*Illustration of terminology management; not the DeepL interface and not a test result.*
DeepL Glossaries define the preferred translation of a company’s words and short phrases. It helps standardize terminology in instructions, product cards, and correspondence. To get started, all you need is a consistent pair of “source term — translation” and the correct language direction. A finished glossary can be created manually or imported; an archive of previous translations requires a different tool — the glossary generator. How DeepL defines a glossary.
FAQ for editors, support managers, and localization owners. Documentation checked October 28, 2026. No DeepL login, import, or translation was performed: the steps below are based on the official help documentation, and the examples were created by the editorial team. Your organization’s plan, permissions, and service availability should be confirmed separately. Use only materials your company is allowed to share with the selected service.
Question navigation
- Glossary or translation memory?
- Which terms should be included?
- How do you create a glossary manually?
- What CSV file should be prepared?
- Why does the generator require a different file?
- How do you apply a glossary to a document?
- Why isn’t it visible in the API?
How is a glossary different from translation memory?
A glossary locks in the translation of a term, while translation memory stores previously translated text segments. If the dispute is about a part name, start with the glossary. If you need to reuse approved sentence-level phrasing, consider translation memory. DeepL describes it as a way to store and reuse translations uploaded in TMX format. Description of translation memory.
| Editor’s task | Approach | Sample input |
|---|---|---|
| Lock in a component name | Glossary | control cabinet → control cabinet |
| Repeat approved sentences | Translation memory | Source segment and approved translation |
| Extract terms from previous translations | Glossary generator | Matched source and translated texts |
| Choose the meaning of a polysemous word | Editorial decision first | A sentence with context and expert approval |
The table below shows suggested real-world scenarios. For example, a supplier manual may include both a standard warning in full and the name of a specific component. In that case, both mechanisms are needed. When used together, DeepL gives glossary entries priority over translation memory — so an outdated terminology entry can also affect an established phrase. Glossary priority.
Which company terms should be added to a glossary?
Start with product names, industry terms, and internal terminology that already has an agreed translation. DeepL recommends collecting these words together with technical, sales, and marketing teams. Candidates also include terms that should not be translated. Guidelines for collecting terms.
Editorial example: for equipment documentation, the specialist approved control cabinet → control cabinet, and for the service interface — work order → work order. Save the source sentence, area of use, and term owner alongside the working glossary. This information helps the team choose the right meaning; it is not included in the import file for ready-made terms.
Do not add every word from an instruction manual to the glossary. If one team understands lead as a potential customer and another as a material, separate the use cases first. Otherwise, a single entry will lock in the wrong meaning. This is a suggested editorial way to organize glossaries, not a promise of automatic industry recognition.
Can you assign multiple translations and set the reverse direction at the same time?
For one source entry, you need to choose one preferred translation; the reverse direction is created separately. DeepL’s help documentation requires entries to be unique. When adding terms manually, you can select Add term for both language directions, which creates an additional entry with the languages and terms reversed. Rules for creating entries.
Do not enter control cabinet / control panel in the target cell if the editor has not made a choice yet. That string does not express an unambiguous rule. The suggested solution is to agree on a more precise source expression or separate the glossaries by subject area.
The reverse pair should also be evaluated for meaning. An English term may have several Russian equivalents in different documents; mechanically swapping the columns does not solve this problem. A successful editorial outcome is an agreed meaning for the needed direction, not just the presence of two entries.
How do you create a small glossary manually?
Open Glossaries, choose manual creation, and enter the source term with the approved translation. According to the web version documentation, the path is: Glossaries → +New glossary → Create a glossary manually → Continue. Then set the name, language pair, the Source and Target values, add additional pairs if needed, and click Create glossary. Documented workflow.
For example, name the glossary Equipment Instructions EN-RU and enter the approved component name. After saving, verify the glossary name, direction, and entry content. Then translate an approved sentence containing this term and read it in full: having an entry does not yet confirm the quality of the final wording.
Manual creation is convenient for a short list of disputed terms. For an already prepared table, use import if your plan supports it. The help page for creation and the separate import page list different plans, so you need to verify the button and upload access in the specific account; you cannot promise import to every user.
How do you prepare a CSV for importing an existing glossary?
For a single language pair, prepare a UTF-8 CSV with two columns and put the first terminology entry in the very first row. It must not contain comments or duplicate source text. In the import window, check the detected delimiter and language pair, then confirm the upload. Import requirements.
Below is a sample training file for EN→RU. Its CSV syntax has been checked locally; upload to DeepL and translation quality have not been verified.
control cabinet, control cabinet
work order, work orderSave this block as UTF-8 CSV without a header row. To open the correct mode, use Glossaries → +New glossary → Upload an existing glossary, enter a name, and choose a file. Signs of correct preparation: two terminology pairs in two columns, readable Cyrillic, and no system row among the terms.
For multiple language pairs, DeepL describes four columns: source term, translation, source language code, target language code. Do not add "department" or "comment" columns to a two-column file for internal tracking: keep the expanded editorial table separate from the import file.
Why does the glossary generator require a different CSV?
Importing ready-made terms and generating a glossary from previous translations use different data. The generator extracts candidates from parallel texts; for its CSV/TSV, DeepL recommends the first row Source,Target or the corresponding tab-delimited version. For importing a ready-made glossary, the first row is already a terminology pair. Generator data structure, ready-made glossary structure.
That is why the "my CSV is correct" error often means you first need to clarify the file’s purpose. In one case, the company already has approved terms. In the other, it provides source and translated sentences to generate terminology suggestions.
Example for discussion with an editor: if there is no approved translation for work order yet, do not upload a random equivalent as a required rule. Use high-quality aligned texts and consider generating candidates. Suggested entries should be reviewed, corrected, or deleted before saving; DeepL provides for this review step. Reviewing generated entries.
Do you need to add all case forms and plural forms?
According to DeepL, there is no need to list every inflected form separately: entries adapt to the grammar of the target language. The glossary does not work as a simple character-by-character substitution. This is a property of the mechanism, not a guarantee that every sentence will be correct. Grammatical adaptation of entries.
For testing, we suggest using several approved sentences with different contexts for the term: the name of the component on its own, a phrase about installing it, and a phrase about maintaining it. Compare meaning, agreement, and terminology. This is an editorial example; no results are provided here.
If a separate phrase sounds unnatural, go back to the accuracy of the source term and the length of the expression. Do not expand the glossary with dozens of case-form duplicates before identifying the cause. The problem may be in the surrounding part of the sentence or in an ambiguity in the source text that a terminology rule does not resolve.
How do you apply a glossary to document translation?
Select the file, set the language pair, and turn on the desired glossary in the file translation panel. DeepL support explicitly requires the source and target languages to match the glossary languages. Selection applies to document translation; the mere existence of a glossary does not mean it is selected in the current task. Using glossaries with files.
Recommended workflow for an employee: take an approved short instruction with a known term, open file translation, check the direction, choose the glossary, and run the translation. Then find the source term and read the corresponding part of the result. You need both the correct term and the preserved meaning of the instruction.
If the term remains unexpected, first check the selected glossary and languages. Then compare the actual spelling in the source text with the entry. A single bad example does not mean DeepL ignores glossaries altogether; but the fact that a glossary is selected does not replace human review of the result.
Why is the glossary from the web version not visible in the API?
Whether it can be used depends on the account type, glossary owner, and granted permissions; there is no universal rule for all subscriptions. The general help describes a special scenario for customers with API Pro and Translator: an admin with a Translator license can manage both types, and a shared team glossary is available to the API under the stated conditions. This does not mean automatic compatibility for any personal glossary. Ownership and access across products.
At the same time, the separate file translation guide describes a narrower path: use an API glossary, and download the web version glossary as CSV and upload it to the API glossary. It also requires passing glossary_id and source_lang with the appropriate language pair. Conditions for API file translation.
For an integrator, the next step is to determine exactly which account is making the request, who owns the glossary, and whether the API has access to it. Do not copy an ID from someone else’s profile at random. This article does not include a verified API request: resolving the ambiguity for a specific organization requires checking its settings and the current documentation for the relevant method.
How do you save a glossary to pass it to another employee?
Download the glossary through its management menu and pass along the CSV together with the language direction and intended use. DeepL documents export through Glossaries, the menu next to the desired glossary, and Download. The exported file lets you keep terminology entries outside the service interface. How to download a glossary.
A file copy and rights to a cloud object are different things. The recipient of the CSV receives a dataset, while access to an existing glossary depends on the account settings. For a working archive, we suggest saving the date, responsible editor, and glossary purpose next to the file, without including these fields in the import table.
If the task goes beyond terminology and you need to connect translation with document workflows, a separate overview of AI document translation and workflowsmay be useful. For preparing to switch services, see the material on moving off an AI service without losing the workflow.
How does AIrassvet connect terminology with workflow?
For regular translation, the path from the source document to an approved result matters. As part of data preparation and AI solution integration, the task may include collecting approved terms by subject area, preparing source materials, and connecting translation to a corporate workstation with delivery of the result to an editor. The specific service and rights to its glossaries are determined separately.
The first step is to choose one type of document, describe the current terminology fixes, the sources of approved translations, and the data transfer restrictions. That provides a factual basis for discussing integration. Discuss the task.
Where should you start with a glossary?
Take a recurring document and align its disputed terms with a subject-matter specialist. Create a small glossary manually, or import ready-made term pairs in the correct mode. Check the direction, glossary application, and the meaning of the translated passage. If the source material is an archive of parallel translations, use the candidate generator and editorial review, and for API integration, set the glossary owner and availability separately.
