Here’s a small thing that quietly wrecks a translation project: you translate “dashboard” on Monday, and by file 30 on Thursday you’ve called it three different things. Nobody catches it until the client does. Then you’re doing an unpaid sweep through 30 files to make them agree.
A glossary fixes that — a simple list of “this source term always becomes this target term.” Translators have used these (termbases, in the trade) forever. The new part is that ChatGPT can build one and enforce it across a whole project in about five minutes. Done right, a proper termbase cuts post-editing time by 20–40%, mostly because you stop fixing the same word over and over. Here’s the workflow, with the exact prompts.
Step 1: Pull the key terms out
Don’t write the glossary by hand. Hand ChatGPT a representative chunk of your source — the first file, or a few pages — and have it find the terms that need to stay consistent.
You are a terminology manager. From the source text below, extract every
term that should stay consistent across a translation project: product
names, technical terms, recurring phrases, UI labels, and brand words.
Skip ordinary words. Return a table: Source term | Type | Notes on meaning/context.
Source text:
[paste your text]
You’ll get a clean list of candidates. This alone surfaces terms you’d have translated on autopilot.
Step 2: Lock in your target terms
Now you make the calls — because you decide what’s right, not the machine. Add the target column, and overrule anything you disagree with.
For each source term below, propose the best [target language] equivalent
for [domain: e.g., fintech UI, informal tone]. Keep these in English: [list
any do-not-translate brand terms]. Return: Source | Target | Keep in English? (Y/N).
Maybe you decide “dashboard” stays English but “settings” gets translated. Fine. Edit the table until it’s your glossary. This is the part that needs a human, and it’s quick.
Step 3: Translate with the glossary locked in
Here’s the move that makes the glossary actually do something. For every file, paste the glossary into the prompt and tell ChatGPT to obey it.
Translate the text below into [target language]. You MUST use these exact
term mappings every time they appear — no synonyms, no variations:
[paste your Source | Target glossary]
Keep tone [formal/informal]. Translate everything else naturally.
Text to translate:
[paste file]
Now “dashboard” comes out the same in file 1 and file 30. Every time.
Step 4: Run a consistency sweep at the end
Even with a locked glossary, drift sneaks in across a big job — especially if more than one person touched it. So before you deliver, run a sweep.
Below are translated segments from several files. Find every case where the
same source term was translated differently across them. Output a table:
Term | Variants found | Recommended single version | Files affected.
Also flag inconsistent capitalization and formatting.
[paste segments]
You get a clean discrepancy report — fix list in hand, no eyeballing 30 files at midnight.
What this means for you
If you’re a solo freelancer. This is the highest-leverage habit you can add. The glossary turns a “hope I stayed consistent” project into a “guaranteed consistent” one, and consistency is exactly what separates a pro deliverable from a machine dump.
If you work in a team. Share one approved glossary before anyone starts. It’s the cheapest way to stop three translators from inventing three words for the same thing.
If you localize software or products. Your UI labels and product terms are the whole ballgame. A locked termbase keeps “Sign in” from becoming “Log in,” “Enter,” and “Access” across screens.
If you only translate occasionally. Even then — for any job over a few thousand words, ten minutes of glossary work saves you an hour of rework. The math always wins.
What this won’t do
Be clear-eyed about the limits:
- ChatGPT isn’t a translation memory system. It won’t remember your glossary next month or across separate releases. For that you still need a real CAT tool or TMS — the glossary lives in your file, and you paste it in each time.
- The term decisions are yours. AI proposes; you approve. Rubber-stamping its suggestions defeats the point.
- Long files lose the thread. On very large texts, re-paste the glossary periodically so it stays “in mind.” Don’t assume one mention at the top holds for 10,000 words.
- Confidential work needs care. Don’t paste sensitive client material into a consumer tool without checking your terms and anonymizing where you can.
The bottom line
A glossary is the least glamorous, highest-return habit in AI-assisted translation. Five minutes up front, consistent terms across the whole project, and a clean sweep before delivery. It’s also the difference between selling “I ran it through AI” and selling “I delivered a controlled, consistent translation” — and only one of those holds its rate.
This pairs with the bigger picture in using ChatGPT as a translator without losing work. And to build the full post-editing and glossary workflow as a system, the AI for Translators course walks you through it step by step.
