# We Gave AI Write Access to Ads. Never Delete Access

The safety line for AI in ad accounts is not read vs write — it is recoverable vs permanent. Drawing it there makes full write access rational.

Blog · Published 2026-08-10, updated 2026-11-11.

> An AI agent managing Google Ads should hold write access but never delete access, because the meaningful safety line is what can be put back, not read-versus-write. Pausing, budget changes and bid changes can all be put back; deletion cannot. AdCopilot enforces the line structurally: of the 63 Google Ads tools implemented in its fork of Google's MCP server (as of v2.46.25), the four remove tools are never exposed to hosted connectors, and any change carrying status REMOVED is refused on the server — in any letter case.

Most debates about AI in ad accounts argue the wrong line. One side wants
read-only access, which keeps you safe and keeps the AI useless for half its
job. The other wants full autonomy, which nobody responsible for a budget
actually ships. The line that matters runs elsewhere: between actions that
can be undone and actions that cannot. Draw it there, enforce it in the
server rather than in the prompt, and full write access stops being brave —
it becomes rational.

## The wrong debate: read-only vs full access

Read-only is a real position with a real cost. An agent that can only read
your account is an analyst who ends every finding with "now go do it
yourself" — the wasted spend it found stays found and unfixed until a human
retypes the fix into the interface. Most of the value of connecting an AI to
[an ad account at all](https://adcopilot.cloud/ai-agent-for-google-ads) sits in closing that loop:
the same conversation that finds the leak kills it — the negatives go in as
you speak, in seconds.

Full access is not a position so much as a shrug. An agent that can do
anything includes an agent that can do the one thing no approval workflow
reliably survives: destroy something permanently on a misread. The debate
stalls because both camps are answering "how much should AI do?" when the
useful question is "which mistakes must remain impossible?"

## Recoverable vs permanent: the line that matters

Sort every Google Ads action by its way back and the safety problem gets
small:

| Action | Way back | What survives a mistake |
| --- | --- | --- |
| Read anything | Nothing to put back | Everything |
| Add keyword, negative, asset | Remove or pause the addition | Everything |
| Create campaign (arrives paused) | Never enable it | Everything — it cannot spend |
| Pause / enable | Flip it back | Structure, history, learning |
| Budget or bid change | Set the old number back | All but the interim spend |
| Delete (status REMOVED) | None | Nothing — it is gone |

Every row except the last is recoverable, and the worst recoverable case —
a bad budget change running for a day — costs money but destroys nothing.
The last row is different in kind, not degree. In the Google Ads API there
is no delete button as such: setting an entity's status to REMOVED *is* the
deletion, and it is permanent. A safety model that treats that row like the
others, just with a sterner confirmation dialog, has misunderstood it.

The asymmetry deserves one more sentence, because it is the whole
argument. A recoverable mistake costs its interim damage; a permanent
one costs the entity plus its history plus every future decision that
history would have informed — a deleted campaign takes its performance
record, its search terms archive and its learning with it. You are not
weighing two amounts of risk. You are weighing a cost against a loss.

## What pause, budget and bid changes have in common: a way back

The writes that matter day to day — pausing a bleeding campaign, raising a
budget before a promotion, tightening a target — share a property worth
naming: each one is a state you can restore. That property is what makes
[an approval loop](https://adcopilot.cloud/blog/why-your-ads-agent-needs-an-approval-loop)
sufficient for them. Approval is a judgement checkpoint, and judgement
failures on recoverable actions are correctable failures: notice, put it
back, move on. The account's history, structure and learning are all still there.

This is why "the AI changed my budget wrongly" is an incident, while "the
AI deleted my campaign" would be a catastrophe — and why the two must not be
governed by the same mechanism. Incidents are managed with process.
Catastrophes are prevented with architecture.

## Why the delete tools are absent, not forbidden

As of v2.46.25, AdCopilot's underlying server implements 82 Google Ads tools: 63
in its fork of Google's MCP server, and 19 Google Ads tools AdCopilot runs itself
(read-only audits and analyses, a change history and an explanation of why Google Ads and Analytics count differently, and one write, rating a Local Services lead). Hosted connectors are
given 77 of them. The five missing ones are the
`remove_*` tools that delete a campaign, an ad, a keyword or an asset, and the gap is
the design: they are not disabled, not hidden behind a setting, not waiting
for an enterprise tier. They are absent from the tool list an agent ever
sees.

Behind that sits a second, independent layer. If any mutate call arrives
carrying a status of REMOVED — whatever the tool, whatever the letter case,
whatever creative phrasing produced it — the server refuses it before it
reaches Google. And the tools that set a status — on a campaign, an ad group,
an ad, a keyword or an attached asset — refuse REMOVED themselves, with a
message that says to pause instead. Three layers, all server-side, none
dependent on the model's cooperation.

The distinction matters because the alternative — a rule in the system
prompt saying *please never delete* — is a politeness convention, not a
safety property. Prompts can be overridden by [injection, confusion or
plain model error](https://adcopilot.cloud/questions/can-ai-delete-my-google-ads-campaigns);
capability absence cannot. It also survives the failure modes nobody
designs for: the new teammate's over-eager prompt, the model update that
reasons differently, the pasted spreadsheet that happens to contain the
word "removed" in a status column. Server-side enforcement does not care
how the request was produced — only what it asks.

You can verify the difference yourself in one minute: ask a connected
agent to delete something, deliberately, and watch the refusal come back
from the server. A system's real safety model is what it does when asked
to do the wrong thing.

## Defence in depth: what surrounds the line

The no-delete line is the floor, not the whole building. Around it,
[the same architecture](https://adcopilot.cloud/security) stacks controls that each fail
independently:

- **Proposals first.** AdCopilot proposes each change first, and your AI client asks before each write, showing the exact tool, account and change, if it is set to ask before AdCopilot's write tools. Reads change nothing,
  so they can be allowed once and left to run.
- **Audit trail.** Every call is recorded: tool, account, success or
  refusal, reason, timestamp. Refusals are evidence, not embarrassments.
- **Scoping.** A connector can be restricted to specific customer IDs, so
  an agent working one client physically cannot touch another.
- **Kill switch.** Per connector and per organisation — access dies in one
  action without touching your Google password.
- **Revocation at the source.** The Google grant itself can be withdrawn
  from your own Google account at any time; the connector holds one
  encrypted refresh token, never your password.

Note the shape: none of these controls depends on another working. An
approval mistake is caught by the audit trail; a leaked address is ended by
the kill switch; and underneath all of it, deletion stays impossible, and the one final write, a lead rating, is proposed first like any other, with your AI client asking before it runs if it is set to ask before AdCopilot's write tools.

For teams the same stack scales sideways: each member connects under
their own Google sign-in, and every Google Ads change they make runs on
that login; a teammate's login adds reach for reads only, and only when the
owner shares it with them and that teammate has allowed it; and every seat
is individually revocable. The no-delete line does not get negotiated per
seat — it is the floor under everyone, which is what makes delegating
access to a five-person team the same decision as delegating it to one.

## What this means for piloting AI on a live account

The recoverable-versus-permanent line changes what a sensible pilot looks like. You do not
need a sandbox account, because the blast radius is already bounded: reads are free, negatives and paused drafts cannot spend, AdCopilot proposes each spend change first ([set your AI client to ask for a hard stop](https://adcopilot.cloud/tutorials/approval-workflows-for-ai-ads-changes)), and deletion is off the table by construction. A live-account pilot in three
moves:

1. **Read week.** Audits only — wasted spend, conflicts, disapprovals. You
   learn how the agent reasons before it touches anything.
2. **Inert writes.** Negatives, paused drafts, assets. Approvals are quick
   because nothing here can spend.
3. **Governed spend changes.** Budgets and statuses with evidence attached,
   one click each, every action logged.

Somewhere in that first week, run the deletion test. The refusal you get
back is the whole argument of this page, executed: write access wide
enough to be useful, and a hard wall exactly where the way back stops existing.
That is what a safety property looks like when it is real — silent,
boring, and impossible to talk the system out of.
[Start free](https://app.adcopilot.cloud/signup) — 5 days on the full Pro plan for your first workspace, no card to start (one is needed after 1,000 operations) — and try it on your own account.

## Questions

### Can a paused campaign be brought back?

Fully. A paused campaign keeps its structure, keywords, ads, history and learning; enabling it resumes where it left off, minus whatever ramp-up the bid strategy needs. That is what makes pause the correct maximum penalty for an AI agent: it stops spend immediately while destroying nothing. Deletion is the opposite — in the Google Ads API, setting status to REMOVED is permanent, which is why that one status is refused outright.

### What if I want the AI to clean up old campaigns?

It can do the whole job except the last click. Ask it to find campaigns with no spend or no conversions over your window, show the evidence, and pause them with your approval — a paused campaign costs nothing and clutters little. If you later want them gone for tidiness, that final removal is a deliberate human act in the Google Ads interface, exactly as rare and considered as it should be.

### Doesn't an approval prompt already make delete access safe?

An approval prompt is a strong control with one weakness: it relies on the approver reading carefully, every time, forever. Approvals catch bad judgement; they do not survive a rushed click on an irreversible action. Removing the capability does. That is why the delete tools are not gated behind approval — they are absent from the tool list, and the server refuses REMOVED mutations regardless of what any client or model asks.

---

This is the markdown copy of https://adcopilot.cloud/blog/the-case-for-ai-that-cannot-delete. The index of the site's pages is https://adcopilot.cloud/llms.txt.
