Keyword Planner is very good at telling you what people type and roughly what it costs. It is no help at all with the question that actually matters, which is whether the person typing it is ready to buy something.
That is the gap. A list of two thousand keywords sorted by volume is not research. It is a spreadsheet you will never act on, and if you upload it to a campaign you will spend most of your budget on people who were never going to buy.
Start with intent, not volume
Every search sits in one of four buckets, and only one of them is worth your money.
- Ready to buy. Someone searching "emergency plumber manchester" has a problem right now. Expensive, and worth it.
- Researching. "how much does a new boiler cost" is a real person who may buy in three months. Cheaper, slower, sometimes worth it.
- Wants to do it themselves. "how to bleed a radiator" will never hire you. This is where most wasted budget goes.
- Looking for a job. "plumber jobs manchester" costs you the same per click as a customer and is worth nothing.
Sorting a keyword list by intent is tedious, judgement heavy work, which is exactly what AI is good at. Here is the prompt we use.
I sell [WHAT YOU SELL] to [WHO BUYS IT] in [WHERE]. Find search keywords with real commercial intent, not research intent. For each one tell me roughly what it should cost and whether someone searching it is ready to buy or just looking. Exclude anything aimed at people wanting to do it themselves or looking for jobs.
The last sentence is the one that earns its keep. Without it you get the do it yourself searches mixed in with the buying ones, and they look identical in a spreadsheet.
Then turn the list into ad groups
A campaign with one ad group containing ninety keywords will show the same three ads to everyone, and none of those ads will match what anyone typed. Keywords that mean the same thing belong together, and keywords that mean something different belong apart.
Ask the AI to group your surviving keywords by what the searcher actually wants, then name each group in plain English. Five to fifteen keywords in a group is about right. If a group has forty in it, it is two groups.
Write the negatives before you spend anything
This is the step that saves the most money and the one almost everyone does after the fact, once the search terms report has already shown them where the budget went.
You can predict most of it. Free, cheap, DIY, jobs, salary, courses, wholesale, second hand, the names of competitors you do not want to bid on, and any meaning of your product that is not your product. Ask the AI what someone might search that contains your keywords but has nothing to do with buying from you, and you will get most of the list before it costs you anything.
What this does not replace
Once the campaign is running, the search terms report tells you what people really typed, which is always stranger than anything you predicted. Research gets you a sensible starting point. The first fortnight of real data is what makes it good.