ThingLink AI features: privacy overview

This article provides an overview of the models and services that power ThingLink AI features, the data they receive, and how that data is processed.

Don't hesitate to reach out to your ThingLink point of contact or submit a request to our support team if you have any questions about the information in this article.

Note that you can control whether your users are allowed to use AI features via the organization settings as described here.

Summary

• AI models process only the user's prompt or other forms of input. They do not access any personally identifiable information about the user.
• The data is not used to train the AI models.
• The data is processed in the EU or the US, depending on the feature.

Feature: 2D image generation 

This covers different tools that allow users to generate 2D images, including background images, images inside hotspots as well as the 'Convert to ThingLink Scene' feature inside the Scenario Builder.

• AI model: Imagen by Google
• What data is collected: the user’s prompt (text input) describing the image. 
• How the data is processed: the user’s prompt is sent to Imagen via API. The model processes the prompt and returns the image output which is saved to the user’s ThingLink account. Google does not receive any personally identifiable information about the user.
• Is the data used for training the model? No
• How long does the model retain the data? The data is deleted after processing.
• Where is the data processed? United States

Feature: 360° image generation

• AI model: Skybox AI by Blockade Labs
• What data is collected: the user’s prompt (text input) describing the image. 
• How the data is processed: the user’s prompt is sent to Skybox AI via API. The model processes the prompt and returns the image output which is saved to the user’s ThingLink account. Skybox AI does not receive any personally identifiable information about the user.
• Is the data used for training the model? No
• How long does the model retain the data? Indefinitely; the data can be deleted via written or e-mail request
• Where is the data processed? United States

Feature: PDF analysis and tag generation inside the Media Editor

• AI model: Gemini by Google
• What data is collected: The user’s prompt (text input) and, optionally, a PDF file. 
• How the data is processed: The user’s prompt (and the PDF file, if applicable) is sent to Gemini via API. The model processes the prompt and returns the text output which is saved to the user’s ThingLink account. Google does not receive any personally identifiable information about the user.
• Is the data used for training the model? No
• How long does the model retain the data? The data is deleted after processing.
• Where is the data processed? United States

Feature: text generation inside the Scenario Builder

This covers text-based features inside the scenario builder ("Write a recap", "Generate questions", "Suggest branches").

• AI model: GPT model by OpenAI (self-hosted on a server controlled by ThingLink)
• What data is collected: The user’s prompt (text input)
• How the data is processed: The user’s prompt is processed by a GPT model on ThingLink’s servers. The model processes the prompt and returns the text output which is saved to the user’s ThingLink account. The model does not receive any personally identifiable information about the user.
• Is the data used for training the model? No
• How long does the model retain the data? The data is deleted after processing.
• Where is the data processed? Ireland

Feature: AI-assisted scenario creation

This covers the 'Create with AI' option available in the Scenario Builder.

The user selects which AI model to use (Gemini or GPT-4). The data is processed differently depending on this choice.

• Option 1: Gemini (default)
  • AI model: Gemini by Google
  • What data is collected: The user’s prompt (text input) and, optionally, a PDF file. 
  • How the data is processed: The user’s prompt (and the PDF file, if applicable) is sent to Gemini via API. The model processes the prompt and returns the text output which is saved to the user’s ThingLink account. Google does not receive any personally identifiable information about the user.
  • Is the data used for training the model? No
  • How long does the model retain the data? The data is deleted after processing.
• Option 2: GPT-4
  • AI model: GPT model by OpenAI (hosted on a cloud server controlled by ThingLink)
  • What data is collected: The user’s prompt (text input) and, optionally, a PDF file. 
  • How the data is processed: the user’s prompt (and the PDF file, if applicable) is processed by a GPT model on ThingLink’s servers. The model processes the prompt and returns the text output which is saved to the user’s ThingLink account. The model does not receive any personally identifiable information about the user.
  • Is the data used for training the model? No
  • How long does the model retain the data? The data is deleted after processing.

Feature: Copilot (ThingLink Creative Copilot)

• AI model: Copilot via Microsoft Copilot Studio
• What data is collected: The user’s prompt (text input)
• How the data is processed: the user’s prompt is sent to Microsoft Copilot Studio via API. The model processes the prompt and returns the response (a text message) which is displayed to the user inside the chat widget. Copilot does not receive any personally identifiable information about the user.
• Is the data used for training the model? No
• How long does the model retain the data? 30 days