception.