An AI-powered Google Ads search campaign for ecommerce: setup in four steps

Telling the assistant “build me a campaign” is like asking a specialist for a plan without showing them the account. The order that works is clear: audit the account and conversion health first, then run keyword research with real volumes, make the critical decisions yourself, and only then have the campaign built.

We walk through the flow using an example home textiles store selling bedding and towels; every step runs in a single chat in Claude or ChatGPT through the Tecrube connector.

The first command should be “audit my account; rank the conversion tracking, bidding strategy and structure issues by priority”. The most common ecommerce finding is a purchase conversion that exists but carries no value; without value there is no ROAS target to set.

If conversions are broken, fix them first. Because Tecrube reads GA4 and Tag Manager through the same connector, “does the purchase event exist in GA4 and does it reach Google Ads” is answered in the same chat.

“Pull search volume, competition and bid range for duvet sets, cotton towels and bedspreads; group by intent” brings Keyword Planner data back as a draft of ad groups rather than a raw list.

The point to watch is keeping brand terms in their own group. If “your brand + duvet set” sits with “cheap duvet set”, both the bid and the ad copy go to the wrong audience.

Start the daily budget cautiously; write the figure you can afford to lose for a month, not the figure the assistant suggests. Geographic targeting should cover everywhere you ship, but regions with long delivery times belong in a separate campaign.

A target ROAS based on sale value only makes sense when conversion value flows correctly; otherwise maximise conversions is safer for the first weeks. Write these three decisions explicitly in chat and the assistant prepares the build around them.

The build command can read: “create a search campaign with duvet, towel and bedspread ad groups; use phrase match, add at least two responsive search ads per group plus sitelink, callout and price extensions; add free, second-hand and wholesale as negatives”.

Tecrube gathers the request into a single preview: campaign settings, ad groups, keywords, RSA headlines and descriptions, extensions and the negative list. You read and approve; the campaign is created paused and you enable it when ready.

Preview → approve → apply holds for campaign builds too. Single-step budget increases above fifty percent are rejected and every change is written to the audit log. The 75 free credits comfortably cover one audit, one research pass and one campaign build.

Seven rules: audit first; no ROAS target without conversion value; brand on its own; a negative list from day one; at least two RSAs; never leave extensions empty; watch only for the first two weeks, then clean search terms from week three.

If you want the four steps in a single opening message, this template works: “I run an ecommerce site selling home textiles. First audit my account and tell me whether conversion value flows correctly. Then research keywords for duvet sets, towels and bedspreads with volume, competition and bid range, grouped by intent. Before building the campaign, give me three options for budget, geography and bidding strategy; I will decide.”

The strength of the message is its last sentence. The assistant does the research but leaves the critical decisions to you, so by the time you reach the build preview you already know why each choice was made. For the first two weeks after launch, only watch; from week three start cleaning up in the same chat with “list the search terms that clicked without buying”.

You can reuse the same template for other product groups; changing the product names and the target audience is enough.