Grouping Keyword Planner data by search intent with AI

Google Ads Keyword Planner turns one seed term into hundreds of suggestions, each with monthly search volume, competition level and a top-of-page bid range. That data is the only reliable basis for building a campaign budget on measurement rather than guesswork.

The trouble is that the raw list is long and unsorted. Seed it with “dental clinic” and people asking for prices, people looking for the nearest clinic, people typing treatment names and people browsing job ads all sit side by side. Dropping the list straight into a campaign crams different intents into the same ad group.

In Tecrube the assistant calls the planner directly: “pull 50 keyword ideas for 'dental implant' with volume and bid range”. The data comes from Google; the model only reads and organises it. Then the second request: “group these by search intent”.

The output is usually four or five clusters: price intent (“implant prices”), location intent (“implant in Kadıköy”), informational intent (“does an implant hurt”), comparison intent (“implant or bridge”) and brand intent. Average volume and bid range per cluster appear in the table; you decide which clusters enter the first campaign. Informational intent usually ends up in the negative list.

Clusters do not all need the same match type. Conversion-near clusters such as price and location keep control with phrase and exact match; low-volume, clear-intent terms use budget efficiently in exact match. Broad match is worth considering only with smart bidding and enough conversion data, not at launch.

Ask the assistant to “suggest a match type per cluster and explain why”. The reasoning leaves the decision with you, and because the estimated cost per keyword is in the table, a choice like “start this cluster on exact match” takes seconds.

Read the match type decision together with the budget. On a campaign opened at $30 a day, broad match can hand half the budget to irrelevant searches on day one; the same budget in exact match brings fewer clicks but a higher conversion rate. Widening match types as budget and conversion data grow is always safer than the reverse.

Once clusters are set, a third request produces the campaign structure: each cluster becomes an ad group, each group gets its keywords in the chosen match type, two responsive search ads, and a negative list compiled from the informational cluster. Tecrube shows the whole structure in one preview; the campaign is created paused and you approve and enable it.

Three requests and one approval cover the whole path from research to launch. The same flow works for Microsoft Ads: the planner data comes from Google, but the clustering and structure apply to the Microsoft campaign. The 75 free credits are enough to run this end to end for one product category.

Alongside monthly volume, the planner gives the curve of the last twelve months. That curve sets the budget calendar: if “air conditioner service” peaks in May, the campaign should launch in April, gather data and get its budget raised in May. In chat, “show the highest and lowest month for each cluster” produces that plan in minutes.

Competition level and bid range should be read together. Low-competition, high-bid terms are usually high-value niches with few rivals; high-competition, high-bid terms are where marketplaces and big brands fight. A small budget should stay out of the second group.

The model doing this reading stays faithful to the data: because every figure in Tecrube's tool output comes from the planner, no number is marked “estimated”. The only things the model adds are commentary and grouping; you see both in the preview and change what you disagree with.

First, treating volume as the only criterion: “dental” has millions of searches but no clear intent, while the low-volume “dental implant price Kadıköy” is far more valuable. Second, reading the planner's volume ranges as exact numbers; ranges are wide and contain seasonality. Real search term data from the first two weeks always corrects the planner's estimate.