← Back to blog

Safe AI integration with an ad account — 2026 checklist

3 fundamental principles

The entire AI ↔ ad account integration rests on 3 principles. Without them, the rest doesn't matter:

1. Least privilege (give the minimum permissions needed)

The agent gets only as much access as it actually needs for its task — and nothing more. Not "full admin access just in case". Read-only by default. Write access only when there is a specific, documented, one-off task and a separate go-ahead from you. If the agent "reports" — it doesn't need permission to edit budgets. Ever.

2. Separation of duties

Different tasks = different agents = different tokens. One "super-agent" with access to Google Ads, GA4, Merchant Center, CRM and email is risk centralization. When it falls over — everything goes. Therefore: the reporting agent separately, the data extraction agent separately, the ChatGPT analysis agent separately (and that last one is best run WITHOUT a live account connection).

3. Audit by default

Every agent action is logged automatically — with identification of "who" (agent), "for whom" (the user whose permissions it uses) and "what" (action). The audit log is not an option you enable when there's a problem. The audit log is the foundation without which you'll never detect a problem before you see the financial impact.

OAuth scopes — granting the minimum

An OAuth scope is simply the "permission range" an app asks for when it logs in. The Google Ads API has one "full" scope (https://www.googleapis.com/auth/adwords) — read + write across the entire account. Grant this scope only when you genuinely need to give the AI write access. In 95% of cases an AI agent only needs to read reports.

Practical approach (at the application + user level):

Agent use case	What to grant
Reports / dashboards	Read-only user on the account + adwords scope (read-only app)
CSV data extraction	Read-only user on the account
ChatGPT analysis via copy-paste	No integration. Copy CSV, paste into ChatGPT — zero OAuth
Auto budget optimization	Standard user + extra restrictions (max daily % change)
Creative editing	Standard user + sandbox first, then production
Everything + adding users	NO. Admin only for the human in charge

Google Ads role hierarchy:

• Admin — full access + adding users. Only for you (the owner) or one trusted person
• Standard — edit campaigns, no right to add other users
• Read-only — read only
• Email-only — notifications only (no panel)

For an AI agent: Read-only in 95% of cases. Exception is an agent that actually executes optimizations — then Standard with additional restrictions (max daily % change, list of allowed action types, daily report to you).

RBAC for team and agency

RBAC (Role-Based Access Control) is simply "who has what access to the account". In Google Ads that's user management. A pattern that works:

Role	Google Ads access	Notes
Business owner (you)	Admin	One Admin per account, ideally with a hardware key (e.g. YubiKey — a physical USB stick used for login) on the Google account
Agency / freelancer running the account	Standard	Campaign edits, no admin
Agency junior / analyst	Read-only	Can learn, can't break things
Internal analyst	Read-only	CSV export → analysis
AI agent (reporting)	Read-only (separate account / token)	NOT on your personal account
AI agent (optimization)	Standard + restrictions	Only after a full audit

RBAC red flags I still see too often:

• One shared admin account for the entire agency — when 1 person leaves, you must disconnect EVERYONE and rebuild access
• AI agent on your personal admin account — when the agent (or its vendor) falls over, you fall over with it
• "Forever" tokens — every integration should have a timeout, even if it means re-authorising every week
• Service-account email as a personal Gmail — a service account (a separate Google account used only for integrations, not for daily work) should sit on your business domain so you can reclaim it if someone leaves
• No rotation — a quarterly review (who still needs access?) is the minimum, not the maximum

Audit log — what to monitor

Google Ads has a built-in Change History (Tools & Settings → Account → Change History). Everything that happens on the account is logged there. The question is not "do I have a log" — it's "is anyone looking at it".

Practice that works:

1. Monthly export of Change History to Google Sheets (Apps Script or manually) — you keep history even if Google ever truncates it
2. Alerts on unusual changes: budget spike > 20% in 24h, mass campaign pauses, new user added, new OAuth token issued
3. Weekly review — 10 minutes on Friday, scan the week's changes. Do I see anything not from me or the agency?
4. Hourly anomalies — changes at 3 AM on a Polish account are suspicious by default

What specifically to monitor:

• Budget changes (> 20% in 24h = alarm)
• Conversion changes (adding / removing goals — critical)
• User changes (new user = alarm, verify the email)
• API access changes — Google Ads Change History logs campaign / budget / conversion / user changes. New OAuth token issuance is not logged there — check it at myaccount.google.com → Security → Activity
• Mass pause / enable of campaigns (bot pattern)

Without an audit log, you won't detect a leak until you see the financial impact — which is already too late.

10-point pre-integration checklist

Before you connect an AI agent to a Google Ads account (or anything valuable), go through 10 points. Every ✓ is a concrete action, not philosophy:

• Minimum scope (read-only by default). If the vendor asks for write — ask "exactly why?". Without a concrete answer → NO.
• Dedicated service account. Not your personal Google account. Set up a separate Google account on your business domain, with its own password and a hardware key (e.g. YubiKey — a physical USB stick used for login), used only for the AI integration.
• Short-lived tokens (max 24h, auto-rotate). If the vendor says "token forever" — red flag.
• IP allowlist where possible. The agent runs from specific IPs — list them in Google Cloud Console.
• Audit log enabled + monitoring. Weekly review minimum.
• Quarterly access review. Each quarter: who still needs access? Anything to remove?
• Incident response plan (see section below). Written before the problem, not during.
• Sandbox environment for tests. Test the agent on a separate Google Ads account (test account) before releasing to production.
• Code review of the integration (if custom-built). An independent pair of eyes on code that touches your money.
• Vendor due diligence. Ask for: SOC2 Type II certification and GDPR compliance (these are standard security credentials you should require from any serious vendor), server location, incident history, breach response plan.

If even 2 points are unmet or the vendor dodges specifics — DON'T connect. This is not pessimism, it's the procedure you routinely expect from banks and insurers.

Incident response — what to do when it leaks

It will leak someday. Maybe not your token, maybe a vendor on your side, maybe an agency. The assumption "this won't happen to me" is the most common reason a first response takes 4 hours of mistakes rather than 5 minutes of action. A plan worth having printed:

Step 1 (first 5 minutes) — Revoke the OAuth token

OAuth tokens for external apps (AI agents, SaaS tools) are managed at the Google Account level, not inside the Google Ads panel. Go to myaccount.google.com → Security → Third-party apps with account access → find the suspect app → Remove access. Revoke any tokens you're unsure about — better too many than too few (you can re-authorize later).

Additionally in the Google Ads panel (Tools & Settings → Account Access) remove any service accounts / users added as part of the integration — that's a separate access mechanism from OAuth tokens.

Step 2 (15 minutes) — Secure the Google account

Change the password on the main Google account + enable a hardware key (e.g. YubiKey — a physical USB stick used for login) if you don't already have one. Also check two-step verification (2FA) — do you have backup methods that aren't the same phone as the primary one?

Step 3 (30 minutes) — Review Change History

Google Ads → Tools & Settings → Account → Change History → last 7 days. Are there unusual changes? Check especially: budgets, conversions, users, campaigns active / paused.

Step 4 (1 hour) — Roll back changes, if any

If you spotted unintended changes — undo them manually. Google Ads Change History shows "before → after", so you know what to restore. If many changes — export changes + restore from a saved copy of the campaign (if you have one).

Step 5 (24 hours) — Notifications and lessons learned

If it's a client account (agency → client) or your account with client data (e-commerce with conversion data) — notify the client. If the leak was on the vendor's side — notify the vendor and demand a post-incident report. Write down lessons learned: how it happened, how long it took to detect, what I'm changing so it doesn't repeat.

Keep this plan in a visible place (Notion / team document / printed). Searching for a procedure during a panic is worse than not having one.

About the sources: Article based on Cowork — AI Agent Permissions 2026, Aicuflow — Enterprise AI SSO/RBAC/Audit, Stacklok — Best MCP platforms 2026, RootCX — AI Agent Governance, WitnessAI — Agent Access Control, Arcade.dev — Secure AI Agent Connections + observations from JP Ads client accounts. Numbers from my practice are labelled "from my experience".