Using the ChatGPT Data Agent to Investigate Marketing Performance

AI generated content

Sources reviewed September 15, 2026

A useful marketing report explains what changed and helps a team decide what to investigate next. The ChatGPT Data agent offers a way to ask those questions conversationally, but the quality of the answer still depends on the data, definitions, and checks behind it. Start with a narrow business question before requesting a dashboard.

What the new Data agent does

OpenAI introduced the Data agent in ChatGPT Work on September 10, 2026. It connects to approved company data, supports follow-up analysis, and can build interactive dashboards. OpenAI lists sources including Google BigQuery and Snowflake, and describes using business context to interpret metrics. Administrators control the relevant plugin access. Read OpenAI's Data agent announcement.

Ask a question that leads to a decision

Instead of asking for a complete marketing dashboard immediately, identify the decision your team faces. For example, a subscription business might need to understand why the cost of acquiring activated customers increased last month.

Break that question into possible explanations. Advertising costs may have changed. The mix of campaigns may be different. Visitors may be signing up but failing to activate. These are hypotheses to examine, not conclusions to insert into a report.

A focused question makes it easier to notice when an answer drifts into unrelated data. It also gives the analyst reviewing the output a clear standard for deciding whether the work is complete.

Define the metric before asking for a comparison

Write down what counts as an acquired customer. Specify the date range, time zone, currency, and whether refunds or cancellations affect the result. For subscription products, distinguish a registration from a paying or activated account.

Here is an example analysis request to adapt to approved data sources:

"Compare acquisition performance for the last two complete months by channel. Use our documented activated-customer definition. Show spend, registrations, activated customers, and cost per activated customer. Identify missing data and explain the calculations before suggesting explanations for any change."

That prompt is a proposed workflow. It does not guarantee that every required field or integration is available. If the source data cannot answer the question, the useful response is a clearly described gap.

Check the analysis before acting on it

Reconcile the total spend with an established report for the same period. Then inspect a small sample of customer records to check how the activation definition was applied. Look for duplicate records introduced when datasets were joined.

Ask whether recent customers have had enough time to complete the conversion process. Comparing mature customers from an earlier period with newly acquired customers can create an apparent decline that is largely a timing issue.

Separate observations from explanations in the final report. "Activation declined in one campaign" is an observation. "The new landing page caused the decline" requires additional evidence. A useful dashboard should make the next investigation easier without presenting an untested explanation as a fact.

Questions about AI reporting

Can a marketing team use this without writing database queries

OpenAI presents the Data agent as a way to investigate connected company data through conversation. Teams still need access to suitable sources and reliable metric definitions. The business question can be written in plain language; the result still needs review. OpenAI's product description.

What should the finished report contain

Include the question, calculation definitions, source period, and any unresolved data gaps. Finish with the specific decision the evidence supports. Keep the dashboard tied to that decision so the next review can establish whether the action helped.