FAQs Archive - Magic Feedback

Q: How does the AI handle languages?

Magic AI Surveys support more than 100 languages. You can select a default language and enable auto-detection of languages as well. If auto-detection is enabled, all questions will be automatically translated to the user's preferred language, as set in their browser, or based on the language they respond in. All answers will be translated into your preferred language in our platform.

Q: How do the AI follow-up questions work?

You decide which questions our AI should be allowed to ask follow-up questions on. This means you have total control over which questions you want to dig deeper into, if relevant. The AI will only ask a follow-up question if the user provides a somewhat short or generic answer that lacks depth or context to make it actionable. For example, if you ask users in a free-text question what you can improve about your service, and a user provides a short, generic answer like "better support," the next question will automatically be an AI follow-up such as, "Can you elaborate on what we can do better in support?" This encourages the user to give a more detailed and actionable response. By default, Magic AI Surveys will only ask one follow-up question per initial question. You can also configure the tone of voice to match your preferences, such as being more empathetic or informal. Additionally, you can provide the AI with the context behind your survey and your objectives, allowing it to focus more on certain topics.

Q: What is the difference between AI surveys and regular surveys?

Magic AI Surveys look and feel just like regular surveys. However, they understand what you're asking in your questions, what users are answering, and the overall context of the survey. This means our Magic AI Surveys can ask relevant follow-up questions in real time, unlocking more actionable answers.

Q: How do you train your model?

We have many AI models we use for different tasks, just like human beings. The AI is trained on best practices similar to how humans would solve the tasks. For example, our AI analysis is done the same way that a senior user researcher would approach analyzing unstructured data.

Q: How is Magic Feedback different from using ChatGPT?

We love ChatGPT and use it every day, and while it's great for copy-pasting some feedback into it, it really doesn't work when you need to analyze a lot of feedback at scale and ensure you don’t accidentally upload any confidential information, potentially leading to data privacy problems. Magic Feedback is built for scale. You can integrate into all your data sources automatically, analyze all your feedback without issues across different sources, and get great reporting, insights, and alerts on top of that. To make a long story short, you use Magic Feedback instead of ChatGPT when you need to analyze a lot of feedback and ensure it’s done in a privacy-safe way, with plenty of customization options for how things are handled.

Q: Which AI model are you using?

We are using a combination of open-source models that we fine-tuned and trained to handle all the tasks required for analyzing customer feedback. This is where the magic happens, so we cannot go into great detail, but we are happy to elaborate as much as we can.

Q: Does your model hallucinate?

No, we do not use OpenAI or any other third-party provider. We use our own AI models on our own servers and have complete control over the results. You should not expect any hallucinations in the results.

Q: Can you handle translations/languages?

We know that customer feedback can come in many different languages so we can handle more than 100 languages without problems. Today, we are analyzing feedback from all around the world and unifying the results of the analysis in your desired common language. We also allow you to change the language at the company or user level.

Q: How do you work with feedback in different languages?

We support more than 100 languages and have no issues analyzing feedback in different languages and unifying the insights into one common language of your choice. Today, we are conducting analyses across the world in many languages, from Finnish to Arabic. You should not expect any issues when working with feedback across languages.