Optimize Google Ads with AI: what to ask beyond Smart Bidding

Your Search campaign runs on Target CPA at US$ 60. Last month it delivered at US$ 84, and Google Ads still says the campaign is optimizing. It is. Smart Bidding is pricing every auction as well as it can with the inputs you handed it. Optimizing Google Ads with AI means auditing those inputs. A conversational AI that reads the account through a connector can check what Smart Bidding never questions: search terms that spend and never convert, campaigns capped by budget, keywords with zero conversions in 30 days. It reads, groups and proposes. You decide. Nothing changes until you say yes.

Smart Bidding and an AI assistant do different jobs

Smart Bidding sets a bid per auction using signals no human can weigh in real time. That part is solved, and you should not try to out-bid it by hand. But it works inside a box you built: the conversion action you defined, the keywords and match types you chose, the budget you set. When the box is wrong, Smart Bidding gets steadily better at buying the wrong thing. How to pick the target itself is covered in Target ROAS in Google Ads. This post is about the box.

The middle column does not replace the left one. It watches the box.

Three audits worth asking for, in order

1. Search terms that spend and never convert

Highest return per minute in almost any Search account. Ask with a date range, a threshold and a breakdown:

"Search terms for the campaign Search – Roof Repair over the last 30 days with cost above US$ 25 and zero conversions, most expensive first. Show the keyword that matched each one and group them by intent."

What the AI does well here: reading a few hundred rows and sorting "roofing jobs", "diy roof leak fix" and "roofing supplies" into buckets, then pointing at the keyword that keeps pulling them in. It is usually a broad match. Where it slips: it does not know your vocabulary. A term that looks irrelevant may be the name of a product line the client sells. The full cleanup workflow, with the classic mistakes, is in negative keywords with AI.

2. Campaigns capped by budget

A campaign with a good CPA that burns its full budget every day is leaving demand on the table. Ask:

"By campaign, daily spend and daily budget for the last 14 days. Which ones spent close to budget almost every day and have a CPA below the account average?"

Read the two-week average, not a single day. As of October 2026, Google Ads Help says most campaigns can spend up to two times the average daily budget on a given day, while the month stays within 30.4 times that budget. One day over budget means nothing on its own.

The AI gives you the shortlist. Confirmation lives in Google Ads: the campaign status shows when it is limited by budget, and Search lost impression share due to budget tells you how much demand is left. Where the AI slips: assuming more budget buys more conversions at the same CPA. Sometimes it does. Sometimes CPA climbs with spend, and only a test tells you which.

3. Keywords with zero conversions in 30 days

"Active keywords with spend above US$ 75 and zero conversions in the last 30 days, with match type, max CPC and Quality Score."

The list usually holds three kinds of rows: a broad match pulling in a neighboring market, an expensive generic term that never sold, and a good keyword with too little volume to judge. The first two are pause candidates. The third waits.

One detail the AI does not always mention: on a campaign using Smart Bidding, Google sets the bid for each auction, so lowering a keyword's max CPC does nothing there. The fix is to pause it, tighten the match type, or add negatives for what it attracts. And pause rather than remove, so you can bring it back in one click.

Prompts for the rest of the audit, from structure to ads to landing pages, are in Google Ads prompts.

Where AI gets Google Ads wrong

It reads fast and answers with confidence, including when context is missing. The mistakes that show up most:

• Judging a campaign mid-learning. Changed the target, the strategy or the budget three days ago? Cost will wobble. Asking "what got worse?" in that window produces a suggestion to change something again, and the campaign never settles.
• Calling "no conversions" waste. Early research queries can convert later, inside the conversion window. On a long sales cycle, 30 days without a conversion is not a verdict.
• Adding up conversions that are not worth the same. If the account counts a phone click and a booked job as conversions, the blended CPA hides both. Tell the AI what each conversion action means.
• Recommending more budget without knowing the operation. It does not know whether the sales team can handle 30 more leads a week. You do.
• Answering about data it does not have. Ask "which campaign and which date range is that number from?" whenever something looks too good.

The safeguards are the same as in the rest of the series: questions with a date range, a check against Google Ads in the first week, and no change without your yes.

What the AI can do in Google Ads through Vazante

Connecting the account to Claude or ChatGPT is covered on the Claude connector page.

A 20-minute weekly routine

• Every Monday: last week's search terms above your cost threshold. Negate what is the wrong intent.
• Every other week: budget-capped campaigns and keywords with zero conversions in 30 days.
• One change per campaign at a time. Then ask "what happened to CPA since the change on the 6th?" and check the log for what changed and when.

This is the diagnose-and-execute part of the full method in optimize ad campaigns with AI, applied to Google. Google keeps the bid. You keep the box around it.

To run this audit by talking to your Google Ads account, see the plans: starting at US$ 39 per month, with the first 7 days for US$ 2.99.