- Security
- AI
- Google Ads
7 questions to ask before giving AI access to your ad account
A tool reading your account is one thing; a tool changing your budget is another. Seven questions to answer before you give AI access to Google Ads or Meta, what a good answer looks like, and how to grant access step by step.
On this page8
Before you give AI access to your ad account, get seven answers: which permissions the tool requests, who approves each change, whether proposals are validated before they go live, who changed what and when, how you revoke access, whether your data trains AI models, and where the limits are when something breaks. If a tool can't answer in writing, don't connect your Google Ads or Meta account.
In short
- Reading and changing are different rights. A tool that only builds reports doesn't need permission to change anything.
- The key question is who decides: every change should arrive as a proposal and go live only once you approve it.
- Every approved change should be logged with who proposed it, who approved it and when, in a record that can't be edited.
- You should be able to revoke access in one step, from the tool and from Google's and Meta's own settings.
- Your ad data shouldn't train AI models, and the privacy policy should say so plainly.
Why reading and changing aren't the same thing
An ad account is one of the fastest ways a company spends money. Imagine an extra zero slipping onto a daily budget that should be $50: by the time you wake up, the campaign may have spent several times what you planned.
So the question isn't “does it work?” but “what happens when it gets something wrong?” A tool that only reads can misread your data; a tool that can change your account turns its mistakes into spend. The seven questions below apply to any tool, each with what a good answer looks like.
Access: what can the tool actually touch?
1. Which permissions does it ask for, and why?
Google and Meta grant third-party access through OAuth: their own consent screen opens and lists what's requested. A tool that asks for your password fails right here. Then look at the scope, where the two platforms differ:
- The Google Ads API uses a single OAuth scope. There's no separate read-only permission. What the tool can do depends on the Google Ads access level of the person who connects it. Connected with Read-only access, it can pull reports but not edit campaigns; with Standard or Admin access, it can do whatever that person can.
- On Meta, reading and managing are separate permissions. If a tool asks for both, ask what it needs management rights for.
- For GA4, Tag Manager and Search Console, read-only is usually enough. Working out why GA4 and Google Ads report different conversions doesn't require write access.
A good answer: The tool explains each permission. If it only reports, it doesn't ask to change anything; if it needs more later, it asks then and says why.
2. Who approves a change?
This is the most important question. “AI automatically optimizes your campaigns” can hide who is actually making the call.
A good answer: Every change arrives as a proposal showing what will change, the old and new value, the reasoning and the expected impact. Nothing goes live until you, or someone you've authorized, approves it. Pending proposals also expire, so one drafted a week ago isn't silently applied under today's conditions.
Assurance: how are changes applied and tracked?
3. Is a proposal validated before it's applied?
A model's suggestion may break the platform's rules or target the wrong account. And after drafting, a colleague may have edited the same budget by hand.
A good answer: The proposal is first run against the platform in validation-only mode. The Google Ads API can check a change without applying it, so errors are caught before the approval screen. At approval, the current value is read again: if the budget changed in the meantime, the old proposal isn't applied.
4. Who changed what, and when?
“Who raised this budget?” shouldn't be buried in a chat history. Google Ads keeps a change history for the past two years, but for API changes the user column may show the API or the tool's name rather than a person. The platform says “a tool changed this”, not who approved it inside that tool.
A good answer: The tool logs every applied change with who proposed it, who approved it, when, and the before and after values. The log is append-only and can't be edited or deleted.
The platform's change history tells you what changed. The tool's log should tell you who decided.
Exit and data: who stays in control?
5. How do I revoke access?
Ask this before you connect. There should be three ways out:
- In the tool: disconnect in one step; the access token is deleted and revoked on the platform's side.
- In Google: open the linked apps page of your Google Account (myaccount.google.com/linkedapps), select the tool and remove its access.
- In Meta: remove the app under Business integrations in your Facebook settings, and review access granted through your business portfolio.
A good answer: All three work, and the tool stops cleanly when access is removed elsewhere.
A trustworthy tool makes it easy to take access back. If it makes it hard, that's an answer too.
6. Is my data used to train models?
Your ad data is a trade secret: what you pay for which searches, which audiences convert. The question covers both the tool and its model provider.
A good answer: Platform data is used only to serve you, and neither the tool nor its model provider uses it to train AI models. The privacy policy says so and, for Google API data, confirms compliance with Google's Limited Use requirements. Where data is stored and which subprocessors (hosting, AI, email) can access it is published.
7. Where are the limits if something goes wrong?
One person's slip shouldn't affect the whole account.
A good answer: Roles decide who can approve, and some roles can see reports without changing anything. The tool can apply only a defined set of action types, and when something fails it says what happened and what state the account is in.
The seven-question checklist
Take this table into your evaluation call. If you hear a red flag, ask for the answer in writing.
| Question | Good answer | Red flag |
|---|---|---|
| 1. Permissions | Each justified, only what's needed | Asks for your password |
| 2. Approval | Every change is a proposal | “Optimizes automatically” |
| 3. Validation | Tested on the platform first | Errors appear after going live |
| 4. Audit log | Proposer, approver, time; uneditable | Only a chat history |
| 5. Revoking | One step, three ways | Requires a support ticket |
| 6. Data | No model training, in writing | Vague policy wording |
| 7. Limits | Role-based approval | Anyone can change anything |
How to grant access step by step
Even with good answers, don't grant the widest access at once:
- Choose who connects. In Google Ads, use a user with no more access than the job needs. For reports only, Read-only is enough.
- Read the consent screen. It should open on Google's or Meta's domain and match what the tool told you.
- Read only for the first week. Ask it to find budget leaks in your search terms report, and compare its conclusions with someone who knows the account.
- Keep first proposals small. Start with changes that are easy to undo, such as a few negative keywords.
- Cross-check the log. Compare the tool's log with the Google Ads change history; every change should appear in both.
- Assign roles. Give approval rights only to decision-makers; everyone else can view.
How ad10 answers these questions
We asked ourselves these questions while designing ad10. The flow is always: proposal, validation, approval, log.
- The assistant reads your accounts through the official APIs and drafts proposals with the old and new value and the reasoning.
- Google Ads budget and negative keyword proposals are first run in Google's validation mode and marked “Validated by Google” on the approval screen. If the budget changes after drafting, the proposal isn't applied.
- Only people with at least the Editor role can approve; Analyst and Client roles can't make changes. The same applies through the Claude or ChatGPT connector.
- Every applied change goes into a log that can't be deleted, with who proposed and who approved it.
- ad10 requests read-only access to GA4, Tag Manager and Search Console. Access tokens sit in an encrypted vault and are deleted when you disconnect.
- The database is in Frankfurt, and every provider we use is on our subprocessors page. Your data isn't used for model training; see our privacy policy.
See the approval flow in our security section and supported accounts in the platforms list.
Frequently asked questions
Can an AI tool change my Google Ads account without my approval?
Technically, a tool connected to the Google Ads API can do anything the connecting user's access level allows, so ask whether approval is mandatory inside the tool.
Which access level is enough if I only want reports?
In Google Ads, Read-only access is enough for reporting, and a tool connected with it can't edit campaigns. On Meta, reading ad reports is a separate permission from managing the account.
Will my campaigns stop if I revoke access?
No. Revoking access only cuts the tool off; campaigns, budgets and applied changes stay as they are.
How can I tell whether a change was made by a tool?
In the Google Ads change history, API changes may show the API or tool name as the user. For who approved it, check the tool's own log.
Is it safe to share my ad data with an AI tool?
It can be, if the policy states that your data only serves you, isn't used for model training, where it's stored and who processes it. If it doesn't, ask in writing.