Contents7
In short
A brand glossary lists the terms whose translation is required, language by language: product names, trade terms, banned words.
It is written before the first translation, validated by someone who knows the market, and checked after each batch.
DeepL, Google Cloud Translation and general-purpose AI models can all use one; the file itself must remain yours.
Imagine a product line called “Pro” on the home page, “Professional” in the menu and “Business” on the pricing page: in English, readers wonder whether these are three offers. Nobody would have decided it: three translations made at different times, by different tools or people, would simply have picked three words. A brand glossary exists so that this choice is made once, by you, before the first translation.
This guide shows what it contains, how to write it and how tools use it. It extends our articles on website translation and opening new markets.
What a brand glossary contains
Definition
Brand glossary
The list of terms whose translation is required, language by language: the words that nobody, translator or machine, is allowed to translate any other way.
A useful glossary fits in a simple table, one row per term, with a few columns:
- The source term, as it appears in your texts.
- Its required translation in each language, or the note “do not translate”.
- The status: required, banned (a word never to use, even if it looks right), or to avoid.
- The context: a short definition, or an example sentence, so that an ambiguous word is tied to the right meaning.
- The validation: who chose, and when.
Four families of terms come first. Brand and product names, which are not translated or are translated in one way only: in its documentation on Cloud Translation glossaries, Google takes the example of Google Home, which must stay Google Home. Trade terms, those of your quotes and contracts. Banned words, such as a competitor’s term or a phrasing that promises too much. And tone choices that are made once, such as the formal or informal “you” in languages that distinguish them. What changes from one country to another beyond the words (prices, formats, forms) is a matter of localization, see localize, don’t just translate.
What the glossary does not contain: whole sentences, which belong to the translation memory, and the style rules of a language, which belong to a style guide. Microsoft, for example, publishes localization style guides for more than a hundred languages, which it describes as collections of rules that define language and style conventions for specific languages, data formats included. The glossary fixes the words; the style guide fixes the manner.
Why translations become inconsistent
A website is rarely translated in one go. A first version by one translator, updates by another, product pages by an automatic tool, an article by an AI model: at each pass, a term can change, and nothing flags it.
The machine does not solve the problem on its own. The organisers of WMT, the annual evaluation campaign for machine translation, note in their 2025 findings that the best systems remain fragile on a field’s terminology:
Domain-specific evaluations […] demonstrate substantial difficulties in maintaining terminological accuracy and consistency.
A glossary does not make a tool more intelligent. It takes away a choice the tool should not have to make.
How tools use it
The three main families of tools accept a glossary, each in its own way:
- DeepL stores terms by language pair: according to its documentation on glossaries, a dictionary maps source phrases to target phrases for a single language pair, in one direction, and the tool flexes entries for case, gender or tense when the target language requires it.
- Google Cloud Translation defines it as a custom dictionary, loaded from a file and applied to the translations that reference it.
- A general-purpose AI model receives the glossary with its instructions, at each batch. There is no dedicated mechanism: the instruction must be clear, and compliance checked afterwards.
In all three cases, the glossary remains a table that belongs to you. DeepL accepts a CSV or TSV file; Google loads its own from a file. Keep the source file yourself: changing tools must not make you lose your choices. We compare the tools themselves, with no favourite, in DeepL, Google or an LLM: which translates your trade best.
The glossary in the translation workflow
- Writing
The terms and their translations, language by language.
- Validation
A person who knows the market decides.
- Translation
The tool receives the glossary with each batch.
- Checking
Every required term is present, every banned term absent.
- Correction
An error is fixed in the glossary, once and for all.
The check is the step people forget. An automated check is enough to spot a banned term or a wrongly translated product name across thousands of pages. Without it, the glossary remains an instruction you hope is followed. On a catalogue, this check adds to those on figures and units, detailed in translating product pages for export.
Writing the glossary in five steps
- List the candidates.
Product and range names, the terms that come back most often on the site, the words in your quotes and contracts, and the terms customers have already got wrong.
- Choose the translation with someone who knows the market.
A translator of the language, a local salesperson, a trusted customer: a term that is right in the dictionary can sound wrong in the country. Also check that it is the word people search for, see international SEO: choose your markets before translating.
- Mark the banned terms and the “do not translate” ones.
These rows often prevent the most visible mistakes.
- Have it validated, and date it.
One person decides, and their name is in the table: the glossary is not a list of suggestions.
- Connect it to the tool and to the check.
Loaded into the translation tool, verified after each batch.
Start short. A glossary of a few dozen terms that are respected does more good than a thousand-line list nobody reads. It then grows with corrections: every term error found in review becomes a row.
Institutions do the same at their scale. Since 2004, the EU institutions and bodies have shared IATE, their common terminology database, which covers fields ranging from law to medicine. A company does not need that scale, but it needs the same principle: a term decided once, shared by everyone who translates. Terminology is a trade and a service in its own right there: the Translation Centre for the Bodies of the EU charges, in its 2026 price list, 4 € per term in a term list and 945 € per day of terminology work.
From the field
Our pipeline rule sets the glossary before the first translation pass: brand names, trade terms, tones to avoid. On Palmora Property, a real-estate listings site in six languages that we operate, the pipeline combines that glossary, a translation memory and a parity check of the content keys across languages. There, the memory brings 95% of segments out of the cache after the first pass; the glossary makes sure what it serves again is right.
Who keeps the glossary up to date
A glossary lives: a product launches, a range is renamed, a term turns out to be misunderstood in a country. It needs an owner, on the company side, who accepts or rejects additions, and a simple channel so that translators and reviewers can flag a term to revisit.
Also write into the contract that it belongs to you. If your translations are entrusted to a provider, ask for the glossary with every delivery, in a format you can read and reuse, just like the translation memory. Translators remain the best placed to keep it alive: they are the ones who see, sentence after sentence, where a term resists.
Where to start
Open a spreadsheet, list your product names and the ten terms that come back most often in your offers, and have their translation validated for your first language. Then reread your already translated pages with that table in hand: the inconsistencies jump out. Next, translating your website with AI, automatic or reviewed says where to place the review, and website translation cost what it represents in the budget of a language.
To write a glossary and connect it to your translation workflow, describe your languages and your current tools: we start from the terms your customers use.
Frequently asked questions
What is the difference between a glossary and a translation memory?
The glossary fixes words, the memory keeps sentences. The glossary is written before the first translation and says how to translate a term; the memory fills up as translations are made and avoids retranslating a sentence already done. The two complement each other: without a glossary, the memory faithfully keeps inconsistent terms.
Should product names be translated?
It is decided name by name, and only once. A brand name most often stays as it is; a descriptive name can be translated if it needs to be understood or searched for in the language. What matters is that the decision is in the glossary, so that the same name is not translated on one page and kept on another.
How many terms should a glossary contain?
There is no right number. Start with product names and the terms that come back most often in the pages that sell, then add each term corrected in review. A short glossary that is respected and checked is worth more than a long list that is ignored.
Does a glossary help with search engine optimisation too?
Yes. The term chosen for a language should be the one people search for in the country, not just an accurate translation. Check it with a keyword research tool set to the country and the language. Once chosen, it goes into the glossary, and the translation uses it everywhere, titles and descriptions included.
Sources
- Its documentation on Cloud Translation glossaries (new tab)docs.cloud.google.com
- Microsoft, localization style guides (new tab)learn.microsoft.com
- Findings of the WMT25 General Machine Translation Shared Task (new tab)aclanthology.org
- DeepL, managing glossaries (new tab)developers.deepl.com
- IATE, terminology database of the European Union (new tab)iate.europa.eu
- Translation Centre for the Bodies of the EU, 2026 budget (applicable prices) (new tab)cdt.europa.eu
Findings of the WMT25 General Machine Translation Shared Task
Read on aclanthology.org(new tab)Translation Centre for the Bodies of the EU, 2026 budget (applicable prices)
Read on cdt.europa.eu(new tab)Jim · founder of Brain Plus AI
Also in translation and new markets
Understand7 min
Translation memory: what a company gains from it
A sentence translated once should not be paid for twice. The translation memory makes sure of that at every site update, provided it belongs to you.
Choose and price7 min
Translating your website with AI: automatic or reviewed?
Having everything reviewed is expensive, and reviewing nothing lets errors through in languages that nobody at your company reads. The split is decided page by page, depending on what each page commits you to.
Understand8 min
Localize, don't just translate: prices, proof and forms
A page can be perfectly translated and still feel foreign: a price in the wrong currency, a date read backwards, a form that rejects a name, a meeting with no time zone. Localizing means adapting what changes from one country to another, not just the words.