Before you connect an ad account to Neuro
Check the permissions you grant, who can access account data, and when a proposed change can run.
On this page
What to check first before connecting an ad account
The moment you connect an ad account to an AI tool, advertisers and agencies tend to worry about the same things: Do we have to hand over our password? Could our data get mixed up with another advertiser’s? Can the AI change our budgets on its own? Is our data being used to train models?
We get delegated permissions, not your password
Neuro does not ask for your ad-account login password. Platforms using OAuth take you to their official sign-in and consent screen; others use dedicated ad API credentials. Check the requested permissions and account scope before connecting.
| Area | The risky way | How Neuro does it |
|---|---|---|
| Login | Hand your ad account password to an outside tool | You approve access yourself, in the platform’s official auth screen |
| Permissions | Grant broad, full-account access | Connect only the scopes the ad API requires |
| Revoking access | You have to ask the vendor to cut it off | Disconnect in Neuro and revoke the permission in the platform yourself |
A public-facing view of the security flow
The model interprets the request and proposes tool calls. Neuro’s server handles account credentials and platform calls. Workspace permissions determine which connections, reports, and files a user can access.
When changes can run
In a regular chat with ad operations enabled, Neuro requests approval for changes. Automatic execution or “Approve all” for a tool changes that behavior. Scheduled operations can also run from the instructions saved in the task.
Review connection permissions and execution settings together. Keep ad operations off for analysis-only work. If you choose automatic execution, specify the targets and allowed changes, then check the results. Prompt instructions do not replace server-enforced permissions.
Does the AI train on your ad data?
Neuro doesn’t collect advertiser data to train its own AI models. The commercial AI API providers Neuro relies on publicly state that, by default, the inputs and outputs sent through their APIs aren’t used for model training. During a security review or contracting, it’s worth confirming this against their latest policy docs.
- OpenAI states that data sent to its API Platform is not used by default to train or improve models.
- Anthropic states that inputs and outputs from commercial products such as the Anthropic API are not used by default for model training.
Even so, each provider’s retention periods, exceptions, and custom contract terms should be reviewed against your own security policy. For enterprise security reviews especially, it’s safer to run through the DPA, sub-processors, data retention periods, and region policies with a dedicated checklist.
Frequently asked questions
Sources
- Google Account Help: Share some access to your Google Account data with third-party apps
Official Google Help article describing how to grant and revoke only the necessary access permissions for third-party apps without sharing your password.
- OpenAI Platform Docs: Data controls in the OpenAI platform
Official OpenAI documentation stating that OpenAI API Platform data is not used by default for model training or improvement.
- Anthropic Privacy Center: Is my data used for model training?
Official Anthropic documentation stating that inputs and outputs for commercial products such as the Anthropic API are not used by default for model training.
Related posts
Can Pango read your Neuro chats?
Neuro chats are locked to support by default. Learn how customer-approved and emergency access work, and what Secret Chat does and does not retain.
Before AI changes your ad budget
How to review AI proposals for budget and bid changes, including approval settings and scheduled execution in Neuro.
Five areas to check in an AI-assisted ad account review
Review goals, campaign structure, budget, conversion quality, and creative while separating verified findings from causes that need more evidence.
When to start a new chat for ad work
Decide whether to continue or split an ad-analysis chat when accounts, goals, and reporting assumptions change. Includes prompts for carrying verified findings into a new task.
When AI groups campaigns incorrectly: check names and UTMs
Find campaigns whose names leave product groups or audiences unclear, then use settings, UTMs, and a mapping table to classify them.