> ## 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.

# AI data privacy

> What InstantCampaign's AI features and connected AI assistants can see of your contacts (aggregates only), how sensitive fields work, where the models run and what the providers' terms say.

InstantCampaign's AI features never receive the personal data of your contacts. Email addresses, names, phone numbers, postal addresses and custom-field values stay out of every prompt and every tool result an AI model reads. The one feature that works per contact, [AI personalisation](/guides/audience-and-campaigns/ai-personalisation), sends coarse traits only (below). The public summary is at [instantcampaign.ai/ai-privacy](https://instantcampaign.ai/ai-privacy).

## What the AI can see

| Feature | What reaches the model |
| - | - |
| **Spark** (in-app assistant) | Tool results with counts, rates, statuses and ids; the help pages; your own messages with any address or phone number masked |
| **Connected assistants (MCP)** | The same tool results as Spark. Tools that act on a contact you name take its address and answer with its id, never the address |
| **Campaign strategist, weekly proposals** | Your brief and brand kit, list and segment names with their sizes, recent send results, A/B learnings, store totals and product titles |
| **AI segment builder** | Your description (addresses and phone numbers masked), the field catalogue, list and campaign names |
| **Design Assistant, AI landing pages, copy and subject lines** | Your brief, the template or block you are editing, your brand kit |
| **AI personalisation** (automations) | The AI text block's brief and fallback, and a trait profile shared by every contact who matches it: engagement tier, language, country, time zone, purchase-count and recency buckets, list and segment names, the option a contact has in a non-sensitive dropdown or yes/no field the step allows, and short trigger values the step allows. Never an address, name, phone number, company, free-text or sensitive field; the first name is added by the template after generation |

A template preview that Spark or an MCP assistant reads is rendered for a stand-in contact called "Preview Contact", never a real one, with any name or address in the sample data masked. A message-log row names the recipient's domain and contact id; its subject is scrubbed of that recipient's own stored values.

Apart from AI personalisation's trait profiles, this is fixed at **aggregates only**: there is no setting that widens it.

## Sensitive fields

Mark a custom field **sensitive** on the [Custom fields](/guides/audience-and-campaigns/custom-fields#sensitive-fields) page. Its values already never reach a model; marking it also withholds its options and description from every tool output, and any segment rule or automation condition that uses it reads `[sensitive]` to a model. Emails, merge tags, segments and the API use it as normal.

## See your workspace's settings

**Settings → Privacy → AI & data** shows:

* the contact context (aggregates only);
* the fields marked sensitive;
* which text and image models the platform uses and where they are hosted;
* whether a residency mode (UK, EU, or UK + EU) is enforced.

The page is read-only apart from **AI personalisation at send time**, a switch (on by default) that can turn AI personalisation off for the whole workspace. Models and residency are chosen by InstantCampaign for the whole platform.

## Where the models run

Two providers, picked per capability: Google's Gemini API (global) and Amazon Bedrock (Claude and Nova in the UK/EU, Stability image models in the US). The pick is binding: a failed call returns an error instead of moving to the other provider, unless the platform has allowed fallback. Under an enforced residency mode, a model hosted outside the boundary is refused, and so is the fallback to Gemini.

## Training and retention

* **Gemini API, paid services**: Google states it does not use prompts or responses to improve its products and logs them for a limited time to detect policy violations ([terms](https://ai.google.dev/gemini-api/terms)).
* **Amazon Bedrock**: AWS states inputs and outputs are not used to train models and are not shared with model providers ([FAQs](https://aws.amazon.com/bedrock/faqs/)); some models keep traffic up to 30 days inside AWS for abuse detection ([abuse detection](https://docs.aws.amazon.com/bedrock/latest/userguide/abuse-detection.html)).

InstantCampaign keeps your AI generation history (prompt and result) for as long as the workspace exists. Spark conversations are not stored on our servers. MCP calls are kept for 90 days in the [API request log](/guides/developers/api-request-log), which only API-key managers can open.

## Limits

* Text you write (a brief, a template, your brand voice) is sent as written. A customer's name typed into a brief reaches the model.
* The REST API returns contact data to integrations you authorise with a key; it is for your systems, not for models.
* A connected MCP assistant runs on its own provider's model and terms.


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