ChatGPT BigQuery Connector: Options and Setup
You can bring Google BigQuery data into ChatGPT through exported files, the code interpreter, or a custom connector. Here is what each route involves and how to preserve your analysis decisions across sessions.
Updated 5 min read
You can bring Google BigQuery data into a ChatGPT session through exported CSV files, the ChatGPT code interpreter, or a custom connector that queries BigQuery directly. There is no first-party BigQuery connector in ChatGPT's standard catalog. Understanding the difference between these routes helps you pick the one that fits your analysis workflow. Tools like fenbs can keep the metric definitions and approved analysis assumptions that need to persist across sessions.
This page covers the three main options, the access each one provides, and the limits to check before you rely on one.
Option one: export and upload data
The quickest way to get BigQuery data into ChatGPT is to run a query in BigQuery, export the results as CSV, and upload the file to a ChatGPT conversation. ChatGPT can then analyze the data, generate charts, write summaries, or answer questions about the rows.
This requires no connector setup and works on any plan that supports file uploads. The limit is that the data is static: it reflects the state of BigQuery at the moment you exported it. If you need fresh data for each session, you need to re-export before each conversation.
Option two: use the code interpreter with BigQuery exports
ChatGPT's code interpreter (also called the data analysis tool) can run Python code on uploaded files. You can export a BigQuery result set, upload it, and ask ChatGPT to analyze it with code. This lets you do more complex analysis than a simple text question, including statistical summaries, pivot tables, and visualizations.
The same static data limit applies. The code interpreter cannot query BigQuery directly. It works only with the data you upload to the conversation.
Option three: build a custom connector
A custom connector lets ChatGPT query BigQuery directly during a conversation. Building one requires developer mode in ChatGPT, a server that handles the OAuth flow and routes ChatGPT's requests to the BigQuery API, and a registered connector with the correct redirect URL. Which ChatGPT plans include developer mode is OpenAI's to confirm: check OpenAI's help center (opens in a new tab).
The BigQuery API uses Google OAuth, so the authenticated account must have the appropriate BigQuery roles (at minimum, BigQuery Data Viewer for read access). The connector can query tables, read dataset metadata, and return results. It cannot run queries that modify data unless you explicitly add write permissions. For the broader connector setup steps, see chatgpt connectors api.
Access limits to check
- Query costs: BigQuery charges per byte scanned. A ChatGPT session that runs many queries can accumulate unexpected costs. Set a project budget or use query cost controls.
- Row limits: large result sets may be truncated by the connector or by ChatGPT's context window. Check the size of your typical result set before building the connector.
- Schema access: the authenticated account must have permission to read the dataset and table schema, not just the data rows.
- OAuth expiry: Google OAuth tokens expire. Your connector server must handle token refresh.
Keeping analysis decisions across sessions
An AI session that analyzes BigQuery data often produces decisions: which metrics to trust, which date ranges to use, which anomalies to ignore, and which findings need follow-up. Those decisions belong with the project, not in a chat transcript that will be gone the next time you open ChatGPT.
fenbs keeps those decisions as Decisions and rules linked to the project. A connected AI can read them at the start of a session and apply the approved assumptions without you having to re-explain them. For privacy considerations when connecting data tools, see chatgpt connectors privacy.
A second practical issue is consistency across analysts. If three people on a data team each use ChatGPT to analyze BigQuery data with different assumptions about which date ranges to use, how to handle null values, or which revenue definition applies, their outputs will disagree. Without a shared record of those assumptions, nobody can trace which analysis produced which number, and decisions get made on figures that the next session will recalculate differently.
Writing those shared assumptions as Decisions and rules in fenbs before the analysis work starts means every analyst's AI session begins with the same context. Outputs become more consistent, and when someone questions a number you can point to the decision that produced it rather than reconstructing the logic from a chat transcript that may no longer be available.
Record analysis decisions once, apply them every session
Add your analysis project to fenbs and write the agreed metric definitions, date range rules, and known data caveats as Decisions and rules. Connect ChatGPT from Integrations. At the start of each BigQuery session, ask ChatGPT to read the fenbs context so it applies the same assumptions your team already approved.
Create a free accountFree for one person · No credit card · Works with Claude, ChatGPT, Cursor and more
How to do this with fenbs
- Create a free account. One person is free with no credit card.
- Add a project for the analysis work. Write the goal and add the current analysis tasks.
- Write Decisions and rules for the metric definitions, approved date ranges, and data caveats your team uses.
- Connect ChatGPT to fenbs from Integrations using the MCP setup guide.
- At the start of each session ask ChatGPT to read the fenbs context. Before stopping, ask it to record any new analysis decisions it made with your approval.
fenbs does not connect to BigQuery. It keeps the project brief, decisions, and progress notes for the AI to read. For related connector workflows with other data and project tools, see chatgpt connectors and apps and chatgpt connectors not working.