Webflow translation memory: what it is and why it matters

Translation memory is the quiet feature that decides whether translating your store costs $20 or $200 per month at the same catalog size. Most evaluators do not ask about it, then wonder six months in why their bill keeps climbing while their content barely changed.

This is the explainer we wished existed when we were picking translation tools. We cover what TM actually is, why Webflow stores benefit from it more than most, the real math on a 200-product catalog, and where TM quietly gets in your way.

What translation memory actually is

Translation memory (TM) is a cache. When you translate a piece of text from one language to another, the tool stores three things: the source text, the translated text, and the language pair. The next time the exact same source text appears in the same language pair, the tool returns the cached translation instead of calling the underlying AI engine again.

Mechanically it is a database table with roughly this shape:

source_hash  | source_locale | target_locale | source_text                  | translated_text                | hit_count
-------------|---------------|---------------|------------------------------|--------------------------------|----------
abc123       | en            | nl            | "Stainless steel frame"      | "Roestvrijstalen frame"        | 47
def456       | en            | nl            | "Free shipping over $50"     | "Gratis verzending boven $50"  | 230
...

The source_hash is a deterministic hash of the source text. Lookups are O(1). A typical TM hit returns in under 5 milliseconds; an AI engine call takes 300 to 2000 milliseconds and bills you for the characters.

Why Webflow stores benefit more than most

Translation memory is useful for any localization workflow. It is dramatically more useful for Webflow stores than for, say, blog content, because Webflow Ecommerce catalogs have unusually high text repetition. Three reasons:

Boilerplate fields. "Free shipping over X." "Add to cart." "In stock, ships in 1 to 2 days." "Customer reviews." These appear on every product page in your store. With 200 products, that is 200 source strings that translate once and reuse forever per locale.

Shared product attributes. "Material: stainless steel." "Color: matte black." "Dimensions: 30 x 40 cm." Product specifications repeat across SKUs, across product lines, across seasons.

Template-driven copy. Webflow's CMS encourages templated product descriptions. The same writer creates 50 products and reuses phrasings. "Designed in Amsterdam, made in Portugal" might appear on 30 different items.

The exact percentage of repetition is impossible to predict without your specific catalog. The way to find out is to count: pick a translatable string from one product (e.g., "Free shipping over X," "Material: Y," "In stock, ships in Z days"), then scan the rest of your store. Stores that lean on templated descriptions and shared specifications will see a lot of repeats; fully bespoke catalogs will see fewer. The architecture of TM means whatever your number is, you only pay to translate each unique string once.

An illustrative example: a 200-product store across 3 locales

A 200-product Webflow Ecommerce store with roughly 60 translatable words per product produces around 12,000 source words. Without TM, a translation tool bills you for the full 12,000 across each target locale on the first sync and again on every subsequent full sync.

For the math below we assume 40 percent repetition (a midpoint of typical Ecommerce catalog ranges; your real number could be lower or higher). The point is the SHAPE of the savings, not the precision of any single percentage.

Without TM:
12,000 source words × 3 locales × (every sync) = 36,000 billed translations per sync

With TM (assuming 40% repetition + 10% per-sync content change):
First sync:  12,000 × 3 = 36,000 billed translations
Second sync: 12,000 × 3 × 0.10 (only changed strings) = 3,600 billed translations
Subsequent syncs: low thousands until you ship a major content overhaul

On a per-word-billed tool (Weglot's Business tier is listed at 50,000 words/mo at the time of writing), TM is the difference between fitting comfortably and getting pushed to a higher tier within a few months as you sync content updates.

On a flat-tier tool, TM is the difference between staying inside your monthly AI translation cap and hitting it on the second sync.

Translation memory vs glossary

These get confused often. They solve different problems.

Translation memory caches translations of full strings. "Add to cart" was translated once; reuse it next time. TM is about avoiding redundant work.

Glossary protects specific terms from being translated. "iPhone Pro" should stay "iPhone Pro" in every locale, not become "iPhone Professionnel" in French. Glossary terms are wrapped in <span translate="no"> tags before being sent to the engine; the engine respects the tag and copies the inner text verbatim.

You want both. TM saves money and time. Glossary saves your brand from embarrassing literal translations.

The interaction matters: if a source string contains a glossary term, that string is excluded from TM lookups (because older cached translations might pre-date the term being protected). The string goes to the engine fresh, comes back with the glossary term preserved, and is then stored in TM for future reuse.

How TM handles content updates

The smart-edit case is what separates good TM implementations from useless ones.

You edit a product description: change "Stainless steel frame" to "Stainless steel frame, brushed finish." The TM lookup on the new string misses (it has not been seen before). The engine translates it. The old entry "Stainless steel frame" stays in TM in case other products still use it.

Two consequences:

1. Only changed strings cost. Editing 1 product description triggers 1 engine call, not 200.
2. TM grows over time. Stale entries accumulate. A clear-all-TM button is essential for resetting after major brand voice shifts.

Where TM gets in your way

Honest gotchas, because nothing is free.

Tone consistency across batches. If you translate batch A in week 1 and batch B in week 6, TM hits on shared strings will use the week-1 translations. If your brand voice evolved, those old translations now look stale. Clear or scope TM to refresh.

Context-dependent meaning. "Free" can mean "no cost" or "available" depending on context. TM matches on exact string, not context. Edge cases where the same string needs different translations in different contexts are real, though rare in Ecommerce catalogs.

Cross-locale leakage on revision. If you edit a translated string manually (because the AI got it slightly wrong), good TM stores your human-edited version. Bad TM keeps serving the AI version. Make sure your tool stores the edit, not the original.

How the tools stack up

Quick survey of how the major Webflow translation tools handle TM (as of 2026):

• StoreLingo: TM built in on every tier (Starter through Agency). Browse, search, and clear from the panel. Stores human-edited overrides. Glossary-protected strings excluded from TM correctly.
• Weglot: TM-like features on Pro tier and above ($108/mo+). Not on Starter or Business.
• Linguana: Partial TM functionality. Reuses translations within a single site but limited cross-locale memory.
• Webflow native multi-locale: No TM. You retype every duplicate manually.
• DIY freelancer: Depends on the freelancer's tool (most professionals use Trados, MemoQ, or Smartcat which have excellent TM). Self-translation via Google Translate, no TM.

Honest take

Translation memory is the feature that determines whether your translation tool gets cheaper or more expensive as your store grows. Without it, every product update is a full retranslation bill. With it, you pay once for the substantial fraction of strings that repeat across your catalog, then pay only for genuinely new content.

If you are evaluating translation tools, ask three questions before signing up:

1. Is TM included in the plan I am considering, or is it gated behind a higher tier?
2. Can I see, search, and clear TM entries from the UI, or is it a black box?
3. If I manually edit a translation, does TM store my edit (so future identical strings get my version) or keep serving the original AI output?

A "yes" to all three usually correlates with a tool that respects your money and your translators' time. A "no" or "we are working on it" answer to any of them is a signal the underlying business model needs your bill to scale faster than your content does.