Google Ads scripts vs. an AI assistant: when each one is right

What is the difference between a Google Ads script and an AI assistant?

A script does the same thing every time it runs, on a schedule, with nobody watching. An AI assistant does something different every time you ask, while you watch. Scripts suit checks you can write down as a rule. An assistant suits questions you could not have written down in advance, and the judgement calls that follow from the answer.

There is a third tool in the same account that most people skip: automated rules. A rule is a script with no code, limited to the conditions and actions Google offers in a form. Most accounts need some of all three, and the expensive mistake is using one where another belongs.

Where do Google Ads scripts win?

Scripts win on anything repetitive, mechanical and urgent. They run inside Google Ads, they cost nothing, and once authorised they run whether you are at your desk or not. Google documents these limits:

• Schedule. Once, hourly, daily, weekly or monthly, at an hour you pick. Google’s own alerting scripts, such as the account anomaly detector and the link checker, are meant to run hourly.
• Run time. 30 minutes per run. A manager account script that fans out with executeInParallel gets 60 minutes and reaches up to 50 client accounts per run.
• Volume. Up to 250 authorised scripts per account, and iterators return at most 50,000 results each.
• Reach. Scripts can read any report the Google Ads query language exposes, write to Google Sheets, fetch URLs and send email.

That makes them the right tool for a link checker that crawls every final URL each hour, a budget pacer that pauses a campaign at 2 a.m. when it has spent its month, or a label job that runs the same way across 40 client accounts every Monday. A person or an assistant doing those jobs by hand would be slower and less reliable.

A typical alerting script is short. This one emails the search terms that spent money in the last 30 days and never converted:

function main() {
  const rows = AdsApp.search(
    'SELECT search_term_view.search_term, metrics.cost_micros ' +
      'FROM search_term_view ' +
      'WHERE segments.date DURING LAST_30_DAYS ' +
      'AND metrics.conversions = 0 AND metrics.cost_micros > 50000000 ' +
      'ORDER BY metrics.cost_micros DESC LIMIT 20'
  )
  const lines = []
  while (rows.hasNext()) {
    const row = rows.next()
    const cost = Number(row.metrics.costMicros) / 1e6
    lines.push(row.searchTermView.searchTerm + ': ' + cost.toFixed(2))
  }
  if (lines.length > 0) {
    MailApp.sendEmail('you@example.com', 'Search terms with spend and no conversions', lines.join('\n'))
  }
}

The catch is the code. The script needs someone who can write and maintain JavaScript, a threshold chosen in advance (here $50, as micros), and a person to read the email and decide what to do. It finds the terms. It does not decide which ones to block.

Where do automated rules fit?

Automated rules are the no-code version, built from a form under Tools → Bulk actions → Rules. A rule checks a condition on a schedule (hourly, daily, weekly, monthly or once) and then pauses, enables, changes a budget or a bid, or emails you. Google allows 100 active rules per user per account.

Use a rule when the condition is a single metric against a fixed number and the action is one of the ones on the list: pause a keyword whose cost per conversion passes $80 with 30 clicks behind it, or raise a budget on weekday mornings. Once the logic needs a loop, a lookup, a second data source or a branch, it has outgrown the form and belongs in a script.

Where does an AI assistant win?

An assistant wins on questions that are new each time. “Why did conversions drop on Thursday?” has no script, because the answer could be a disapproved ad, a tracking tag, a budget cap or a competitor’s bid, and you only know which after looking at all four. An assistant connected to the account pulls each report, compares the periods and tells you which segment moved.

It is also the better tool when:

• The output is a decision, not a list. Reading 23 zero-conversion search terms and saying which are job seekers, which are a competitor’s brand and which simply have not converted yet is judgement. A script can sort them; it cannot read them.
• Nobody on the team writes JavaScript. The question is typed in plain language, and there is nothing to maintain when Google changes a report.
• The work crosses platforms. A script lives in one Google Ads account. A question like “where did we spend the most per lead last month, Google or LinkedIn?” needs both.
• The job happens once. Restructuring an account, drafting ad copy for a new product or auditing an account you just took over are one-off pieces of work. Writing a script for them costs more than the work itself.

Where an assistant loses: it does nothing until you ask, it is no faster than you at a task that needs to happen at 2 a.m., and a change it makes needs a human to check it before it reaches the account. An assistant that can write to an account without that check is one nobody should connect.

How do the three compare?

How do you decide which one a task needs?

Ask three questions about the task, in order:

1. Does it have to happen when nobody is looking? Yes means a schedule: a rule if the condition fits the form, a script if it does not.
2. Could you write the whole procedure down today? Yes means it can be automated. No means it needs someone, or something, to look first.
3. Is the output a decision? Yes means a person signs it off, whichever tool did the finding.

In practice the three stack. A script or a rule watches overnight and raises the flag. An assistant answers why it went up and drafts the fix. A person approves the fix.

How does Adako combine the two?

Adako is an MCP server that connects Claude, ChatGPT and other assistants to Google Ads. It covers the assistant column of the table above and part of the script column.

On the assistant side, Google Ads in Adako answers questions like the ones in this post from live account data. Reads are free on every plan. A change, whether it is a negative keyword, a paused campaign or a new budget, arrives as a proposal with the fields that will change. Nothing reaches Google until you approve it, it executes once, and Adako reads the result back from Google.

On the schedule side, Adako monitors are saved checks such as “CPA above $60” or “spend 30% over its baseline”. They run once a day against a synced copy of the account, email you when they fire, and can propose a pause or a budget cut for you to approve. They never change the account by themselves.

What Adako does not replace: a script that has to run hourly, crawl every landing page or act at a specific minute. Monitors are daily, and their only actions are proposals. If a job needs to change the account at 2 a.m. without a person, keep the script.

The short version

• Scripts: recurring, mechanical, urgent work you can write as code. Free, hourly at best, 30 minutes per run.
• Automated rules: one threshold, one action, no code.
• AI assistants: new questions, judgement, one-off projects and work across platforms.
• Anything that changes money or delivery gets a human approval, whichever tool found it.