If you run a bilingual Canadian store, the short answer is this: Shopify's Translate & Adapt now refreshes machine translations on its own when the English source changes. The update rolled out to US and Canadian merchants on 27 October. Turn it on, and a product description you edit in English will have its French version updated without anyone opening the app. For Quebec, where a French storefront is a legal obligation rather than a nice-to-have, that closes the exact gap that caused most of the trouble: not the first translation, but the eleven edits that came after it.
We have audited enough Canadian stores to know the pattern. The French side was done properly at launch. Then the merchandiser rewrote forty product descriptions in March, added six collection pages in August, and swapped the hero copy three times before Black Friday. None of it crossed over. By Boxing Day the French storefront is a museum of last year's catalogue.
What the auto-sync release actually changes
Before this, Translate & Adapt did detect drift. When you edited English source content, the app flagged the French version as outdated and waited for you to click something. That flag lived inside the app, which is a place most teams visit roughly never. The translation did not break, it just quietly stopped matching.
With automatic updates enabled, the app re-runs machine translation on changed source content by itself. Same engine, same output quality, no human in the loop. The practical difference is that your French storefront now decays at the speed of your review process rather than at the speed of your attention span.
Worth being clear about what it is not. This is machine translation refreshing machine translation. It does not improve quality, it does not understand your brand voice, and it does not know that Quebec says magasiner where France says faire du shopping. It solves currency, not accuracy.
Why the drift, not the launch, is the Bill 96 problem
Quebec's Charter of the French Language, as amended by Bill 96, requires that commercial content available to Quebec consumers in another language also be available in French, on terms no less favourable. A website counts. So does a product description, a shipping policy, a returns page and a promotional email.
Enforcement through the Office québécois de la langue française is largely complaint-driven. Someone shops your store, hits a product page where the title is in French and the full description is in English because it was rewritten in September, and files a complaint. You then have a remediation window and a paper trail showing the English version was more complete than the French one for four months. That is a worse position than never having translated at all, because the gap is documented in your own edit history.
Launch-day translation is the easy part. Any agency can do it, and the Canadian stores we work with almost always get it right the first time. Maintenance is where compliance actually lives, and maintenance is what nobody scopes.
One aside, because it comes up: if you employ 25 or more people in Quebec, there is a separate francization registration obligation with the OQLF. No app fixes that. Different problem, talk to a lawyer.
How to auto-sync translations in Shopify, step by step
Install or open Translate & Adapt from the Shopify admin. The setting for automatic updates sits with the language you have configured, not at the app level, so if you are running both French and Spanish you set it per language. Enable it for French (Canada). Then do the following before you trust it:
- Create a throwaway product, give it an English description, let the French version generate, then edit the English and watch whether the French follows within a few minutes. Do not assume behaviour from a release note. Test it in your own store.
- Take a product where someone has hand-corrected the French, edit the English source, and confirm the human version survives.
- Check a theme section string, a collection description and a metafield, because coverage differs by resource type and the gaps are not obvious from the UI.
Budget twenty minutes. The failure mode you are looking for is not "it did nothing", it is "it did something to content you did not want touched".
Does Translate & Adapt retranslate your edits?
This is the question everyone asks second, and it deserves a direct answer: translations you have edited by hand are treated differently from machine-generated ones. The app tracks which strings a human touched, and automatic updates target the machine output. Your reviewed French copy is not supposed to be overwritten when the English changes.
Which creates a second-order problem nobody warns you about. If you hand-corrected the French description for your best-selling product in May, and the merchandiser rewrites the English in November, the French now stays frozen at the May version. Protected and wrong. Auto-sync will not save you there, because it correctly assumes you knew what you were doing.
So keep a short list of every page and product where a human overrode the machine. In practice that is your top thirty revenue pages plus the legal copy. Those are the ones that need a manual check whenever the English moves. Everything else can ride on the automatic refresh.
The content the app does not reach
Translate & Adapt covers store content: products, collections, pages, blog posts, theme strings, navigation, email notification templates. It does not cover everything a Quebec shopper sees between the cart and the confirmation email, and the leftovers are where English shows up in the most embarrassing places.
- Discount titles. "Boxing Day Sale – 25% off" is a merchant-created string that renders in the order summary and the confirmation email. Machine translation will not help you here, and incidentally Quebec says l'Après-Noël, not Boxing Day.
- Shipping method names. Carrier-calculated rates return whatever the carrier sends. Flat rates and custom methods use whatever you typed. During the last Canada Post disruption we watched merchants add Purolator and UPS fallback rates in a single afternoon, all of them named in English, all of them live at a French checkout by dinner.
- App-generated content. Reviews widgets, subscription selectors, loyalty tiers, size charts rendered from a third-party app. Each app handles localisation its own way, and some do not handle it at all.
- Metafields. Translatable only if the metafield definition is marked as translatable. Flip that on before you populate it, not after.
- Images with text baked in. Your festive banner says SALE in 90-point type. No translation layer touches a JPEG.
- PDFs. Spec sheets, care guides, warranty documents.
Filter and facet labels are a quiet one too. If your filters are driven by tags or metafield values, the values themselves need translating or a French shopper browses a sidebar of English colour names. If you are rebuilding that layer anyway, pick a filter app that supports translated values and French search synonyms; ours is FilterPro, and there are others. Just check the translation handling before you commit, because retrofitting it is miserable.
French checkout content for Canadian stores
Shopify localises the checkout's own interface strings, so the buttons, field labels and Shop Pay flow appear in French once the buyer's language is French. Shop Pay adoption being what it is in Canada, a good share of your Quebec traffic never sees a form field anyway.
What breaks is the content you supplied. Checkout UI extensions with custom text. Delivery method names. Gift card messages. Order status page custom blocks. The refund and privacy policies linked from the footer of the checkout, which are merchant content and frequently English-only even on stores that did everything else right.
Tax lines are worth a look as well. A Quebec order carries GST at 5% and QST at 9.975%, and the French invoice should read TPS and TVQ. On a $200 order that is $10.00 and $19.95 respectively, $229.95 total. Small thing, but it is the line a Quebec customer reads most carefully, and it is the sort of detail that tells them whether the French side was built or bolted on.
Keeping French product descriptions current when the catalogue moves
Some arithmetic, because the scale of the drift is not intuitive. Take a 600-SKU store. Count the translatable fields that actually matter per product: title, body, SEO title, SEO description, and two metafields for materials and care. Six fields, 3,600 strings.
Edit 8% of the catalogue in a quarter, which is a modest rate for anyone doing real merchandising, and 48 products fall out of sync. That is 288 strings. At 90 seconds per product to open the app, re-run and glance at the output, catching up costs 72 minutes.
Seventy-two minutes is nothing. The reason this was a problem is that in most quarters, nobody spent the seventy-two minutes. Auto-sync removes the task entirely for machine-translated fields, which is the correct way to fix a task that depends on someone remembering.
What remains is a sampling job. Pull ten random French product pages a month and actually read them. We catch something about a third of the time on the first pass, usually a product name that should not have been translated, a unit conversion the engine invented, or a brand term rendered into something unfortunate.
Reviewing machine translations before they go live
Auto-sync publishes. There is no staging step, no approval queue. For most of a catalogue that is the right trade, because an imperfect current French description beats an elegant stale one, legally and commercially. For a few surfaces it is not.
The pages we insist on a human pass for: the homepage, the top ten revenue products, the shipping and returns policy, the size guide, every email notification template, and any page making a claim about ingredients, safety or warranty. Call it forty pages at roughly 300 words, so 12,000 words of human-reviewed French. That is a real invoice from a real fr-CA translator, and it is money well spent exactly once, after which auto-sync maintains everything around it.
Things to brief that translator on, because machine output misses them consistently: Quebec vocabulary over France vocabulary (courriel, not e-mail; panier, not chariot), English terms left untranslated in a French sentence, and the holiday naming. Set your store language to French (Canada) rather than plain French so the locale is at least declared correctly, and get the hreflang and translated meta descriptions right while you are in there, because a French storefront that does not rank in French is doing half a job.
Translating into eight languages, not two
The free app's automatic translation covers a small number of languages. If your answer to "how many languages" is eight, you are past what it does for free and into one of three routes: a paid translation app with its own engine and glossary support, translations pushed through the translations API from whatever system holds your source content, or a Plus plan where the published-language ceiling is higher. Check the current cap under Settings and Languages in your own admin before you plan around a number you read somewhere.
For a Canadian merchant the honest advice is usually to stop at two and do them properly. English and French, both current, both reviewed, is worth more than six half-maintained locales. The exception is cross-border: if a meaningful share of revenue comes from the US, your work is duty, de minimis and landed-cost display, not a third language.
What to do before Boxing Day
Campaign content is the highest-risk category on a bilingual store because it is written fastest and lives shortest. Landing pages built on 23 December, discount names typed into the admin at 11pm, a banner image with English text. None of it goes through the review process the evergreen catalogue gets.
Two weeks before the promo, build the French versions at the same time as the English, in the same ticket. Name discounts bilingually or neutrally. Render promotional text as HTML over an image rather than inside it. Check the shipping method names you will use if a carrier falls over, because you will not be translating anything at that point.
Then switch on automatic updates, run the throwaway-product test, and write down which pages carry human-edited French so you know what the sync will leave behind. If you would rather have someone else go through the checkout, the notification templates and the metafield definitions line by line, our free audit covers it, and we will tell you plainly if there is nothing to fix.