An active Experiences account.

API credentials, a Client ID and Client Secret. Contact your account team if you don't have them; they are issued per brand.

Claude Desktop.

If your credentials cover more than one brand, you also need the Business ID of the brand you want to work with. Ask your account team, or leave it blank if you only have one brand.

Which campaigns are live, and how are they performing this quarter?

Where do people drop off in the Product Finder quiz?

Which recommendation do respondents see most often?

Where does this campaign's data go, and what fields are we sending?

Is our nightly SFTP export still running, and how many rows did it send?

Which attribute does the skin_concern mapping on our Klaviyo integration read?

Campaigns — list them, find one by name, and get its launch links and distribution channels.

Performance — views, engagements, completions, leads, opt-ins, time spent, and the derived engagement, completion, and lead rates, over all time or a date range you specify. Ask about a single campaign and you also get which outcome respondents were shown and how often, and how many people reached each question versus answered it.

Integrations — every integration configured for the brand, with the service it sends to, the destination host, the field mappings applied on the way out, and any static values injected into the payload.

Scheduled exports — the batch exports that ship session data on a cadence to SFTP, S3, Google Cloud Storage, Azure, or email, including their schedule, which fields they send, and the last ten runs with row counts and status.

Attributes — the attributes your experiences collect, which is how a field name in an integration mapping resolves to the attribute it reads.

Drop-off includes results screens. The per-question breakdown covers every screen a respondent interacts with, not only questions. Each row reports a resource_type. On an Outcome or Reward screen, a low response rate means people saw their result and didn't click a call to action. That's normal, not a place the campaign is losing people.

Large accounts are summarized. Where an account holds more records than fit in one answer, the connector states that and reports the true total rather than quietly returning a subset. If Claude mentions a partial result, ask it to narrow the question, by campaign name, launched status, or a specific attribute, rather than asking for everything again.

It cannot change anything. Every operation is a read.

It never returns credentials. Integration and export secrets are not read at all. Destinations are reported as a host rather than a full endpoint, since endpoints often carry tokens. An email export reports how many recipients it has, not their addresses.

It does not expose individual respondents. There is no access to individual submissions, to the answers a particular person gave, or to any personally identifiable information (PII). All reporting is aggregate counts.

Confirm your credentials carry the scopes needed for reporting, scheduled exports, and attributes; contact your account team to request missing scopes.

Explore the example prompts above to get familiar with what the connector can answer before relying on it for a specific report.

Your credentials might not carry the scopes for what you asked. Reporting, scheduled exports, and attributes each need their own permission granted to your credential. Contact your account team with the name of the tool that failed.

Your credentials cover more than one brand and no Business ID is configured. Add it in the connector's configuration.

Check that the Business ID matches the brand that owns the campaign.