Technology
How AI helps you read multilingual feedback
How to read feedback in many languages with AI translation, sentiment analysis and voice and video transcripts, while your team keeps every decision.
The most practical way to handle feedback in many languages is to let visitors write in their own language, then give your team three AI tools that speed up understanding: translation, sentiment analysis and voice and video transcription. In EnaFeedback, the visitor interface is available in Turkish, English, German and Arabic, and admins can read comments written in other languages through AI translation in the panel. AI makes reading faster, but your team still decides where a ticket goes and how it's closed. This post explains what each tool does, how to turn it on or off, and where its limits are.
The real problem with multilingual feedback
Feedback at a hospital's international patient unit, an airport terminal or a campus full of exchange students doesn't arrive in one language. And language isn't the only obstacle:
- A complaint in a foreign language can sit unread in the panel.
- A short, angry message and a long, calm suggestion look the same in a list.
- Nobody has time to listen to what's said in attached videos or voice notes.
AI is useful because it removes these three barriers. Making the call is still a human job.
The visitor side: four languages, with right-to-left Arabic
The visitor and staff interface comes in four languages: Turkish, English, German and Arabic. When Arabic is selected, the interface switches to right-to-left. Visitors can change the language themselves, and admins can turn off any language except the organization's default.
The interface's fixed text, such as buttons, form labels and messages, comes already translated into those four languages. Content you write is different: content pages, survey questions and module text are written in your organization's language. To show them in the visitor's language, you can turn on dynamic content translation, which targets the same four languages.
| Surface | Languages |
|---|---|
| Visitor and staff interface | Turkish, English, German, Arabic |
| Admin panel | Turkish, English |
| Mobile app | Turkish, English, German, Arabic |
| Dynamic content translation (for visitors) | The same four languages |
| Comment translation (for admins) | From the comment's language into the language the admin reads |
If a visitor leaves contact details, the email and SMS sent when their ticket status changes are also available in these four languages.
The admin side: comment translation
When a German tourist or an Arabic-speaking patient's relative writes in their own language, the message lands in the panel in that language. With comment/message translation turned on, admins can read it in their own language. This feature is off by default and is turned on in the organization's AI preferences.
Translation is a reading aid, not a replacement for the original. For an important ticket, especially a complaint, it's a good habit to glance at the original too, or check with a colleague who speaks the language.
Translations are on record
Translations made in the panel can be reviewed in the Translations view under Records. The AI call log stores the provider, model, latency and status, but never the prompts or content.
Sentiment analysis: seeing what to look at first
Sentiment analysis is on by default. It estimates whether a visitor's text is positive, negative or neutral. That helps in two places:
- In analytics: The feedback analytics screen shows the tone of the text alongside the category, so messages filed as "Suggestion" but written in harsh language don't slip by.
- On the positive outcome screen: If the thank-you block shown after submission is set to "Conditional," it appears only for positive comments. A tolerance setting lets you decide what happens when the tone is neutral or unclear.
Be clear about what sentiment analysis doesn't do. It doesn't set ticket priority (priority comes from the category the visitor picks), it doesn't route tickets to a team, and it doesn't close them. Routing is always done by an authorized user, with a note.
Voice and video transcripts
Video attachments and voice notes (for example, those sent from the mobile app) can be turned into text with AI. Voice transcription and video transcription are on by default. With a transcript, your team can read and search the content without playing the recording.
If a transcript is in another language, transcript translation can render it in the admin's language. It's off by default. For videos you upload to content pages, if you don't upload a caption file, captions are generated automatically from the AI transcript where possible.
Transcripts and translations are cached for 14 days, so the same item doesn't trigger repeated AI calls.
Choosing a provider and staying in control
EnaFeedback's AI features run on one of four providers:
| Provider | How | Supported content |
|---|---|---|
| ENA AI Gateway | Default, no setup needed | Text, audio, video |
| OpenAI | With your own API key | Text and audio |
| Google Gemini | With your own API key | Text, audio and video |
| Anthropic Claude | With your own API key | Text only |
If you bring your own key, it's stored encrypted with AES-256-GCM. Only one provider is active at a time, and there's no automatic fallback to another provider.
Usage is protected by built-in daily call limits per feature: 20,000 for sentiment, 5,000 for translation and 5,000 for transcription. These limits are fixed and act as a safety net against runaway usage.
Which features to turn on, and when
| Situation | Suggested setting |
|---|---|
| Most visitors write in your organization's language | Sentiment on, translation off |
| Foreign-language comments arrive regularly | Turn on comment translation |
| Your content pages serve international visitors | Turn on dynamic content translation |
| Video attachments and voice notes are common | Keep transcripts on; add transcript translation if needed |
| Your policy rules out AI | Turn everything off; the product works fine without AI |
Example: the journey of a complaint written in Arabic
Here's how the tools work together in practice. A patient's relative scans the QR code in a polyclinic waiting area and switches the interface to Arabic. The screen flips to right-to-left, and the form labels appear in Arabic.
- The visitor picks the "Complaint" category, writes a message in Arabic and attaches a photo. The ticket is created at high priority.
- Sentiment analysis flags the text as negative, and the feedback manager sees that next to the category in the panel.
- With comment translation on, the manager reads the message in their own language, with the original right beside it.
- The manager routes the ticket, with a note, to the right resolution team. The team handles the issue and concludes the routing.
- If the visitor left contact details, they get an email or SMS in Arabic when the status changes. If not, they can track it with their request code.
In this flow, AI steps in at two points, to speed up reading and prioritizing. Every other step is a human decision.
People stay in charge
None of the AI features make decisions. The visitor picks the category, the category sets the priority, an authorized user routes the ticket, and the team or an admin closes it. There's also no AI summarization feature; your team reads the actual text.
This approach has two benefits. The first is accountability: every routing and closure is written to the audit log along with who did it. The second is trust: a bad translation or a wrong guess about tone can't quietly close a ticket.
FAQ
How many languages does the visitor interface support?
Four: Turkish, English, German and Arabic. The admin panel is in Turkish and English. Admins can read comments written in other languages through AI translation.
What data goes to the AI provider?
It depends on which features you turn on: the visitor's text for sentiment analysis, voice and video attachments for transcription, and the text being translated for translation. You decide which features are on and which provider is used.
Does AI route tickets automatically?
No. Routing is always manual and requires a note. AI only produces information that makes reading and prioritizing easier.
Can I turn AI off completely?
Yes. Each of the six feature switches can be turned on or off separately. With all of them off, feedback, ticket management and reporting work as usual.
The bottom line
Multilingual feedback is a manageable workload with the right tools. Visitors write in their own language, AI helps your team read and prioritize faster, and decisions stay with your people. See the AI insights and multilingual and accessible pages for more, or request a demo to try it with your own data.