The daily routine
The six-check morning audit reads only, then proposes fixes. For a hard stop, set your AI client to ask first (Needs approval in Claude, Allowlist in Cursor).
The daily routine is a fixed, read-only audit followed by proposals. The assistant inspects and reports, then proposes, and AdCopilot can never delete. For a hard stop, set your AI client to ask before AdCopilot's write tools: Needs approval in Claude; in Cursor, set the run mode to Allowlist and keep them off the allowlist (Cursor's cloud agents never ask).
Ask for it with:
Run my AdCopilot morning audit.
That runs AdCopilot's morning-audit prompt, which every connection has with
nothing to install. Two more prompts come with it: growth-audit covers Google
Ads plus whichever of Search Console, Analytics and Tag Manager you connected,
and weekly-report is the Monday summary. Where the account's own analysis
tools cover a check (wasted spend, negatives, pacing), the assistant calls
full_audit and quotes its numbers instead of doing the arithmetic itself.
Before you start
- If the assistant does not know an account's monthly plan cap (budget target), it asks once. Without it, spend is reported as a raw number and the cap check is marked "no cap set."
- It identifies accounts with
list_accessible_customers, and consultsget_resource_metadatarather than guessing a field name.
The six checks
Run in order, per connected account. Checks 2–6 use the
search tool and are read-only. cost_micros ÷ 1,000,000 = amount in the account currency. Date
ranges are never written into the query: pass start_date and end_date as
YYYY-MM-DD and the connector adds the quoted segments.date filter itself.
-
Accounts in scope —
list_accessible_customers. -
Spend vs plan cap — month-to-date cost against the cap.
resource: campaign,fields: campaign.name, metrics.cost_micros, withstart_date= the 1st of the month andend_date= today, which reaches Google as:SELECT campaign.name, metrics.cost_micros FROM campaign WHERE segments.date BETWEEN '2026-09-01' AND '2026-09-16' -
Wasted-spend ranking — highest-cost search terms with little to show, flagging near-zero conversions.
resource: search_term_view,fields: search_term_view.search_term, metrics.cost_micros, metrics.conversions, metrics.clicks,orderings: metrics.cost_micros DESC,limit: 50, and the last 30 days as two dates:SELECT search_term_view.search_term, metrics.cost_micros, metrics.conversions, metrics.clicks FROM search_term_view WHERE segments.date BETWEEN '2026-08-17' AND '2026-09-15' ORDER BY metrics.cost_micros DESC LIMIT 50 -
Negative-keyword candidates — the clearly off-intent terms from that ranking worth blocking.
-
Disapprovals — ads switched off by policy. No date range on this one:
SELECT ad_group_ad.ad.id, ad_group_ad.policy_summary.approval_status, ad_group_ad.policy_summary.policy_topic_entries FROM ad_group_ad WHERE ad_group_ad.policy_summary.approval_status != 'APPROVED' -
Pacing — project month-end spend from MTD cost and the fraction of the month elapsed, compared to the cap.
The report
Each account ends with a compact table and a plain verdict per row — OK, WATCH, or ACT:
Account: <name> — <id> (<currency>) Window: <start> to <end>
Check Result Verdict
Spend vs plan cap $X MTD of $Y cap OK
Wasted spend $X on N zero-conversion terms ACT
Negative candidates N off-intent terms ($X) ACT
Disapprovals N ad(s) disapproved ACT
Pacing projecting $X — Z% of cap WATCHTurning ACT into proposals
For each ACT row the assistant writes a specific, approvable proposal and shows it to you before making it:
- Wasted spend → pause the offending keyword
(
update_keyword→ PAUSED) or lower its bid. Never removed. - Negative candidates →
add_negative_keywordswith the exact terms and level. To reverse one, the assistant proposes taking it back out (remove_negative_keywords, or its shared-list or account-level sibling), and your AI client asks before that runs too if it is set to ask before AdCopilot's write tools. - Disapprovals → surface the ad and its policy reason; the fix is usually a corrected ad you review.
- Pacing over cap → reversible budget or bid changes; if a campaign must stop, pause, never delete.
Your AI client asks before each change runs if it is set to ask before AdCopilot's write tools (above). See Approval-gated writes.
After an approved change, the assistant reports how to undo it — almost always the opposite call.
Make it a habit: the daily check-in and the Monday summary
Two artefacts, one habit. The daily check-in is this audit, run every morning. The Monday summary is the same read over the past week. On every plan, including Free, AdCopilot already emails you the daily check-in itself (Recipe C below), and the Monday summary too. To have them inside your own assistant as well, pick the recipe for the client you actually use — the schedule then lives in your client.
Recipe A — Claude (a scheduled task, or a Claude Code routine)
Your Claude runs AdCopilot's read-only morning check by itself, every weekday at 8:07 (a few minutes past the hour, because runs set exactly on the hour can start late). The Check-ins page of your AdCopilot dashboard has the same text filled in with your account and time zone; the version below works as it is.
- Let it run without asking. In Claude, open Customize → Connectors → AdCopilot → Tool permissions. Set Read-only tools to Always allow, because a scheduled run cannot answer a prompt. Under Write/delete tools, set AdCopilot: workspace context (records visit) and AdCopilot: save business notes and progress to Always allow too: they keep AdCopilot's own notes and never change Google Ads. Keep the other Write/delete tools on Needs approval: AdCopilot proposes each change first, and that setting makes Claude ask before it runs.
- Save the task. Send this request in a new chat in Claude (claude.ai, the desktop app or Cowork), and confirm the task Claude sets up. Scheduled tasks come with a paid Claude plan.
Schedule a task for every weekday at 8:07 AM my time with these instructions: AdCopilot morning check, read-only. Call get_org_context first, with source "claude-scheduled" and task "AdCopilot morning check". Do not use any tool that changes my account, and do not ask me questions; if something is unknown, say so and carry on. For the Google Ads account I have switched on in AdCopilot:
1. Yesterday's spend, clicks, conversions and cost per conversion against my 7-day average.
2. This month's budget pacing: on track, over or under, and any campaign limited by budget.
3. The 5 search terms from the last 7 days that cost the most with no conversions, with their cost.
4. Any ad that is disapproved or limited, with Google's stated reason.
5. Any change made in the account in the last 24 hours, and by whom.
Start with one line: "All clear" or "Needs a look". End with the single most valuable fix, written as a yes/no question I can answer when I open this.- Run it once now and check it finishes without stopping to ask.
In Claude Code it is a routine. Add AdCopilot at claude.ai/customize/connectors first: a routine runs in the cloud and cannot see the plugin's own connector. Routines run tools without asking, so keep the instructions read-only, exactly as written. Then type:
/schedule weekdays at 8:07am AdCopilot morning check, read-only. Call get_org_context first, with source "claude-scheduled" and task "AdCopilot morning check". Do not use any tool that changes my account, and do not ask me questions; if something is unknown, say so and carry on. For the Google Ads account I have switched on in AdCopilot:
1. Yesterday's spend, clicks, conversions and cost per conversion against my 7-day average.
2. This month's budget pacing: on track, over or under, and any campaign limited by budget.
3. The 5 search terms from the last 7 days that cost the most with no conversions, with their cost.
4. Any ad that is disapproved or limited, with Google's stated reason.
5. Any change made in the account in the last 24 hours, and by whom.
Start with one line: "All clear" or "Needs a look". End with the single most valuable fix, written as a yes/no question I can answer when I open this.Its first step tells AdCopilot a scheduled run is working, so the assistant stops offering to set one up. It also stops AdCopilot making any change while the run is on (up to 20 minutes): to apply the fix the run suggests, answer its yes/no question in a new chat. For the Monday summary, add a second task for every Monday at 8:07 AM with:
AdCopilot weekly summary, read-only. Call get_org_context first, with source "claude-scheduled". For last Monday to Sunday: spend, conversions and cost per conversion against the week before, the three biggest changes, and three things to do this week.
In a live chat the same summary is AdCopilot's weekly-report prompt.
Recipe B — ChatGPT (Tasks)
ChatGPT does not list connector prompts, so the task is written in plain English instead:
Every day at 8am, run my AdCopilot growth audit and message me the summary.
Every Monday at 8am, send me my AdCopilot weekly report.
Check that the task lists the AdCopilot connector, then trigger it once by hand. Honest caveat: if ChatGPT lets you — we have not yet verified that a scheduled ChatGPT Task can call a connector's tools, so if the first run comes back empty, use Recipe A instead.
Recipe C — AdCopilot running it for you (every plan)
On every plan, including Free, AdCopilot runs the daily check-in itself, server-side, and emails it to the workspace owner every day. It is on by default: AdCopilot switches it on for a workspace with Google Ads connected. Each run is a short, read-only audit of about 14 Google Ads reads, none of them counted against your plan's daily operations. The email shows yesterday's spend, clicks, impressions and conversions against the 7 days before, the keywords that spent the most with no conversions and got the most clicks, impressions and conversions over the last 30 days, and what needs your attention, or says the checks came back clean. What is new or changed since the last check-in comes first. Once an assistant is connected, a short task for the day follows it. Conversion-tracking checks that have not changed since the last check-in are listed together under "Still open from before". Checks the last check-in flagged that now pass are listed under "Fixed since your last check-in". Setup findings that have stayed the same for a week (conversion tracking, disapproved ads, account setup) leave the daily email and are listed once a week, in the weekly summary, on the day it is set to (Monday's check-in when the summary is off, or the next check-in when no email went that day), under "Still open from your setup"; one that changes comes back the next morning, and so does one after a few days without a check-in email. A keyword you have since paused says so, and a search already blocked as a negative keyword is left out. If you added AdCopilot to the Claude app (claude.ai or Claude desktop) yourself, each finding and the task open Claude with the question already typed in; with ChatGPT, Claude Code or a teammate's Claude, the task is a question to ask your own assistant. If you use the Claude app or ChatGPT, a day with a finding worth fixing opens the email with it: "Fix it in Claude" opens Claude with a request to read the account first, propose the fix and change nothing until you say yes; when the check already has the change ready, the request asks Claude to propose that change. With ChatGPT, "Open ChatGPT" opens it, and the email prints that request for you to copy into the chat. It changes nothing. How your data moves says what it reads and keeps.
Pause it on the Check-ins page of your AdCopilot dashboard, or with the unsubscribe link in the email; a paused check-in is never switched back on for you; Resume on the same page switches it back on. The Monday summary is switched on for you beside it, every Monday at the same hour, and that day it comes in place of the check-in, so you get one email; pause it on the same page. Neither can be switched on from inside an AI client.
With more than one Google Ads account switched on, the check-in and the Monday summary read one of them. Choose it under Check-in account on the Check-ins page, or ask your assistant to set it. Until you do, AdCopilot picks one for you: of the accounts whose spend it has measured, the one that spent the most, otherwise the one it read before or the one your assistant has worked in most. If Google says the account it reads isn't active, the check-in reads the next one in that order instead, and that email says so in one line, with the link to choose another. An account you chose yourself is never changed for that. A manager (MCC) account can't be chosen, because it has no campaigns of its own. If an account you chose turns out to be one, or is switched off (for example when a plan change leaves fewer accounts on), the check-in reads another and the email says so in the same way.
It is on every plan, including Free, and so are Recipes A and B.
Accounts and scope
Which Google Ads accounts an assistant can act on — what your Google sign-in can open, the accounts you switch on, and the number your plan allows.
Negative-keyword sweep
Find the search terms that spend with nothing to show and block the clearly off-intent ones with negative keywords. Every negative is proposed first.