Building a Google Ads search campaign for ecommerce with AI: a step-by-step walkthrough

Telling an AI “build me a campaign” is like asking an expert for a plan without showing them the account. What comes out is a generic template. The right start is asking the assistant to look at the account first, ask questions second and propose last.

The scenario in this article is fictional: a home textiles store building its first search campaign with Tecrube. The steps reflect the real flow; the numbers are illustrative.

First request: “Audit my account; is conversion tracking healthy, is Merchant Center linked, is there anything to learn from past campaigns?” Tecrube lists the conversion actions, checks whether the purchase event sends a value and whether add-to-cart was accidentally marked primary.

In our example the purchase conversion exists but the value is empty. That gets fixed before the campaign; otherwise a target ROAS strategy will never work. This step underpins every decision in the following weeks.

Second request: “Pull search volume and estimated CPC for the duvet set, bedspread and pillow categories; separate branded and non-branded terms.” Tecrube fetches Keyword Planner data, clusters terms by intent and proposes the first negative list in the same table: “sewing”, “pattern download”, “wholesale” and the like.

The lesson: make research a step of its own, separate from the build. Do not move to structure before you have seen and approved the keyword choice in a preview.

Three decisions are never left to the AI: budget, geographic target and conversion goal. Tecrube asks them as questions; in our example 60 dollars a day, nationwide with separate ad groups for the three largest cities, purchase as the goal and maximise conversions to start.

Switching to target ROAS needs at least thirty conversions in the first thirty days; the assistant reminds you of the threshold and offers the switch as a preview when the time comes.

Fourth request: “Build the campaign with these decisions: one ad group per category, phrase and exact match, fifteen headlines and four descriptions per group, sitelink and price extensions.” Tecrube shows the whole structure in one preview: campaign settings, ad groups, keywords, ad copy, extensions and the negative list.

In the preview two headlines are adjusted for the brand, one ad group is dropped, approval is given. The campaign is created paused; pressing the go-live button is the last step and it is always yours.

Once live, one question every morning for the first week: “List yesterday’s irrelevant search terms and propose negatives.” Three minutes of approval a day protects the first month’s budget. Every change is logged and the weekly summary sits in the /app panel.

Seven rules in short: audit first, research second, critical decisions are yours, structure in preview, campaigns are born paused, daily term cleanup, no strategy change before the threshold. The flow can be tried on 75 free credits, and the same steps work from Claude, ChatGPT, Claude Code or Cursor through the MCP connector.

The store’s gain came from sequence more than speed: no campaign before measurement was fixed, no structure before research was approved. AI sped up every step, but keeping the order was in the owner’s hands.

Follow the same order for your own campaign. Tecrube keeps the chat history, so when you repeat the flow for the next category, your earlier decisions are used as context.

Why a Search campaign instead of Shopping? They are not alternatives. Search is a good start while your product feed is not ready or when you want to capture category-level intent. Once the feed is healthy, Shopping is added alongside.

Can I use the suggested headlines unchanged? You can, but you know your brand voice best. Editing at least a few headlines in your own words during the preview helps your ads stand out from competitors.

How many credits does a campaign build use? Credit usage depends on the number of analyses and actions. This flow of audit, keyword research and campaign build usually stays within the 75 free credits; the panel shows estimated usage before every action.

When should I judge the first results? In the first week only clean search terms; leave bids and budgets alone. Waiting at least two weeks and close to thirty conversions before a real evaluation prevents early, wrong decisions.