The AdCopilot blog

A Google Ads AI agent playbook: audit wasted spend, build campaigns and adjust bids in plain language inside Claude or ChatGPT, safely, on your approval.

The waste in most accounts never shows on the Overview page. Ten leaks hiding in reports advertisers rarely open, and one AI session to find all ten.

Forty prompts written for an AI that can read your account — audits, negatives, pacing, RSAs and builds — plus the patterns that make them work.

RSAs killed the clean A/B split. The honest replacement is asset-level analysis plus deliberate variants — and AI does the tedious half of both.

AI is eating the PPC work that required attention but not judgement. The professionals who thrive are redefining the job around what it cannot touch.

The monthly reporting grind is mostly narration — explaining what numbers mean. Narration is exactly what a connected AI drafts well, per client.

AI Max trades keyword control for query expansion. The right call depends on whether your negatives and tracking can police what it brings in.

Auto-apply is Google optimising for Google's definition of optimised. Some toggles are hygiene; several quietly restructure your account.

Q4 prep is a September project, not a November scramble — and every classic checklist item is now a prompt away from done.

Broad match finally works — with Smart Bidding, real conversion data, and someone reading the search terms. Skip the third leg and it reverts to a money pit.

Google spends to daily budgets while you answer to monthly ones. The gap between those clocks is where overspend hides — a pacing prompt closes it.

DSA campaigns auto-upgrade to AI Max on Google's schedule whether you plan or not. Migrating deliberately keeps your URL controls and data intact.

Most accounts run zero experiments because setup and readout are chores. An AI removes the chores and leaves only the decision — the valuable part.

AI Max went GA, ads entered AI Mode, Google shipped its own agent, and every ad platform released an MCP server. What each shift means for operators.

Drops have a diagnostic order: measurement, change history, auction, demand, constraints. Walk the five layers and find the culprit in minutes.

Reporting is a question-answering problem, not a dashboard problem. How a connected AI builds the weekly readout and answers the follow-up.

Scripts answer questions you thought of last month. An MCP agent answers the one you have now. Where each wins, and the hybrid that beats both.

Four ad platforms shipped official MCP servers within months. Google's is read-only — which defines exactly where third parties still matter.

There is no AI Overviews checkbox. Eligibility flows from campaign types and asset quality — an account-hygiene problem you can actually work on.

What ChatGPT can do for Google Ads without account access, and the four connected workflows that turn it from copywriting intern into analyst.

Claude's long context fits an entire account in one conversation. How to connect it, the whole-account questions to ask, and the guardrails.

Approving every AI action sounds safe and scales terribly. The working pattern is tiered: log the trivial, confirm the material, forbid the irreversible.

The morning triage across a client roster — who overspent, who tanked, who got disapproved — collapses into one cross-account prompt.

N-gram analysis used to need a script and a pivot table. A connected AI tokenises, aggregates and — crucially — makes the negative-keyword call.

Quality Score is a diagnostic readout, not a KPI. Map every low-QS keyword to its failing component and the vanity metric becomes a fix list.

Smart Bidding automates the bid, not the judgment around it. Targets, learning resets, constraints and seasonality are still yours — made visible by AI.

Third-party keyword tools estimate demand you might get. Your search terms report records demand you already paid for. Mine it with AI first.

PMax punishes daily fiddling and rewards disciplined weekly review. The four-prompt sequence that covers spend, queries, assets and brand leakage.

Not which MCP servers exist — which five layers cover a marketing team's actual week: ads, analytics, CRM, content, comms, composed in one client.

Google's change history plus a connector-side audit trail gives you two independent ledgers of what the AI did. Reconcile them weekly.

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.

Full autonomy sounds like the goal until an agent spends real money. The approval loop is what makes an autonomous Google Ads agent shippable.

The hard part of RSAs is not fifteen headlines — it is writing them against real search terms and pushing them live without a copy-paste chain.