> ## Documentation Index
> Fetch the complete documentation index at: https://docs.instantcampaign.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# EU AI Act transparency and residency

> See which AI models serve your workspace and where they run, how AI images are marked, add an optional AI-content line to marketing emails, and export the AI decision log.

The EU AI Act's transparency rules (Article 50) apply from **2 August 2026**. InstantCampaign gives you the facts and the switches you need to work out your own position. Find them under **Settings → Privacy**:

* **AI Act & residency**: which model and provider serve text, images and video, where they run, how generated images are marked, and the optional AI-content label.
* **AI decision log**: every AI output your workspace received, with a CSV export.

Both pages are for **Owners and Administrators**. Changing the AI-content label needs permission to change workspace settings.

<Warning>
  These pages describe how the product behaves. They aren't legal advice and don't certify compliance. In particular, it is unclear whether the Act's labelling rule for AI-generated text reaches promotional email. Ask your own counsel.
</Warning>

## What each part covers

| Article 50 duty | What InstantCampaign does |
| - | - |
| **50(1)** Tell people when they interact with an AI | Spark, the in-app assistant, always shows "You're chatting with an AI assistant". Your recipients never talk to an AI: forms, landing pages, the preference centre and emails contain no chatbot. |
| **50(2)** Mark AI-generated images in a machine-readable way | Images generated or edited with AI carry IPTC/XMP metadata (`DigitalSourceType = trainedAlgorithmicMedia`) and an id that links back to the generation record. |
| **50(4)** Disclose AI-generated text published to inform the public on matters of public interest | Optional footer line on marketing emails, off by default. |

### Image marking and its limits

Marking happens the moment an image is generated, before you see it. It covers AI Studio, the builder's AI image tools, the Design Assistant, AI landing pages and images Spark generates. The page shows how many images from the last 90 days carry the marking.

Not covered yet:

* **No signed C2PA manifest.** A C2PA manifest must be signed with a certificate, and InstantCampaign doesn't hold one yet.
* **No invisible watermark.** The Commission's Code of Practice expects more than one marking technique where needed.
* **Metadata can be stripped.** Some tools and websites remove it when an image is re-saved or uploaded.

## The AI-content label

Under **AI-generated content label**, choose:

* **Off** (default): nothing is added.
* **Add the built-in line**: "Parts of this email were created with the help of AI.", in the recipient's language.
* **Add my own text**: your wording, up to 300 characters.

The line is added to **marketing** email only (campaigns, journeys and marketing sends through the API), below the footer. Transactional email never gets it.

To label a single email instead, add `{{ ai_disclosure }}` where you want the text and leave the label **Off**. The variable shows your own text if you set one, otherwise the built-in line in English.

<Note>
  Article 50(4) asks for a disclosure on AI text "published with the purpose of informing the public on matters of public interest", unless a person reviewed it and holds editorial responsibility. Marketing email that you review before sending may fall outside that, so the label is a voluntary transparency choice. Ask counsel before relying on either setting.
</Note>

## Residency and fallback

The table lists the model, provider, hosting (EU, UK, US or Global) and region for each capability, and whether the current residency mode allows it. Residency is enforced by the model router: a model outside the boundary is refused, never silently used. The page also shows whether a failed request may move to another provider.

## The AI decision log

Each row is one AI output: when, the kind (generation, segment AI, weekly proposals, landing page and so on), who asked, the model and provider, the region it ran in, a short summary and the id of what it produced. Summaries hold **types and counts only**, never your prompt text or any contact data.

* Filter by **Kind**, then click **Export CSV** to download the rows.
* Rows are kept for **24 months**, then deleted.

The processing record (**Settings → Processing record**) lists AI processing too, built from the same data, once your workspace has used AI.

## Owner checklist

<Steps>
  <Step title="Check where your AI runs">
    Open **AI Act & residency** and confirm each capability's region meets what you promised your customers.
  </Step>

  <Step title="Decide on the label with counsel">
    Leave it **Off**, turn on the built-in line, or write your own.
  </Step>

  <Step title="Keep the marking intact">
    Use generated images as they are. Re-editing them in another tool can strip the metadata.
  </Step>

  <Step title="Export the log for your file">
    Download the AI decision log CSV with your processing record when you review compliance.
  </Step>
</Steps>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.