September 29, 2026 · 5 min read
Before an AI Changes Your Ads, Ask for This Approval Card
Use a worked approval card to inspect the exact account, proposed settings, risks and verification plan before an AI changes your advertising.
By the AdvisorPPC Team · Reviewed by Claude
An approval is useful only when you understand the action it authorizes. "Improve my account" leaves too much unspecified: the account, the settings, the timing and the amount of money affected. A practical alternative is a short change card you can read without understanding an API.
This article uses one hypothetical budget reallocation to show the information an owner should demand. It is a buyer's review method, not a claim that every connector currently enforces every control described here. The broader guide Approval Gates for AI PPC Agents explains how the surrounding workflow should be designed.
Begin with one decision and a named account
Imagine two campaigns for a local business. Campaign A has an average daily budget setting of $60 and Campaign B has $40. An assistant recommends changing them to $50 each because B has produced more qualified bookings recently.
The change card should name the business account, both campaigns and the currency. It should say that $10 moves from A's daily setting to B's. "Rebalance by 10" is ambiguous: ten dollars, ten percent and ten units of another currency are different instructions.
The proposal should also identify shared budgets or other dependencies. If either campaign uses a shared budget, the apparent two-row edit may affect other campaigns. The reviewer needs the actual object being changed, not just a friendly campaign label.
Require the calculation and its limits
In this hypothetical example, A spent $420 and produced six paid bookings, a $70 cost per paid booking. B spent $280 and produced seven paid bookings, a $40 cost per paid booking. That difference makes B worth investigating. Thirteen bookings are still a small observation set, and the figures do not prove that extra budget will produce bookings at the same rate.
Assume the average contribution before acquisition is $100 per booking. A's observed contribution after advertising is $180: six times $100, minus $420. B's is $420: seven times $100, minus $280. The card should state that the contribution estimate comes from the owner and that attribution has not established incrementality.
Ask whether B has additional capacity, whether both periods are mature, and whether a tracking change affected the results. A cheap booking for a fully occupied time slot is not automatically a better business outcome.
Make the proposed state readable
| Field | Hypothetical approval card |
|---|---|
| Purpose | Test a small allocation toward qualified booking demand |
| Account | The selected business account, confirmed by the owner |
| Current state | A: $60 average daily setting; B: $40 |
| Proposed state | A: $50 average daily setting; B: $50 |
| Scope | These two budget objects only, dependencies checked |
| Evidence | Matched-period paid-booking reconciliation |
| Review time | After a predefined observation period with conversion lag considered |
| Stop condition | Tracking failure, unexpected scope, capacity issue or agreed risk threshold |
| Recovery | Reinspect current state, then propose restoring recorded settings |
The $100 sum of daily settings is not a proof that daily charges cannot exceed $100. Budget mechanics and campaign types matter. Read Why Google Ads Spends More Than Your Daily Budget before presenting a settings total as a hard spending cap.
Recheck immediately before execution
Approvals age. Another person may change a budget, pause a campaign or fix the measurement after the card is prepared. A robust workflow should re-read the current values before applying the approved change and stop if the relevant state has changed.
For the owner, the question is simple: "Does this approval apply only to the values on this card?" If the answer is no, ask how the scope is bounded. A broad, persistent permission should have explicit allowed actions, account scope, limits and expiry.
If you decline the card, that should end this proposed action. It should not be treated as permission to apply a smaller variation automatically. Approval language should be specific enough that both a person and a system can identify what was accepted.
Ask what happens when only part succeeds
Google's API supports several mutation methods with different execution behavior. Bulk mutations can be configured to commit valid operations independently, while batch jobs use partial-failure behavior. The actual request path therefore matters when someone claims a multi-step action is all-or-nothing. Mutate resources.
The change card should describe the recovery process if A updates but B does not. It should not quietly retry an uncertain operation until the desired screenshot appears. The receipt must identify which operations succeeded, failed or remain unresolved.
This is why a status such as "submitted" should remain separate from "verified." A request can be accepted while the business result is still unknown.
Close with a receipt you can independently inspect
After execution, ask for the resulting settings, verification time and any partial failures. Keep the original card beside the receipt. Google's change history records campaign changes, including those made through its API, and offers undo for eligible changes rather than every possible operation. Review your account history.
Use the guide PPC Change Receipts: Prove What Automation Published to decide what your team retains. Restoring a setting does not refund spend already incurred or erase the consequences of a published ad, so recovery should never be described as risk-free.
Before choosing paid editing capability, compare AdvisorPPC's current access scope with the operations you need. Ask to see the proposal, approval and verification behavior for your actual client. A useful AI assistant makes a decision easier to inspect before it makes that decision easier to execute.