We use AI to make your campaigns more powerful — so contacts can reply with their voice, so answers get understood automatically, and so your messages can be spoken out loud. This article explains, in plain language, where AI is used and exactly what information we share to get the job done.

Reminder: AI is powerful, but it is not perfect. It can get things wrong (and sometimes very wrong), especially with short or unclear replies. We recommend testing each feature with your own use case before relying on it.

We only share what the task needs

Whenever a feature uses AI, we send only the information needed for that specific task, nothing more.

In particular, we do not send a contact’s phone number, first name, last name, or any other custom field, unless you put them into the content yourself—and that part is critical.

How you might share sensitive information with AI inadvertedly

If you want to evaluate privacy risks around AI usage, please consider what you share when setting up the survey, and maybe more importantly: what your contacts might be sharing when replying.

When it comes to the survey design, the personalization feature needs particular care. For example, you might set up a question with answer categorization, and the question includes the contact’s name (“Hi FIRST NAME”). In this case the first name will be shared with AI as part of the question. Personalization allows you to include arbitrary fields in a message, and whatever info they contain. This gives you a lot of flexibility—and with that power comes the responsibility to consider where that data goes.

Note that when an answer is processed by AI, it will process whatever the contact replied. You need to evaluate the privacy risk aforehand. For example, if you ask someone for their name, then the expectation of course is that AI will process the name in the answer.

Remember: We share your data with AI only when you use an AI feature, and only the data that you or your contacts provide to that feature.

Where we use AI, and what we share

Below we discuss each AI use case, and what data is shared.

1. Transcribing audio files

Inside the app you can let AI transcribe audio files. For example, when you upload an mp3 for a voice question, you can generate the transcript if you don’t have it ready.

What we share: the audio file you wish to transcribe.

2. Transcribing voice replies

When you enable transcription for WhatsApp questions or on voice calls then we end up with an audio file (a recording in the case of a voice call) spoken by your contact. We then use AI to detect the language and turn the recording into text.

What we share: the transcript of the question if any, and the audio file.

3.Speech recognition

In Voice calls you have the option to enable speech recognition. We use AI to figure out if the contact selected one of the answer choices when talking.

What we share: the answer recording, the transcript of the question, the transcript of the answer, and all possible answer choices.

4. Detecting answers (matching and categorizing replies)

When a contact replies in their own words, you may use AI to figure out what they meant; either matching their reply to one of your answer choices, or sorting it into a category.

• Matching an answer: if your choices are “female” and “male” and the contact
  writes “I am a woman,” AI identifies “female,” and your campaign continues just as
  if they had picked it directly.
• Choosing a category: for the question “What is your favorite time of day?” with
  categories morning, afternoon, and evening, a reply of “when I eat breakfast” might
  be categorized as “morning.”

What we share: the question as it is sent, your answer choices or categories,
and the contact’s reply (including the audio and transcript, depending on the medium).

5. Text-to-speech (turning your text into spoken audio)

We use AI to read your written message aloud in a natural-sounding voice.

What we share: the text you want spoken.

6. Chatbot (AI-powered conversations)

We use AI to hold a back-and-forth conversation with your contacts based on the instructions and knowledge base you set up.

Since the chatbot is entirely AI-powered, there is no way to avoid AI-exposure. Still below are the three datasets that you share with AI when using the chatbot campaign.

What we share:

• instructions & prompt: apart from the raw text, personalization variables might expose sensitive information.
• knowledge base: each chatbot is given access to a custom knowledge base, which is used to generate replies. Whatever is in that knowledge base is potentially exposed to AI.
• conversation messages: AI reads any message between the chatbot and the contact.

Have questions? Reach out to our support team any time.