How IGP one uses AI

How IGP one uses AI

IGP one is AI. It helps you plan content, write posts, stories, reels and captions, suggest and send replies to comments, messages and story replies, capture leads, find opportunities and summarise results in your weekly report. In our mobile apps, you can also talk with it by voice.

It works from your brand kit and brand memory: the business facts, prices, products and tone you have approved. It does not invent prices, offers, testimonials or results.

Our AI providers

We use large language models from Anthropic and OpenAI through their business APIs. Requests include only the context needed for the task.

These providers do not use data sent through their business APIs to train their models, and we have not opted in to any training. They may keep data for a short period for abuse and safety monitoring under their terms.

We do not use your content or your customers' messages to train our own models.

Voice mode

Voice mode lets you talk with the same IGP one agent in our iOS and Android apps. Your audio is streamed to OpenAI's realtime service, which recognises your speech and produces a spoken reply. The voice you hear is AI-generated.

We save the text transcript in your chat history, like a typed conversation. We don't store audio recordings, and we don't use your voice to identify you. OpenAI may keep data briefly for abuse and safety monitoring under its API terms.

Voice mode follows the same rules as chat: publishing, changing a price, creating an offer or spending money still needs your approval.

You stay in control

Approval is the default. Nothing is published without your approval unless you turn on auto-publishing for the accounts and content you choose.

Automatic replies work only within the rules and limits you set, and you can pause them at any time.

Changing a price, creating an offer or spending money always needs your explicit approval. The agent can never charge your card on its own.

Edited content goes back to review, and every sensitive action is logged.

Transparency and AI labels

In the app, content and replies created by AI are labelled as AI.

When IGP one replies to your customers in Instagram messages, including replies to comments and to story replies, the conversation identifies that replies are AI-assisted, so people know they are not talking to a human. This supports the transparency rules in Article 50 of the EU AI Act, which apply from 2 August 2026, and similar rules elsewhere. You may adjust the wording, but you must not remove the disclosure where the law or platform rules require it.

If you publish AI-generated images, audio or video that realistically show people, places or events, you must disclose that they are AI-generated.

People who message your business can ask to talk to a person. IGP one then stops replying in that conversation and alerts your team, who continue it in the Instagram app. Automated replies are sent only within the time Instagram allows: 24 hours after the person's last message or, for a single private reply to a comment, 7 days after the comment.

Safety measures

Content from websites, documents and customer messages is treated as untrusted input. It can never change your permissions, settings or the agent's rules.

The agent follows guardrails: no cold messages, no invented facts, no sensitive data collection and no actions outside the permissions you granted.

We monitor for misuse and errors and improve our safeguards over time.

Limits of AI

AI output can be wrong, incomplete, outdated or similar to content produced for others. Review important business decisions before approving.

IGP one does not give legal, medical, financial or tax advice, and it does not make decisions with legal or similarly significant effects on people.

IGP one is a general-purpose marketing tool. It is not designed for high-risk uses under the EU AI Act, such as recruitment, credit scoring or access to essential services, and you may not use it for them.

Reporting a problem

If IGP one produced something harmful, inaccurate or inappropriate, email [email protected]. We review every report.