Approve AdPilot proposals, then apply: risk levels and who clicks
In this article
AdPilot doesn’t write directly to your Google Ads account: each change goes through a proposal you approve (source). This article covers the steps a proposal goes through, how the risk level changes the way you approve, who in your organisation can click, and why you might still be view-only after pasting the Full address.
1. The five steps of a change
According to the Docs page, every change takes exactly one path, with no detour for “urgent orders” (source):
| Step | What happens |
|---|---|
| 1 · Propose | The AI drafts the change and says exactly what it intends to do, on which account. Nothing has touched Google yet. |
| 2 · Pre-check | The change is checked against safety rules, the limits you set and the account’s actual state right now; accounts in finance or crypto get an extra check against that industry’s policy rules. |
| 3 · Approve | A real person agrees or declines. |
| 4 · Apply | The only place in the whole system that writes to Google Ads. |
| 5 · Undo | If it’s wrong, you can roll it back — and know exactly what’s being rolled back. |
Each proposal comes with a reason and a risk level (source).
2. Pre-check: when a proposal is blocked
A blocking issue means the proposal can’t be applied, unless an owner or approver overrides it with a reason (for block types that allow an override) (source). For anything it flags, AdPilot explains why and suggests a fix (source).
3. The risk level decides where you click
Low-risk changes: you agree to them yourself in the chat. High-risk changes: an owner or approver approves them on the web.
- Low risk: you agree right in the chat (source).
- High risk: someone has to open the approval page on the ALFA web app and click by hand; the AI can’t approve it itself (source).
- Nobody approves: the proposal touches nothing and expires on its own after 24 hours (source).
Whether a given change is low or high risk: check the risk level shown on that proposal.
Using AdPilot for Meta Ads? It uses the same approval gate: see Approve and undo Meta Ads changes in AdPilot.
4. Who in your organisation can click
An organisation has three roles (source):
| Role | What they do |
|---|---|
| Owner | Can edit invoicing details, order plans, and decide who joins the organisation. |
| Approver | Approves or rejects proposals before they touch an ad account. |
| Operator | Gives instructions and creates proposals. Can apply low-risk changes by agreeing in the chat with the AI; a high-risk change needs an owner or approver to approve it on the web. |
The owner or an approver is also the one who clicks “Allow changes” for each newly linked account, because new accounts are read-only (source). See Connect Google Ads to AdPilot.
Organisation roles are separate from platform roles: ALFA administrators run the platform, and they have no approval rights inside your organisation (source).
5. Apply and read the result
Apply is the only place that writes to Google Ads. AdPilot reports the result item by item — if only part succeeds, it says exactly which items failed — and records it in the log. Each apply goes into the log: who approved, when, and how it turned out; for Google Ads, runs that failed or were blocked are recorded too (source).
Mistakes can be rolled back. Where something can’t be reversed, such as deleting a campaign, AdPilot says so (source). An undo is applied automatically, with no wait for approval, when the real level of the call is “Full”, the account hasn’t been edited since the change was applied, and either the reverse change is low-risk, or the original change was low-risk and was applied no more than 24 hours ago. Removing an asset group that is running or has already had impressions always waits for an approver, even within 24 hours. Any other undo, and any undo on an account edited since, waits for an approver like any other proposal (source).
6. Your organisation’s maximum level in Guardrails
Pasted the Full address and can still only view? According to the Connect AI page, that isn’t a bug. The real level for each call is the lowest of four things (source):
- the permission you gave the AI app,
- the address you’re using,
- your organisation’s maximum level in Guardrails,
- the plan you’re on.
The most common case: your organisation set a Guardrails maximum below the address you pasted — then the Guardrails level wins (source). The Full address reads, creates proposals and applies; high-risk changes still need a person to approve them on the web (source).
Why a single door
With two write paths you have to guard two places, and the one nobody watches is the one that breaks. A single door is easy to guard, easy to log, and when something goes wrong there’s only one place to look (source). See also Use AI to run Google Ads and keep approval.
Checklist
- The organisation has at least one owner or approver to click high-risk proposals
- The account to change has had “Allow changes” clicked
- Read the risk level and reason on the proposal before agreeing
- Proposal blocked at pre-check: read the reason AdPilot gave
- After applying: checked the result item by item
- Still view-only: checked the four things, Guardrails maximum first
AdPilot for Google Ads · Docs: The approval gate · Connect your AI
💬 Need help? Join the ALFA Software support group on Telegram (ask in your own language) — or message our support bot privately about your account or invoices.
Checked against ALFA’s Docs, AdPilot for Google Ads and Connect AI pages on 8 Nov 2026.
Sources
- ALFA Software — Docs (The approval gate; Accounts, organisations and roles): https://alfasoftwares.com/en/tai-lieu#cong-duyet
- ALFA Software — AdPilot for Google Ads: https://alfasoftwares.com/en/adpilot-google
- ALFA Software — Connect AdPilot for Google Ads and Meta Ads to your AI: https://alfasoftwares.com/en/noi-ai