Translate Your Shopify Store: Translate & Adapt vs DeepL vs ChatGPT

Start with Shopify's free Translate & Adapt app, because it writes translations straight into your store, then test DeepL and ChatGPT on 20 real product descriptions in each language. Use whichever scores best with a native speaker. Mixed use is common: Translate & Adapt for structure, DeepL or ChatGPT for the pages that sell.

The honest answer to "which is best" depends on your languages, your catalogue size and how much hand editing you can afford. So run a test before you pick a winner: quality varies by language pair and by product type, and I would not trust anyone's ranking for your store until you have checked it on your own products. The costs below come from the vendors' own pages; the quality scores are illustrative so that you can see how to read your own.

Follow me on Instagram@sagnikteaches

What each route actually is

Translate & Adapt is Shopify's own app. Shopify's help centre says it installs automatically when you add a language, covers products, collections, articles, policies and pages, and offers manual translation or AI translations from Shopify, with a weekly Auto-sync in some markets. The App Store listing shows it as free, with a side-by-side editor, 100-plus languages, and translation from the theme editor or any resource page in admin. It also lets you adapt content between markets that speak the same language.

Connect on LinkedInSagnik Bhattacharya

DeepL is a dedicated translation service. Its plans are Free (50,000 characters a month), Individual, Team, Business and Enterprise, with a separate API for developers. Glossaries, file translation and the formal or informal tone setting vary by plan, and the Free plan allows a single glossary with only a handful of entries, so check the plan table if your catalogue depends on fixed terms. DeepL does not sit inside Shopify admin: you move text through a CSV file or through a third-party app that uses the DeepL API.

Subscribe on YouTube@sagnikteaches

ChatGPT is a general assistant that can translate and also rewrite for tone and market. It needs the most care with formatting and consistency because it processes whatever you paste in. It is also the only one of the three you can tell, in plain words, to keep your brand voice, avoid a banned phrase or localise units.

How to run a fair side-by-side test

Run a short test in one target language, with one reviewer, before you commit. The steps below take half a day.

1. Pick 20 strings. Use a spread: 6 product descriptions (best seller, a technical product, a cheap one), 4 product titles, 3 collection descriptions, 3 policy or shipping paragraphs, 2 email notification texts, 2 button or theme labels.
2. Translate each string three ways. Use Translate & Adapt's automatic translation for the first, DeepL for the second, ChatGPT for the third with a fixed prompt (below).
3. Hide the source of each version. Label them A, B and C in a spreadsheet, shuffled differently per row, so the reviewer does not know which tool wrote which.
4. Have a native speaker score each version from 1 to 5 on accuracy, naturalness and brand fit, and mark any error that would confuse or mislead a customer.
5. Count critical errors separately. A fluent but wrong size, material or return period is worse than clumsy wording.

A fixed prompt for the ChatGPT arm, so the test is repeatable:

Translate the following online-shop text from English into [language] for customers in [market].
Rules:
- Keep all HTML tags, variables in curly brackets, numbers, product codes and URLs exactly as they are.
- Use the glossary below for these terms and no others: [term = approved translation, ...]
- Use the [formal / informal] form of address.
- Convert nothing; do not change units or prices.
- If a phrase is ambiguous, translate it literally and add a note after the text beginning NOTE:.
Text:
[paste string]

An illustrative scoring sheet

Imagine a print shop selling custom stationery, testing its first target language. Twenty strings are scored by a native reader. The figures below are invented to show how the sheet reads, not a result from a real test.

Reading it: the gaps are small, and the real decisions sit in the critical errors column and in the comments. Say the reviewer's two Translate & Adapt errors were a wrong paper-weight term, and a return window shown as 14 days when the source said 30. The ChatGPT errors were a dropped HTML tag and an invented sentence. That pattern tells you what to guard against: an approved glossary for materials, a check of every policy number, and a tag-count comparison for pages that went through ChatGPT. A different shop, a different language and a different reviewer may produce the opposite ranking.

What it costs for a catalogue of 120 products

Estimate the volume in characters, because DeepL counts characters. For an illustrative shop with 120 products at 150 words each, plus titles and 6,000 words of collections, policies and pages, the total is roughly 24,000 words. At about 6 characters a word including spaces that is near 144,000 characters per language. The cost structure then looks like this.

The software cost is small next to your review time. Fifteen minutes per product to check one language across 120 products is 30 hours, so decide up front which pages deserve a full review. See what adding AI to a Shopify store costs for the wider picture.

Theme text, checkout and emails: the parts product tools skip

Product descriptions are the easy part. The places stores usually fail are the strings a tool does not see.

• Theme text. Buttons, labels, search placeholders and cart messages live in the theme. Translate & Adapt lets you edit them from the theme editor, but themes differ in which texts are exposed. Open your store in each language and click through browse, cart and account pages looking for any English left behind.
• Checkout and notifications. Shopify's help centre says that once a language is published, checkout and email notifications appear in the customer's language, and you can edit and preview notification translations for each language. Place a test order in each language and read every email.
• Customised email templates. One merchant review on the app listing, dated September 2026, says the app pulled Shopify's original email template instead of a customised one. I cannot tell how widespread that is, so test a customised notification in every language you publish rather than assuming it carried over.
• Policies and legal pages. Translate them, but ask an adviser if a legal reader in that market needs the wording to be exact. A machine translation of a returns policy is a business promise.
• Images with text. Product images and banners with embedded English text are not translated by any of these tools.
• Sizes, units and currencies. Never let a tool convert a measurement on its own. Decide a unit convention per market and write it into your glossary.

When each method wins

Translate & Adapt wins when you want content inside Shopify with no file shuffling, when you sell in a few languages, or when you need to adapt wording between English-speaking markets. Its side-by-side editor makes manual fixes practical, and the language limit for Shopify's automatic translations is the thing to check first.

DeepL wins when your reviewer rates its wording as the most natural in your language pair, which is exactly what the blind test shows. Its glossary and tone settings help with consistency, though the larger glossaries sit on paid plans. The cost is workflow: files in, files out, and a check that import did not mangle tags.

ChatGPT wins when you want more than translation: a product description reshaped for a market, a pun replaced rather than translated literally, or a reply in the customer's tone. It loses when you need hundreds of items processed consistently without drift. If you use it for the whole catalogue, keep one conversation or project per language with your glossary pinned at the top, and run an automated tag count after each batch. Shopify's own assistant has its own role in your admin, covered in the Sidekick tutorial.

A hybrid that many small stores can run

1. Add the language and let Translate & Adapt create the first draft for pages and policies.
2. For your top 10 products by revenue, replace the draft with DeepL or ChatGPT versions, chosen by your test.
3. Pin a glossary of 30 to 60 terms: product names that stay in English, material names, size words, brand taglines.
4. Send the top 10 plus checkout and the returns policy to a native speaker for review.
5. Publish the language, place test orders and fix leftovers.
6. Translate the long tail of the catalogue in batches, spot-checking one in ten.

Catching formatting damage before it goes live

Machine translation can corrupt structure silently. Three cheap checks catch most of it. First, compare the number of tags in source and translation: if the English has 14 occurrences of an opening paragraph tag and the translation has 12, something was lost. Second, search the translated file for curly-bracket variables and confirm each source variable appears once, unchanged. Third, view three translated pages on a phone and a desktop, looking for run-together words and broken lists. In a CSV, a stray comma or an unescaped quotation mark can shift every later column, so open the file in a spreadsheet before import and check the row count equals the source.

For a simple tag-count check you can ask ChatGPT itself: paste the source and the result and ask it to list any HTML tag or variable present in one but not the other. Treat its answer as a lead, and confirm with a search in the file.

How a native speaker should check it

A review request that says "please check this" gets slow, vague feedback. Give the reviewer a task with a limit: a spreadsheet with source in column A, translation in column B, a column for severity and a column for the corrected text, 40 to 60 items chosen by traffic. Ask them to mark each item as accurate, awkward or wrong. A back-translation can support the review, but only as a prompt for questions. Ask the reviewer separately about three things a tool cannot judge: whether the tone suits the product (playful stationery versus workwear), whether any phrase sounds like a machine, and whether any word could offend or confuse in that market. Pay for their time and keep a log of the corrections they make, since repeated ones belong in your glossary.

A realistic example: the reviewer of the stationery shop's translated pages repeatedly corrects the word for a paper type that every tool rendered as a generic word. That single entry goes into the glossary, and every later batch uses it. Ten such entries remove most of the recurring complaints, which is much cheaper than reviewing every page.

A mistake that cost a returns policy its meaning

Picture a shop that translated its returns page with a free tool and published it without review. The source said "Returns accepted within 30 days of delivery, unused, in original packaging". The translation read, in the target language, as 30 days from order. Customers then asked to return items that had arrived late, and the owner either refused and upset them or accepted losses. The fix was to rewrite the sentence with a reviewer, add the date rule to the glossary and run a test order. The lesson is that policy pages deserve the first native review, not the last.

Adapting wording for markets that share a language

Translation is not the only job. Translate & Adapt lets you adjust content between markets that speak the same language, so a shop selling to two English-speaking markets can change spelling, seasonal wording and delivery promises without a second store. A simple ChatGPT prompt handles the first draft: "Rewrite this product description for customers in market B. Use their spelling conventions, replace any idiom that would not be understood, keep every fact and number unchanged, and list the changes you made." The list matters, because it lets a person see whether the tool changed something it should not have, such as a dimension or a delivery claim. Then paste the approved version into the market-specific field and leave the rest as it is.

Search and URLs for translated pages

Shopify's help centre says each published language gets its own URLs, and your store shows the translated version where translations exist, falling back to your default language where they do not. Two practical points follow. First, an untranslated page in a translated store looks half-finished to customers and to search engines, so publish a language only when its core pages are done. Second, translate the page titles and meta descriptions as well as the visible text, since those are what appear in search results; Translate & Adapt lists SEO fields as translatable. Write them for a searcher in that language, not as a literal copy of the English keyword, and ask your native reviewer which phrase a shopper would really type.

A six-week rollout for one new language

• Week 1: run the 20-string test, choose the tool mix, build the glossary of 30 to 60 terms.
• Week 2: translate the top 10 products, collections, shipping and returns, then send them for native review.
• Week 3: apply the corrections, translate theme text and notification emails, place test orders on a phone and a laptop.
• Week 4: publish the language, switch on any market-specific currency and pricing, and watch the first conversations and orders.
• Weeks 5 and 6: translate the next 40 products, spot-check one in ten, and compare conversion on translated pages with the English baseline.

Track two numbers from day one: how many sessions arrive on the translated pages, and what share of them check out. If sessions are healthy and checkout is poor, the problem is probably wording, shipping terms or payment options, and your reviewer should look at the cart and checkout path again. If sessions are poor, no amount of polishing the descriptions will help, and the work is in how people find the language version.

Keeping the translation up to date

Every change to your English store creates a stale translation. Shopify's help centre says Auto-sync can keep automatic translations updated weekly in some markets, but hand-edited translations and edits made outside the app need a process. Keep a short rule: when a product description or price-related text changes in English, flag the translated page, retranslate it and review the changed sentences. Once a quarter, run a check of your five best-selling pages in every language, compare with the English and fix drift. If a measurable share of your orders now come from translated pages, move the language up the priority list, and consider adding translated support through AI-assisted multilingual replies so the experience does not end at checkout. If you would like a hand choosing languages and pages, that is what my AI implementation consultation covers.

Questions about translating a Shopify store

Is Translate & Adapt free?

The app is listed as free on the Shopify App Store, and Shopify's help centre says it installs automatically when you add a language. The app listing says Shopify's automatic translations cover up to two languages, or eight in some markets, while manual translations have no such cap. Check your admin for the limit that applies to your store.

Can I use DeepL directly inside Shopify?

Not through a built-in connection that I could verify. The usual routes are exporting translations as a CSV, translating the text with DeepL and importing the file, or using a third-party app that calls the DeepL API. Check an app's reviews and pricing page before installing it.

Will ChatGPT break my product page formatting?

It can. If you paste text containing HTML tags or placeholders, a model may alter, translate or drop them. Instruct it to leave tags, variables in curly brackets and numbers untouched, then compare the tag count of source and result before importing.

Does translating my store help search ranking?

Translated pages get their own URLs, which gives them a chance to appear in searches in that language. Quality matters because poor wording harms conversion even if the page is found. Translate product pages that already sell before translating the whole catalogue.

Do I need a native speaker if I use a good tool?

For anything customers pay on, yes, at least as a sample check. Tools make confident errors in tone, units and product terms, and only a native reader notices. Pay for a short review of your highest-traffic pages rather than the entire store.

Further reads

• How to Write Shopify Product Descriptions With AI That Convert — Write strong source descriptions before you translate them.
• Can an AI Chatbot Handle Order Tracking and Returns? — Automate order-status questions once the store is translated.
• Can AI Handle Restaurant Menu Translation Without Mistakes? — A similar accuracy check on another short-text translation job.
• Taking a Local Shop Online With AI: Product Listings in a Weekend — Get a small shop's listings live before you translate them.
• Best AI Chatbots for Small Shopify Stores (2026) — Add a chatbot once your store speaks more than one language.
• AI for Tour Guides: Multilingual Enquiry Replies and Review Responses — Another small-business example of multilingual replies with AI.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Shopify App Store listing for Translate & Adapt (including its reviews), Shopify help pages on languages and the Translate & Adapt app, DeepL plans page and its USD price listing, and the ChatGPT pricing page, checked 5 November 2026. The example quality scores are illustrative, not results from a real test.