First, select only the terms that truly matter in your company's documents. According to the DeepL guide, the best starting point combines brand and product names, industry terms, terms important to the audience, and words that must not be translated.
According to locize, a glossary works best when it is a curated list of decisions rather than a loose table of words. Each concept should have an approved form, acceptable variants, and forbidden forms. In practice, this means the term list should be built from two sources at once: the source documentation and existing translations, because that is where inconsistencies are easiest to spot.
According to the Dictem study on terminology management, an entry should contain at least a definition, context of use, part of speech, forbidden synonyms, and the term's status. This matters especially for languages with rich inflection, because a equivalent alone is not enough if the tool is to maintain correct forms and style.
DeepL emphasizes that a glossary should work in a specific context and preserve grammatical correctness. locize, in turn, indicates that nouns and proper names are the best candidates, while verbs are weak candidates because MT systems struggle with their conjugation. Therefore, in practice it is worth recording:
According to M21Global, the process does not end with creating a list. It requires auditing existing resources, agreeing the glossary with internal teams, formalizing the delivery format, and continuous updating. This means the terminology owner should periodically check whether new products, features, and campaigns introduce terms that are not in the glossary.
DeepL also describes a practical working model in which glossaries, style rules, and translation memories can work together as a control layer at enterprise level. If a glossary is to support AI in business, it must be integrated into the tool the team uses every day, rather than kept in a separate file without an update process.
Dictem warns that overly large glossaries can worsen translation quality, so it is better to maintain a small core of the most important terms than to expand the database without selection. The same material also highlights the need to avoid duplicates and entry errors, because an MT system applies the correct term more easily when the database is clean and unambiguous.
The safest model, therefore, is a glossary that is:
Lub System helps B2B companies implement AI, automation and IT solutions end-to-end - from strategy to deployment. See our services or get in touch to discuss your case.