How Do You Translate Zendesk Tickets? Native AI Translation, Apps and AI Agents (2026)
Zendesk translates ticket conversations natively on Suite Team and Support Team plans and above, one click per conversation. This guide covers turning on AI translation step by step, its limits from our own sandbox test, the per-agent apps that fill the gaps, and an AI agent that drafts the reply in the customer's language.
Key takeaways
- Zendesk translates ticket conversations on Suite Team and Support Team plans and above: an admin turns on AI translations for all channels, then agents click Translate on each conversation.
- In our 8 October 2026 sandbox test, a three-sentence Japanese ticket showed no translation banner, and a reply sent before the second Translate click went out in English.
- Translation apps price per agent: Translator AI is $3, Swifteq €7 and Macha's AI Ticket Translations $8 a month, all checked on 8 October 2026.
- Macha's test agent drafted German and Spanish replies with English translations as internal notes, at about $0.40 a ticket including setup and a dedicated success manager.
Zendesk translates ticket conversations natively on Suite Team and Support Team plans and above. An admin turns on conversation translation (and, since August 2026, AI translations for every channel), then an agent clicks Translate on each conversation: the customer's messages show in the agent's language and the agent's replies go out in the customer's. For a team that sees a handful of foreign-language tickets a week, that's enough and costs nothing extra. If you want the translation to happen without the click, to stay on the ticket as a note, or the whole reply drafted in the customer's language, the realistic options are per-agent Marketplace apps (Translator AI, Swifteq Translate Conversations, Macha's AI Ticket Translations), an enterprise translation layer such as Language I/O, or an AI agent that writes the reply itself: eesel AI, or Macha from $299/month for 750 tickets, with setup done for you.
| Option | Native or add-on | What it actually does | Writes to the ticket | Setup | Pricing unit | Entry price (checked 8 Oct 2026) | Where it falls short |
|---|---|---|---|---|---|---|---|
| Zendesk conversation translation (AI translations) | Native, Suite Team and Support Team and above | Agent clicks Translate; customer messages show translated inline, agent replies are translated on send | Sends the translated reply; the original stays in the event log | Self-serve, one admin setting | Included in the seat | Included (Suite Team $55, Support Team $19 per agent/month, annual) | Agent turns it on per conversation; no banner on short messages; incoming chat translations aren't saved |
| Zendesk AI agents | Native, billed per automated resolution | Answers customers in their language from the flows and knowledge you build | Public replies from the AI agent | Self-serve builder | Per automated resolution | 5 resolutions per agent/month on Team plans, then $2.00 each pay-as-you-go | Charges per resolution; only as good as the content you give it |
| Translator AI (Infigos) | Marketplace app | One click: the last customer message into your language, or your draft into theirs | Internal note; replaces composer text | Self-serve install | Per agent | $3 per agent/month after a 14-day trial | Translates only the most recent public message; a click every time |
| Swifteq Translate Conversations | Marketplace app | Auto-translates new customer messages, translates replies back, glossary and custom prompt | Composer text | Self-serve install | Per agent | €7 per agent/month, 7-day trial | Agent still writes every reply; enterprise tier is quote-only |
| AI Ticket Translations by Macha | Marketplace app | Translates in the sidebar when the ticket opens, optionally as an internal note; one-click reply translation | Internal note, tags | Self-serve install | Per license | $8 per license/month ($6 billed annually), 7-day trial | An agent still writes and sends every reply |
| Language I/O | Enterprise translation layer | Chat, ticket and help center translation in 150+ languages with glossary enforcement | Through its Zendesk integration | Sales-led | Not published | Not published | Demo and contract before you can test it |
| eesel AI | AI agent layer | An AI agent or copilot drafts the reply from your knowledge, in the customer's language | Drafts, replies, tags | Self-serve | Credits (1 ticket = 1 credit) | $299/month for 500 credits | You write, test and tune it yourself |
| Macha | AI agent layer on your help desk | An agent reads the ticket, tags its language and drafts the full reply in the customer's language with an English translation, or replies directly once you allow it | Internal note, public reply, tags, status, fields | Done for you by the Macha team, with a dedicated success manager, included | Per ticket | $299/month for 750 tickets (about $0.40 a ticket) | Every drafted ticket counts toward your monthly volume, and it answers only from the policies and knowledge it has |
There are two different jobs hiding under "Zendesk translation". One is reading and writing: an agent who speaks English needs to understand a German ticket and send a German reply. The other is answering: someone, human or AI, has to know what the reply should say. Native translation and the per-agent apps solve the first job only. An AI agent does both, which is why it costs per ticket instead of per seat.
How does Zendesk's built-in ticket translation work?
According to Zendesk's conversation translation article (updated 30 September 2026), translation works on every channel except voice: Zendesk messaging, live chat, social channels, Sunshine Conversations, and email, web form and API tickets. It's available on Suite Team, Growth, Professional, Enterprise and Enterprise Plus, and on Support Team, Professional and Enterprise. There is no translation add-on to buy.
How it works under the hood:
- The customer's language is detected from their recent comments. Zendesk says there's no guaranteed length at which detection works, and longer messages do better.
- The agent's language is the language on their Support profile. If an agent types in a different language from their profile, Zendesk warns that translations "may be inaccurate or behave unexpectedly".
- The agent opts in per conversation. When the two languages differ, a banner offers to translate. If detection fails (a short message, for example), no banner appears and the agent has to use the ticket's options menu.
- The customer isn't told that translation is on. Agent messages simply arrive in their language. An admin can add a "Translated by AI" note, but only on email, web form and API tickets.
- The record keeps both versions. The agent's original and translated messages go in the event log; customer messages are stored only in their original language.
There are now two engines behind that banner. The older one is AWS Translate. On 8 October 2026 the setting's own description in Admin Center said AWS covers inbound and outbound messages on live chat and messaging, but only inbound messages on email, web form and API tickets. Zendesk's own setup article says AWS also translates outbound email, so the two sources disagree; if you stay on AWS, test both directions on an email ticket before relying on it. The newer option, "AI translations for all channels", translates inbound and outbound on every channel and replaces AWS entirely. Zendesk made it generally available for live chat and messaging between 3 and 14 August 2026, according to its announcement. The same setting notes that glossary support for AI translations comes with the Copilot add-on, which is $50 per agent per month billed annually on Zendesk's pricing page on 8 October 2026.
Translation is included in the seat because it keeps the multilingual queue inside Zendesk, while the control most teams eventually want, a glossary that keeps product names and policy terms consistent, sits behind the AI add-on.
How do you turn on translation in Zendesk?
We did this in our own sandbox, d3v-macha, on 8 October 2026. Translation had been off. The admin part takes a minute:
- In Admin Center, go to Objects and rules > Tickets > Settings.
- Expand Comment options for agents.
- Tick Allow agents to translate conversations. On its own, this turns on AWS Translate.
- Tick AI translations for all channels to use AI translation on every channel instead.
- Optionally tick Show AI translation disclosure so customers see "Translated by AI" on email, web form and API replies.
- Click Save.
Then the agent side. We created three synthetic tickets through the API, from made-up customers writing in German, Spanish and Japanese, and opened each one as an agent whose profile language is English.
The German ticket showed the banner straight away: "Detected language German" with a Translate button. One click, and the message rendered in English inline, with a link back to the original.
For the Spanish ticket we turned translation on, wrote the reply in English and submitted it. The customer-facing copy went out in Spanish. In the ticket, the agent sees the Spanish version with a "Show your original message" link.
Three things from that test are worth knowing before you roll it out:
- Short messages may get no banner. Our three-sentence Japanese ticket showed none. The agent has to open the ticket's options menu, click Translate, check the detected language in the dropdown, and click Translate again in the banner. We stopped after the first click on our first try, and the reply went out in English.
- It's a per-conversation switch, not a queue setting. When we reloaded the German ticket, the banner was back offering to translate and the customer's message showed in German again. A reply typed then would have gone out untranslated, which is easy to miss on a busy shift.
- The API sees the agent's original. Reading ticket #1216's comments through the Zendesk API returned the English text the agent typed, not the Spanish the customer received. Anything that exports or audits tickets through the API, such as a QA tool, reviews a reply the customer never saw.
What are the limits of Zendesk's native translation?
Zendesk lists its own limitations in the same article, and they're specific:
- Incoming chat and messaging translations aren't saved. They run again each time the ticket is updated or reloaded, so results can vary.
- Translated messages can't be redacted, on any web, mobile or social messaging channel.
- Sunshine Conversations Shorthands and third-party widgets aren't translated.
- Phonetic text fails: Hindi works in Devanagari script, not in Roman script. That matters for markets where customers type their language in Latin letters.
- If translation fails on an outgoing email, Zendesk sends it in the original language after five minutes.
Language coverage is the other limit. Zendesk's language support article (updated 29 September 2026) lists 32 locales for live conversation translation, from Arabic to Vietnamese, counting two English, two Chinese, a Canadian French and a Brazilian Portuguese entry. It doesn't publish a separate list for the AI translation option. Zendesk's AI agents, by contrast, are advertised on its pricing page as translating their responses into "over 80 supported languages".
None of this makes native translation a bad choice. If your agents handle a few Spanish and German tickets a week on email, it's the right first step: it's free with the seat and it's already in the ticket. The limits start to cost money when foreign-language tickets are a steady share of the queue and someone has to remember to click, twice, on every one.
Which Zendesk Marketplace apps translate tickets?
Per-agent apps fix the clicking and the visibility. They all sit in the ticket sidebar or composer, price per agent, and leave the answering to your team. We checked each listing on 8 October 2026.
Translator AI by Infigos is the simplest and cheapest: $3 per agent per month after a 14-day free trial. One icon does both directions. If the editor has text in it, the app translates your draft into the customer's language and replaces the editor content; if the editor is empty, it translates the most recent public customer message into your language and posts it as an internal note. It only takes the latest message, so a long back-and-forth needs a click per message.
Swifteq Translate Conversations is €7 per agent per month with a 7-day free trial, according to Swifteq's page. It auto-translates new customer messages in real time, translates the agent's reply back, and adds a glossary, a custom prompt for outbound translation, per-agent language preferences and usage reports. Swifteq says it's powered by ChatGPT.
AI Ticket Translations by Macha is our own app, $8 per license per month or $6 billed annually, with a 7-day free trial and no card. It translates the moment a ticket opens, in the right-hand sidebar, and can also save the translation as an internal note so a teammate or a QA reviewer who reads only English can follow the thread later. Agents draft in their own language and translate into the customer's with one click before sending. Admins can set an ignore list (so English tickets aren't translated), control the translation prompt, choose how translated tickets are tagged for Explore reports, and see top languages and translation counts. Its app page lists 180+ languages, and it installs from the Zendesk Marketplace. You buy licenses only for the agents who need it, not for every Zendesk seat.

Language I/O is the enterprise end. Its Zendesk page covers chat, tickets, messaging and help center articles in 150+ languages, with glossary enforcement, PII protection and a choice between translation engines. Pricing isn't published; you start with a demo. It fits a contact center with regulated content or a long tail of languages, not a ten-person team.
The pricing logic is the same across all four. Seat pricing charges for agent headcount, not for how many foreign-language tickets arrive. Ten agents cost $30 a month on Translator AI, €70 on Swifteq and $80 on AI Ticket Translations ($60 billed annually), whether those agents see five translated tickets or five hundred. That's cheap insurance for a small team and a fixed cost the vendor earns whether or not you use it.
Can an AI agent answer Zendesk tickets in the customer's language?
Yes. A language model reads German as easily as English, so an AI agent doesn't translate the ticket, work out the answer and translate back. It reads the ticket, decides what to do and writes the reply in the customer's language in one pass. What it needs from you is the same thing an English-speaking agent needs: the policies and the order data. Three options run on Zendesk:
Zendesk AI agents answer messaging and email conversations from the flows and knowledge you build, and Zendesk's pricing page says they translate responses into over 80 languages, with custom translations for standard responses. Billing is per automated resolution: Team plans include 5 per agent per month and Suite Professional 10, then pay-as-you-go resolutions are $2.00 each (8 October 2026). Per-resolution pricing means the bill rises with every ticket the AI agent closes, which is a different incentive from a seat or a ticket.
eesel AI runs an AI agent and a copilot on top of Zendesk that draft replies from your help center and documents. Its own Zendesk auto-translate guide positions it as drafting the reply first and handling the language as part of it. Pricing is credits: $299 a month for 500, where a ticket or chat is one credit however long. It's self-serve: you build, test and tune the agent yourself.
Macha runs agents inside your existing help desk. The agent reads the ticket through the Zendesk connector, then tags, notes, replies or hands off according to written instructions. The next section shows one doing exactly this on our sandbox.
Worked example: a Macha agent that drafts replies in the customer's language
We built a test agent on our Macha Demo workspace, connected to the d3v-macha Zendesk sandbox, on 8 October 2026. We left it inactive, with no trigger, and gave it three Zendesk tools: Get Ticket, Update Ticket Tags and Add Internal Note. No public-reply tool, so it can't send anything to a customer. It runs on GPT-5.4, Macha's default model.
These are the instructions, verbatim. The three policies are made up for the test; in a real setup they'd come from your help center and past tickets.
You draft replies for Zendesk tickets written in any language. You never post a public reply.
1. Read the ticket with zendesk_get_ticket and identify the language of the customer's latest message.
2. Add the tag lang_<two-letter code>, for example lang_de, with zendesk_update_ticket_tags.
3. Answer only from these policies:
- Returns and exchanges are free within 30 days of delivery. The customer replies to our email and we send a prepaid return label. An exchange ships as soon as the return parcel is scanned.
- Refunds go back to the original payment method within 5 business days after the return reaches our warehouse.
- We can't change a shipping address ourselves. A teammate checks whether the order has shipped and updates it in Shopify.
4. If the policies answer the question, add one internal note with two parts: "Draft reply (<language>):" followed by the full reply in the customer's language, signed "The Support Team", then "English translation for the agent:" followed by the same reply in English.
5. If the request needs a teammate (as the policies say) or the policies don't cover it, write the internal note in English: one sentence on what the customer wants, then a short holding reply in the customer's language that the agent can send. Add the tag needs_human.
6. Keep product names, order numbers and addresses exactly as the customer wrote them. Use the formal register in German and Japanese.
In rule 4, the draft carries its own English translation, so the person approving it can check what it promises without a translation tool. Rule 6 covers the two things machine translation most often gets wrong in support: mangled product names and the wrong level of formality.
We ran it with Macha's Test run in autonomous mode, which simulates a "ticket created" event on the sandbox and lets the agent work the ticket exactly as it would from a live trigger. The first scenario was a German exchange request: Lena (a made-up customer) received the "Nordlicht" rain jacket in size M, it's too small, can she swap it for an L and does she pay for the return?
The agent read the ticket, added lang_de and wrote one internal note. The German draft opens "Guten Tag Frau Hoffmann", uses the formal Sie throughout, keeps "Nordlicht" in quotes as the customer wrote it, and says the exchange and return shipping are free within 30 days, that she should reply for a prepaid label, and that the new size ships once the return parcel is scanned. The English translation underneath says the same thing line for line.
The second scenario tested the handoff. Lucía writes in Spanish that she entered her old address and asks us to change it to Calle Mayor 14, 3ºB, 28013 Madrid before the order ships. Our policy says a teammate does that, and Macha doesn't edit Shopify orders, so the right answer is a handoff. The agent added lang_es and needs_human, wrote a one-line English summary with the new address copied exactly, and drafted a short Spanish holding reply: we'll check whether the order has shipped, and if not, a teammate will update the address.
We also ran the agent from the Test run chat on a Japanese refund-timing question (ticket #1225). The note came back with a formal Japanese draft (it opens お世話になっております), the 5-business-day refund rule, the product name 「Aero Run」 kept as written, and the English translation below it.
Compare the two workflows for the German ticket. With native translation, an agent opens the ticket, clicks Translate, reads it, decides the policy applies, writes the reply in English and sends it translated. With the agent, the decision and the reply are already on the ticket in German and English before anyone opens it; the human reads the English, checks it against the policy and sends. On a trigger, Macha can post the reply directly once you trust it, by adding the public-reply tool and changing rule 4.
Macha is the done-for-you option here: the Macha team analyzes your past tickets, finds which languages and categories come in, builds the knowledge base from your help center and past replies, writes and tests the agent on a sample of real tickets, and runs it in safe mode (internal notes only, like the run above) until the drafts are right. That's done-for-you setup and a dedicated success manager on the trial and every plan. Pricing is one plan priced on volume, from $299 a month for 750 tickets, about $0.40 a ticket, which includes setup and a dedicated success manager who handles your changes, such as a new language or a new policy. Against eesel, $299 gets you 750 tickets on Macha and 500 on eesel: $0.40 a ticket against $0.60, with setup done for you included. Macha fits teams already running Zendesk, Freshdesk, Gorgias, Front, Gladly, Crisp, HubSpot or Intercom who get the same questions in several languages and want the reply drafted, not only translated. It's the wrong fit if you only need agents to read the odd foreign-language ticket; native translation does that for free. The trial is $50 of free usage, about 125 tickets, with no credit card. You can start a trial and connect Zendesk in about 15 minutes.
Which Zendesk translation option should you choose?
- A few foreign-language tickets a week, mostly email: turn on Zendesk's AI translations for all channels and train agents on the two-click path for short messages. It's included in the seat.
- A steady multilingual queue and agents who write the replies: add a per-agent app so translation happens on open and stays on the ticket. Translator AI at $3 if you want the cheapest one-click tool; Swifteq at €7 if you want a glossary without Copilot; Macha's AI Ticket Translations at $8 if you want sidebar translation on open, internal notes for QA and language reporting.
- A glossary and strict terminology on native translation: that's the Copilot add-on at $50 per agent per month, or Language I/O if you need enterprise controls and a long tail of languages.
- The same questions arriving in many languages: stop translating and let an AI agent draft or send the answer. At 750 tickets a month, Macha costs $299 with setup included; per-resolution AI agents cost $2.00 for each resolution beyond the included ones.
Language is also a routing signal. If you'd rather send German tickets to a German speaker than translate them, our guide to Zendesk intelligent triage and AI routing covers routing on the detected language.
How do you roll out translated replies safely?
- Start with drafts as internal notes. Whether it's an app saving translations as notes or an agent drafting replies, nothing should reach a customer until a person has read a week of them. Our guide to AI draft replies with agent approval covers how to graduate from drafts to sending.
- Have a native speaker spot-check your top two languages. Ask them for tone and formality, not grammar. The usual failure is a reply that's correct but too casual for a German or Japanese customer.
- Fix names before go-live. List the product names, plan names and policy terms that must never be translated, and put them in the glossary (Copilot, Swifteq) or the agent's instructions (rule 6 above).
- Tag translated tickets. A
lang_xxtag on every ticket lets you report on volume, handle time and CSAT by language in Explore, and tells you which language deserves a native-speaking hire. - Decide who checks outbound in the API's blind spot. Native translation stores the agent's original in the comment and the delivered translation elsewhere, so a QA process that reads comments through the API reviews the English. Review in the Agent Workspace, or keep the translated text on the ticket as a note.
For the broader picture of what to automate first on Zendesk, see how to automate Zendesk with AI; the order-status, refund and long-ticket guides in the same series (WISMO, refunds, ticket summaries) all work the same way in any language.
Frequently asked questions
Can Zendesk auto-translate tickets? Not on its own. Once an admin turns it on, Zendesk detects the language difference and shows a banner, but the agent clicks Translate per conversation, and on short messages may need to start it from the options menu. Zendesk AI agents do answer in the customer's language automatically, and an AI agent layer such as Macha can draft the reply in the customer's language on every new ticket from a trigger.
Which Zendesk plans include ticket translation? Conversation translation, including AI translations for all channels, is on Suite Team, Growth, Professional, Enterprise and Enterprise Plus, and on Support Team, Professional and Enterprise. Suite Team was $55 and Support Team $19 per agent per month billed annually on 8 October 2026.
Does Zendesk translation need the Copilot add-on? No. Translation itself is included. Glossary support for AI translations is tied to the Copilot add-on, $50 per agent per month billed annually.
Can Zendesk translate help center articles and macros? Yes, separately. Admins turn on AI article translation under AI > Knowledge > Translations, and agents generate translations from the article editor, with a list of terms to exclude. Macros, triggers and custom ticket fields are translated by hand with dynamic content, on Growth and Professional plans and above.
How many languages does Zendesk translation support? Zendesk's language support article lists 32 locales for live conversation translation as of 29 September 2026, and its AI agents are advertised as translating responses into over 80. Language I/O claims 150+, and AI Ticket Translations by Macha lists 180+.
What's the cheapest way to translate Zendesk tickets for a small team? Native translation, since it's in the seat. If agents keep forgetting to switch it on, Translator AI is $3 per agent per month. If the real cost is the time spent writing the replies rather than reading them, an AI agent priced per ticket is the cheaper fix: Macha is about $0.40 a ticket with setup and a dedicated success manager included, and the trial covers about 125 tickets before you pay.
How we researched this
Zendesk facts come from support.zendesk.com articles 4408832500506 (how conversation translation works and its limitations), 9837507006106 (turning on AI translation), 6327770807450 (agent steps), 11064240343706 (the August 2026 messaging announcement), 4408821324826 (language support), 9380429475738 (AI article translations) and 4408883892762 (dynamic content for ticket fields), plus zendesk.com/pricing, all read on 8 October 2026. Third-party details come from the Zendesk Marketplace listings for Translator AI and AI Ticket Translations by Macha, swifteq.com/zendesk-translate-tickets, languageio.com/translation/zendesk, eesel.ai/pricing and eesel's Zendesk auto-translate guide, read the same day. Prices change; we dated each one, and Swifteq's is in euros as published.
What we ran live on 8 October 2026: we turned on "Allow agents to translate conversations" and "AI translations for all channels" in our d3v-macha Zendesk sandbox, created synthetic tickets through the API from made-up customers writing in German, Spanish and Japanese, translated them in the Agent Workspace, sent two English replies with translation on and off, and read the stored comments back through the API. We then ran the Macha agent shown above with Test run in autonomous mode on tickets #1234 (German) and #1235 (Spanish), and from the Test run chat on #1225 (Japanese). The AI Ticket Translations app is installed on our sandbox but its trial had lapsed, so we didn't run it for this guide; its image above is the product image from its page. We didn't test Translator AI, Swifteq, Language I/O, eesel or Zendesk AI agents; those descriptions come from the vendors' own pages. All names, emails, products and addresses are made up. Arithmetic: 10 agents × $3 = $30, × €7 = €70, × $8 = $80, × $6 = $60 a month; $299 / 750 = $0.40; $299 / 500 = $0.60.

