> ## 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.

# Spark, the in-app assistant

> Ask Spark, the assistant built into InstantCampaign, about your results, your audience and how the product works, and let it draft campaigns, automations, lists and segments for you to approve.

**Spark** is the assistant behind the sparkle button in the bottom-right corner of every dashboard page. It reads your workspace through the same tools the [MCP server](/guides/developers/mcp-server) offers to outside assistants, and it explains the product from these help pages.

## What to ask

* **Results.** "How did the September sale do?", "Which links got clicked most?", "What time of day do people open?", "Which channel brought the most purchases last month?"
* **Audience.** "How big is the VIP segment?", "How many contacts can I actually send to from these two lists?", "Which automations is anyone stuck in?"
* **Setup.** "What should I set up next?", "Is my domain warmed up?", "Which send policy does the transactional API use?"
* **How-to.** "How do I add double opt-in to a list?", "What does the *default to* action do in an attribution rule?" Spark searches the help docs and links the page it answered from.
* **Drafting.** "Create a campaign for our autumn collection", "Draft a welcome series for new subscribers", "Make a segment of people who clicked in the last 30 days but never bought."

## Approvals

Anything that changes your workspace is shown as a card with **Approve** and **Cancel** before it runs. Nothing is sent from the chat: a campaign Spark creates waits on its plan page for someone to approve the send, and an automation it drafts stays a draft until someone activates it.

## What it cannot see

Personal contact data never reaches the model. Names, email addresses and phone numbers are removed from everything Spark reads, and from your own messages, before the model sees them. Ask Spark for counts, statuses and results; open the Contacts page to look at a person.

The footer of the panel names the model that answered and where it is hosted. Your workspace's AI provider and region are chosen by the platform.

## Troubleshooting

<AccordionGroup>
  <Accordion title="It says it cannot find something I can see in the app">
    Spark only knows what its tools return. Give it the name as it appears in the app, or the page you are looking at. If a feature is new, ask a how-to question and it will search the help docs.
  </Accordion>

  <Accordion title="Replies are cut short">
    Ask for one thing at a time. A question that needs several lookups is answered in steps; the small wrench lines under a reply show what it looked up.
  </Accordion>
</AccordionGroup>