A daily Google Ads audit, in your chat, every morning

How the daily routine works

Three steps. About ten minutes the first time to connect and pick a path, a minute or two to schedule it in your own AI client, and nothing to do most mornings after that except read.

1. Connect Google Ads. Your own Google sign-in — connect at app.adcopilot.cloud. It reaches only the accounts you grant, and it can never delete. Already connected? Skip to the next step.
2. Get the playbook. Pick the built-in morning audit prompt from your client's prompt picker — nothing to install, it is part of every AdCopilot connector and knows the full audit sequence and the safe-change etiquette below. (On Claude Code or claude.ai you can load the adcopilot-daily skill instead — covered just below.)
3. Schedule it, in your client. Only your client can create this recipe's schedule. Paste the read-only task text from the daily routine guide into your client's own scheduler — the skill points you to both. Also covered below.

The zero-install way: the built-in prompt

Every AdCopilot connector already exposes the whole checklist as a ready-made prompt — no skill, no plugin, nothing to install. Select it from your client's prompt picker (Claude, ChatGPT, Copilot, Gemini CLI, or any other MCP-capable client) and it runs the same six checks. This is the path we recommend for almost everyone.

Prefer a skill? Load it yourself

On Claude Code or claude.ai you can load the adcopilot-daily skill, which packages the same playbook plus the safe-change etiquette. It is not published anywhere public — email [email protected] for the file, then on claude.ai use Customize → Skills → Create skill and paste it in. Invoke it by name, or just ask for “my AdCopilot morning audit.” It does the same thing as the prompt above.

Setting up the schedule

Here is the honest part: the skill cannot create the recipe's schedule itself — only your own AI client can. (AdCopilot's own daily check-in email is separate: it reads your account once a day, changes nothing, and you can pause it on the Routine card.) What it does is point you to your client's scheduler and to the task text in the daily routine guide, which you paste as it is. That text is read-only — it says “This is read-only — don't make any changes” in plain words — so a scheduled run reports and lists the fixes, and you approve a change in a chat, never on the timer:

• Claude Code: open the scheduler in any session, choose Daily at whatever time suits you, and make sure the AdCopilot connector is enabled for the routine.
• Claude Desktop: sidebar → Routines → New routine, same daily time, same pasted text, permission mode left on approval — not auto-approve.

Run the audit once by hand first, so you see the report format and can tell it your monthly plan cap. From the next morning on, your client's scheduler runs the audit under your sign-in and queues proposals; you just read and reply.

What the morning audit covers

Six checks, in this order, for every account you have connected:

That is the same sequence a one-off AI audit runs — the only difference here is that your client runs it on the schedule you set, instead of waiting for you to ask.

What lands in your chat

Every run ends with one compact table per account and a plain verdict — OK (nothing to do), WATCH (noted, no change yet) or ACT (a proposal follows):

Account: Example Co. — 1234567890 (USD)

Check                 Result                            Verdict
Spend vs plan cap     $2,140 MTD of $2,500 cap           OK
Wasted spend          $312 on 6 zero-conversion terms    ACT
Negative candidates   4 off-intent terms ($188)          ACT
Disapprovals          1 ad disapproved                   ACT
Pacing                projecting $3,050 — 122% of cap    WATCH

Illustrative example — not data from any account, and not a promise about what your own report will find. Each ACT line becomes a specific, approvable proposal underneath the table: the account, the campaign or keyword, the exact change, and what it is expected to do. Nothing above has touched the account yet.

One rule you keep, one it cannot break

Whether the audit runs by hand or on a schedule, the same two rules apply — a routine that could quietly get around them on a timer would be worse than not having one at all. The first holds only as long as your settings keep it; the second holds whatever any client allows:

1. Your AI client asks first, on Needs approval. The scheduled run performs the audit and queues proposals — its task text is read-only. Each change is proposed in a chat, and your AI client asks before it runs: keep write tools on Needs approval, and the routine's permission mode on approval, not auto-approve, because a tool set to always allow runs without asking. AdCopilot has no “while you sleep” mode of its own.
2. It can never delete. The connector cannot remove anything — the removal tools are not in it at all, and the server refuses any call that would set something to removed. The safe equivalent it reaches for is pause, which a click puts back. A genuine deletion stays a Google Ads interface action, on you, by design.

Every applied change comes back with how to put it back — almost always the exact opposite call, like re-enabling a paused campaign. The honest exceptions are additions it cannot take back, like a negative keyword, a location or an ad schedule window: the connector can add one, not remove it, so putting that back means the Google Ads interface, and it says so rather than pretend otherwise.