What you'll learn in this article
- The two numbers I write down before I open any tool, and why they decide everything after
- My collection pass for how to find keywords in Google Ads: four sources, in order, and what each is good for
- The two-question validation test every candidate has to survive before it reaches a campaign
- How I read a SERP as evidence, and the inference that separates a buyer query from a reader query
- Why the launch list is deliberately small, and how the search terms report takes over from week three
Every new campaign I take on starts with the same question, and it's never really about the tool. The question of how to find keywords for Google Ads gets answered in most tutorials with a screenshot of Keyword Planner and a list of buttons, which is a bit like answering "how do I cook" with a picture of a hob. The mechanics take fifteen minutes to learn. The workflow around them, the part that decides whether the campaign makes money, takes longer and almost nobody writes it down.
So this is mine, as I actually run it, in the order I run it: the sequence I've settled into after building launch keyword sets across lead gen and ecommerce, and after watching enough of my own guesses get corrected by real query data to know which guesses are usually wrong.
The through-line is this: finding keywords is a research step, but validating them is an inference step, and the second one is where the money is made. Collecting candidates is cheap and largely automated. Deciding which of them represent a person worth paying for is a judgement nobody can outsource, and it's the only part of the process I never rush.
Before I open the tool: the two numbers that decide everything
I don't open Keyword Planner first. I open a blank document and write down two things: what a conversion is worth to this business, and what the offer genuinely is in the buyer's language rather than the company's. Everything downstream answers to those two lines, and skipping them is why so many keyword lists look defensible in a spreadsheet and fall apart in an auction.
The value number matters because it converts an abstract question into an arithmetic one. If a lead is worth 200 and the sales team closes one in five, a click that costs 8 needs to convert at better than roughly one in five to break even. Suddenly a keyword isn't good or bad, it's affordable or unaffordable at a given conversion rate, and I can look at a CPC estimate and know immediately whether I'm even in the game. Without that anchor I'd be doing what I see in most new accounts: judging keywords on how relevant they feel.
The offer line matters for a subtler reason. Businesses describe themselves in category language, buyers search in problem language, and the gap between the two is where the useful long tail lives. A client selling "workforce management software" has customers typing about shift scheduling and rota chaos. If my seeds come from website copy, the tool returns polished category variants and none of the phrasing people actually use.
Out of that I write ten to fifteen seed phrases in buyer language. That's the whole pre-tool step, usually thirty minutes, and it's the highest-leverage half hour in the workflow. Google's own best practices for finding new keywords makes a similar point from the other direction: seeds that are too broad give the tool nothing to work with, seeds that are too narrow box it in. In practice mid-specificity seeds in customer vocabulary produce the best expansion, and that matches what I see every time.
The collection pass: how to find keywords for Google Ads without judging them yet
Now the tool. My pass is deliberately wide and deliberately unjudged, because mixing gathering with filtering is how bias creeps in early and quietly. I want a big messy pool first, then one clean decision phase. Anyone asking how to find keywords in Google Ads gets pointed at Keyword Planner, and rightly, but it's one of four sources I use and it's not the first.
I start with the search box itself. Autocomplete and related searches are the only free source showing real phrasing at real frequency, unsmoothed and ungrouped, and ten minutes there beats an hour of tool exports for long-tail language. "People also ask" goes in the same pass, read differently: those are research-stage queries, so they're less a keyword source and more a preview of my negative list.
Then Keyword Planner, seeded with the phrases from step zero plus the client's URL and, when it's informative, a competitor's. Its unique value isn't the idea generation, third-party tools do that fine. It's that the volume and CPC figures come from the same auction I'm about to bid into, banded and rounded but native rather than modelled. When I'm weighing a click cost against the conversion value I wrote down, that's the number I want.
The fourth source is the client, and it's the one most people skip. Sales calls, support tickets, the questions that come up before someone buys: uncontaminated buyer vocabulary, which routinely surfaces terms no tool suggested. I ask for it on every account now.
By the end I have two or three hundred candidates in one column. They're unsorted and mostly unusable, and that's fine, because none of them have been judged yet. If you want the wider framing around this stage, I set it out in my walkthrough of keyword research for Google Ads, which this piece narrows down to the finding-and-validating half.
Validation by intent: the two questions every candidate has to survive
This is the part that actually answers how to find keywords for Google Ads in a way that survives contact with a real budget. Collection gives me candidates; validation decides which ones get money. Every term in the pool faces two questions, and both are cheap to ask and expensive to skip.
First: what does the results page tell me about this query? I search it myself and read what Google has decided to serve. If the page is guides, definitions and Wikipedia, Google has concluded the intent is informational, and it's concluded that from far more data than I'll ever have. Fighting that read with a conversion-focused ad is a fight I lose slowly and expensively. If instead the page is full of ads, product listings and comparison pages, the commercial intent is confirmed by the strongest available evidence. The SERP is a free intent report and I treat it as the primary validation signal.
Second: what would this searcher do next on my landing page? I picture the person behind the query arriving at the page I'd send them to. If the honest answer is "read a paragraph and leave", the term fails, however relevant it looks. If it's "check the price, see whether it fits, enquire", it passes. That test is the only one catching terms that are topically perfect and commercially useless, the most common failure mode in every account I inherit.
Terms sort into three buckets. Buy now, and those launch. Maybe later, and those get parked for phase two once I have data. Never, and those aren't waste, they're my starting negative list, built out of the junk the research surfaced. Building it at this stage rather than after launch is what stops looser matching quietly spending on queries I already decided to refuse, and the mechanics of applying it properly are in my rundown of negative keywords in Google Ads.
The bucket sizes tell the story. Two hundred and fifty candidates typically produce twenty to forty launch terms, forty or so parked, and the rest negatives or deletions. New advertisers find that ratio alarming. I find it reassuring: a validation step that rejects almost nothing wasn't a validation step.
Reading the signals: the inferences that do the real work when you find keywords in Google Ads
Everything above is procedure. What makes it work is a handful of inferences I've built up from watching predictions get corrected by data, and they're worth stating explicitly because they're what the procedure is made of.
High CPC is information, not a warning. When Keyword Planner shows a wide, expensive range, that's other advertisers who have run this term and kept paying, which is the closest thing to a profitability signal available before spending anything. Cheap clicks in a competitive vertical usually mean nobody found them worth defending. I'd rather investigate an expensive term than assume a cheap one is a bargain.
Volume estimates are a weak signal at the term level. Planner groups close variants and reports in bands, so a term showing ten a month often sits inside a family worth far more. I use volume to size a theme and ignore it when comparing individual terms.
Query grammar tells you the stage. Questions signal research. Modifiers like price, cost, near me, buy and comparison phrasing signal readiness. Bare product nouns are ambiguous and need the SERP test. That pattern-reading is most of what people mean by "reading intent", and it's the core of how to find keywords in Google Ads that actually convert.
Last one, the most useful and least comfortable: my confidence in a keyword is a hypothesis, and it will be wrong at a predictable rate. The terms I'm most certain about underperform often enough that I now launch with the expectation of being corrected. That expectation is what makes the post-launch loop feel like the process working rather than the plan failing.
From validated list to launched campaign
A validated list still isn't a campaign, because each surviving term carries a second decision: how tightly I want it matched. My rule falls straight out of the validation work. Terms I'm confident about get the tightest matching, since I've already done the thinking and don't need help exploring. Terms where I suspect there's vocabulary I haven't found get looser matching on purpose, as a controlled experiment with the negative list sitting underneath. The setting itself isn't the risk; unearned freedom is.
Grouping follows intent rather than topic. Terms that share a stage of readiness share an ad group, because they can share a message and a landing page. A price query and a "what is" query can sit in the same topic and need completely different ads, and splitting them by intent is what makes the copy write itself. This is the same logic I apply account-wide in my keyword strategy for Google Ads, where intent decides structure and volume decides nothing.
The launch list stays small on purpose. Twenty to forty terms concentrates budget enough that each accumulates real data within weeks. Spread the same spend across three hundred and nothing reaches significance. Discipline here isn't caution, it's what buys readable data later.
Week three onwards: when the account starts finding keywords for you
Everything up to launch is a hypothesis, and I hold it loosely. Within two or three weeks the search terms report has enough volume to start correcting me, and from that point it becomes the best keyword source I own, because it isn't modelled or estimated. It's what actually matched, what actually cost money, and what actually converted.
My weekly pass never stops. I read the report, promote converting queries into their own tightly matched spots, push irrelevant ones into the negative list, and note where my pre-launch reasoning was wrong. That last part compounds: when a term I was confident about underdelivers, the useful question is which part of my inference failed, because that's what would otherwise repeat on the next account.
After a couple of months the launch list is mostly gone, replaced by terms the account discovered. That's the workflow succeeding. Learning how to find keywords for Google Ads isn't a task you finish before launch, it's a loop you maintain, and the pre-launch work exists to make the first weeks cheap enough that the loop gets to run.
So the whole thing, condensed: write down what a conversion is worth, seed from buyer language, collect wide from four sources without judging, validate every candidate against the SERP and the landing page, launch tight, then let the search terms report correct you every week. That's it. The tools change every couple of years and that sequence hasn't needed to.