How I Manage 43 Google Ads Accounts With Claude Skills

I manage 43 Google Ads accounts single-handedly, under one MCC (Manager Account) structure in Turkey; what makes that possible isn’t logging into each account individually — it’s handing off repetitive tasks to Claude Skills and stepping in only at the moments that actually need a decision. It runs on a single login-customer-id header and a handful of skill files. I haven’t come across another example of a single person managing this many accounts with AI agents at this scale — it usually takes a whole team.

How It’s Set Up: MCC and login-customer-id

The classic way to access 43 separate Google Ads accounts is to log into each one individually, or switch customers between screens. I don’t do that. All the accounts sit under a single MCC (Manager Account, 8601621268); by adding the login-customer-id header with the MCC ID to every query sent to the Google Ads API, I get access — both read and write — to all 43 accounts from a single session.

That’s the foundation Claude Skills sits on. When a skill runs, it first takes which account it’s dealing with (the account ID) as a parameter, then applies the same code path to that account. I don’t write 43 separate scripts for 43 accounts — a single skill takes the account ID as a parameter and runs the same logic on every account. That’s also where the approach differs from classic automation: a script says “do this on that account,” while a skill says “look at performance first, then suggest something if needed” — there’s a reasoning step in between. For a broader framework on this distinction, see the Agentic AI Guide hub page.

Which Tasks Became Skills

Not every task is a candidate for a skill. I turned four recurring tasks into skills, because each one repeated on a weekly or monthly cadence and I was following the same steps every time:

• Weekly performance report — pulls spend, conversion, and CPA for all 43 accounts into a single table and flags accounts showing deviation.
• Budget monitoring — lists campaigns approaching their daily budget cap or running out of budget early in the day.
• Keyword scanning — scans the search terms report and produces negative keyword candidates.
• Bid strategy review — compares Target ROAS/Target CPA performance against the weekly trend and prepares a bid-change suggestion (doesn’t apply it — see the next section).

Each skill consists of a SKILL.md file and whatever scripts it needs, describing which data to pull, how to summarize it, and what format the output should take. I go into the actual file structure and the other skills I’ve built on the Claude Skills Examples I Use as a Consultant page.

Discipline at Scale: What Needs Approval, What Doesn’t

Tracking 43 accounts by hand is already impossible — even checking every account once a week alone would take a full working day just to look. But leaving every account on autopilot is a different risk: budget decisions, bid strategy changes, and campaign pause/start actions affect real money in real time.

So I’ve drawn a clear line. The side that runs without approval: read operations — reporting, deviation detection, keyword scanning, preparing bid suggestions. The side that requires approval: any write operation — budget changes, bid strategy changes, pausing/starting a campaign or ad group. A skill produces a suggestion; I read the current value first, see what will change and what the rollback path is, then apply it. This is the Google Ads equivalent of the “backup first, then approval” discipline I use on the server side.

What Went Wrong in a Skill’s First Version

To be honest: the first version of the bid strategy skill was too aggressive. On a low-conversion-volume account, it looked at a single week of data and suggested raising the bid by over 30% — it had mistaken statistical noise for a real trend. I caught it before applying it, because the approval step was already there; but the skill itself was flawed, not a one-off fluke. The fix came with two constraints: the skill now doesn’t produce a suggestion without at least 3-4 weeks of data, and I capped how large a single bid change it can suggest. The lesson was: leaving a skill alone just because it “works” isn’t enough — you need to audit it a few times, the same way you’d check the first reports of someone new on the job. Now every new skill runs in “suggest only, don’t apply” mode for a few weeks before it gets full authority live.

Why This Works for You Too

Even if you’re thinking “I don’t have 43 accounts, I have 2-3 campaigns,” the idea here still applies — the discipline stays the same even at a smaller scale. For a small business owner, the real problem usually isn’t the number of accounts, it’s time: logging into Google Ads once a week, reading the report, noticing a deviation, asking the right question. The skill logic flips that around — the report waits for you, and you only step in when a real decision is needed.

That’s also exactly where the difference between handing this to an agency versus a consultant lies: an agency usually hands the repetitive work off to a junior; a consultant (me, in this case) hands it off to a skill and keeps control. The result is the same for you — fewer surprises, faster-caught problems — but the process is transparent: which skill checks what, and which decision stays with you or with me, is clear from the start.

You Can Ask Me for This Work

The skill setup described here — MCC access, automating recurring reports, clarifying approval points — can be built for your own accounts too. This is typically 2 hours to 2 weeks of remote, hourly-billed work; the scope depends on your number of accounts and existing Google Ads structure. For details and to book time, check the AI Consulting page; my general approach to managing Google Ads with AI, on the Managing Google Ads With AI page, might also interest you.

Frequently Asked Questions

What is Claude Skills, and what does it have to do with Google Ads management?

Claude Skills are instruction-and-script packages given to Claude (Anthropic’s AI model) so it performs a specific task the same way, to the same standard, every time. On the Google Ads side, this turns something like “prepare this report in this format every week” from a one-off prompt into a persistent, consistent process. See What Is Claude Skills for a full definition.

Does this method work for small businesses too, or only at large scale?

It works — and setup is actually simpler at small scale. For a single-account business, you don’t need the MCC/login-customer-id layer at all; you can connect directly to the account and set up the same skill logic for weekly reports or budget monitoring. The difference is that on my side, 43 accounts are managed by the same skill from a single point; on yours, that’s just 1 account.

Can AI make budget or bid changes in my account — isn’t that risky?

Not in the system I’ve built — and I built it that way deliberately. Skills produce suggestions, they don’t apply them. Budget, bid strategy, and campaign status changes always go through my approval. The automated part is collecting and analyzing data; the risky part — making the decision that affects money — stays with the human.

How many accounts does this method scale to — is there a limit?

The practical limit has less to do with the number of accounts than with how different their structures are. Since most of the 43 accounts share a similar industry/campaign structure, a single skill logic fits all of them. Very different business models coming together (e.g. e-commerce + local service + SaaS) would likely require 2-3 variants instead of one skill — but the 40-50 account range hasn’t forced that split yet.