6 min read
This is an illustrative story. Lucía is a composite based on common situations freelance translators describe, not a real individual, and the numbers are realistic examples rather than reported results.
It is the second week of November, and Lucía has a spreadsheet open with 412 rows. Each row is a product from a small online homeware shop: a name, a short description, a care note. The shop wants all of it in Spanish before its holiday campaign goes live in Mexico and Spain. Lucía is a one-person translation business, and this is the biggest job of her year.
She has done jobs like this before, and she knows where they go wrong. It is not the hard sentences. It is row 37 calling a fabric “lino” and row 288 calling the same fabric “lienzo,” because she translated them nine days apart. The client notices. Customers notice. A store that cannot name its own products consistently looks careless.
If you sell, write or translate anything at volume, her fix will be useful to you too. Let’s walk through it step by step.
The problem was memory, not skill
Lucía is a very good translator. Her issue was that a 412-row catalog is too long to hold in your head. Over two weeks, small inconsistencies creep in: a color called “arena” in one place and “beige arena” in another, a care instruction phrased three different ways, a product line name translated when it should have stayed in English.
Professional translators usually solve this with expensive translation memory software. Lucía had been putting off that purchase for years. What she needed, she realized, was simpler: a single list of agreed terms, and a reliable way to check every row against it. That is something a general AI assistant can help with, as long as a human stays in charge of the decisions.
Step 1: she built the glossary before translating anything
Instead of starting at row 1, Lucía pasted the full English catalog into an AI assistant and asked one question:
“List every term that appears more than twice in this catalog: materials, colors, product line names, care instructions and measurements. Group them by type. Do not translate anything yet.”
In a few seconds she had about 90 recurring terms. This is the part many people skip, and it is the part that matters most. You cannot be consistent about terms you have not identified.
Then she did the work only she could do. She sat with the list for an hour and decided, term by term, how each should appear in Spanish. Some stayed in English because they were brand names. Some needed one version for Mexico and a note for Spain. She sent the 90-term glossary to the client, who approved it with two changes. That one email prevented dozens of later corrections.
Step 2: she translated in batches, with the glossary attached
Lucía translated the catalog herself, 40 to 50 rows a day, using AI as a drafting helper on the simpler care notes. The key habit: every time she opened a new batch, she included the approved glossary in her instructions, with a line that said “Always use these exact terms. If a term is missing from the glossary, flag it instead of choosing one.”
That last instruction is the same principle we recommend in our guide to making AI write in your own voice: give it your rules in writing, every single time, rather than hoping it remembers.
Step 3: the consistency check that caught 31 problems
When the full translation was done, Lucía ran one final pass. She gave the assistant the glossary and the finished Spanish catalog and asked:
“Check every row against the glossary. List any row where a glossary term was translated differently, any English word left untranslated that is not on the keep-in-English list, and any product name spelled two different ways. Give the row number and the exact text. Do not fix anything.”
The answer came back with 31 flagged rows. About two thirds were real problems: a color translated two ways, a measurement written in inches where the Spain version needed centimeters, a product line name that had slipped into Spanish. The rest were false alarms the assistant had misread. Lucía fixed the real ones herself in about 40 minutes.
Notice the words “do not fix anything.” When you use AI to check work, ask it to report, not to rewrite. A checker that quietly edits your text can introduce new mistakes while fixing old ones, and you lose track of what changed.
What changed for her business
In this illustrative example, the results were practical rather than dramatic:
- The client’s review round dropped from roughly two days of back and forth to a single short email, because the terms had been approved up front.
- The glossary became a product. Lucía now offers a “terminology setup” as a paid first step on every catalog job, and clients see it as professionalism, not an extra.
- The next job started faster. When the same shop added 60 products in December, the glossary was already there.
There is a bigger reason this matters for online sellers. AI shopping assistants increasingly read product pages to answer customer questions. A catalog that names the same thing three ways is harder for those systems to describe accurately. Our story about a solo tea shop fixing what AI assistants said about its products shows the same lesson from the seller’s side.
How to borrow this for your own business
You do not need to be a translator to use this method. Anyone with lots of repeated text benefits: product listings, service menus, proposal templates, help pages. Here is the short version:
- Extract first. Ask AI to list every repeated term in your existing content. Do not let it change anything.
- Decide yourself. Choose one approved version of each term. Write down which ones stay in English or keep a brand spelling.
- Attach the list every time. Paste the glossary into every writing or translation prompt, with “flag missing terms instead of guessing.”
- Check, do not auto-fix. Run a final report-only pass, then make the corrections by hand.
If you sell in more than one language, platforms like Shopify let you publish your store in multiple languages, and a glossary makes those translations far easier to maintain. And if a job is high-stakes, such as legal terms, safety instructions or medical claims, hire a professional; the American Translators Association has a directory of credentialed translators.
The small habit behind the big job
Lucía did not let AI translate her catalog. She let it remember for her, and then check her. That is a gentle, very safe way to start using AI in skilled work: keep the judgment, delegate the bookkeeping of consistency.
What is the one word your business keeps spelling or describing three different ways? Start your glossary with that one today.



