Artificial intelligence policy

Our role under the AI Act

TrustInsider does not develop or train general-purpose artificial intelligence models: it integrates third-party models into its service, accessed through their programming interfaces, and is responsible for the features it builds on top of them. No TrustInsider feature falls within the high-risk systems of annex III of Regulation (EU) 2024/1689 or the practices prohibited by art. 5: the service analyses how brands and products are presented, it does not assess people.

Two distinct uses of artificial intelligence

• Measuring assistants: we ask third-party AI assistants, through their APIs or by observing the answers they give to the public, the questions a customer might ask about the brand and its competitors, and we record what they answer and the sources they cite. These questions contain no personal data about end customers.
• AI-based features: we use language models to classify the answers collected, summarise sources, answer questions in Ask, propose recommendations and prepare content drafts and editorial images.

Interface observation takes place as a person would do it, without circumventing technical protection measures and in compliance with the applicable terms. The providers of the models and assistants are listed in the sub-processor table of the Data Processing Agreement.

Your data does not train models

We do not use customer content, integration data or conversations with Ask to train or fine-tune models, whether ours or third parties’. We use providers’ commercial APIs under terms that exclude the use of submitted data for training; we send models only what the request needs and mask confidential data in logs. A provider that does not offer these guarantees receives no customer data.

Transparency: how you recognise AI

• Ask and the site assistant are artificial intelligence systems, not people: they answer using language models and can make mistakes (art. 50(1) AI Act).
• Editorial images generated with AI carry the label “Image generated with AI” in the caption and the alternative text; they do not depict real people or customers.
• Summaries, recommendations and drafts in the reserved area may be produced by a model: they are proposals to be checked, and none goes out without the customer’s rule or approval. Texts published by the customer remain under its editorial responsibility.
• Third-party AI assistant answers shown in reports are quotations of what those assistants said, with date and model: they are not TrustInsider’s opinions.

Human oversight and limits

No outward action starts without a rule or approval from the customer: changes happen in manual mode, with approval on paid plans and automatically only on the Enterprise plan, within caps and rules the customer sets and can undo. We do not take automated decisions with legal effects on individuals. Models can make mistakes or invent: this is why every measurement shows sample, date and, where possible, an interval, and every important statement links to its source.

Skills and internal governance

The people who design and operate the artificial intelligence features receive appropriate training on how they work and on their risks (art. 4 AI Act). Every model is qualified before use, with its cost and quality limits, and every replacement is recorded.

Reports

If an answer, summary or image seems wrong, misleading or harmful to you, write to support@trustinsider.com: we will check and, if necessary, correct or remove the content.