Why keyword theory in Ads breaks down a month after launch
When you set up a Google Ads account for the first time, the logic seems simple: choose the words that describe your product, bid on them, and wait. By the third month, the search terms report shows you that you spent money on queries you would never have imagined. That gap between what you think your customers are searching for and what they actually type into Google is at the heart of everything. This article doesn’t explain what a keyword is or how match types work; you’ll find that in any manual. What follows are specific situations, decisions that went wrong, adjustments that worked, and the real limits of what can be controlled with keywords in a campaign.
The case of the spare parts ecommerce store: when exact match saves you
A home appliance spare parts project had a broad match campaign on “washing machine filter.” The average CPC was around €0.40 and conversions were coming in, but the cost per acquisition was eating into the margin. When reviewing the terms report, searches for “how to clean the washing machine filter,” “washing machine filter not working,” and “washing machine filter second-hand price” appeared. None of those queries corresponded to someone willing to buy a new spare part.
The change wasn’t adding negatives willy-nilly. It was splitting the structure into two branches: one with exact match for commercial variants (“washing machine filter bosch,” “washing machine filter balay”), another with phrase match for specific models, and an expanded negative list with instructional verbs (“how,” “clean,” “repair”). The CPA dropped within a few weeks without touching bids. The practical lesson: for a physical product with clear part references, exact match is still the most profitable tool, even though Google’s interface pushes you toward “phrase” and “broad.”
Local services: the keyword that wasn’t geographic
A dental clinic wanted to rank for “cheap dentist” and “dental implants price.” Both converted poorly and brought in curious clicks without appointments. When analyzing the actual terms, they discovered that patients who booked appointments were searching for “dental emergency + city” or “dentist open today.” The keyword with the highest intent wasn’t the most searched one, but the one that expressed an immediate need.
They restructured: an emergency campaign with phrase match keywords that included the neighborhood name, location extension to a tight radius, and an ad with the phone number as the main extension. The call conversion rate went up because the message matched the urgency. What didn’t work was trying to capture “cheap implants”: the volume was high, yes, but the intent was to compare prices, not to book. In healthcare services, the price keyword almost always attracts the curious person who doesn’t want to pay for a consultation.
Industrial B2B: keywords that only exist in the supplier’s catalog
Here comes the mistake that only experience reveals. An industrial machinery company was bidding on the exact technical name of its equipment, as it appears in its PDF catalog. The problem: nobody searches for “7.5 kW oil-injected rotary screw compressor” on Google. Buyers were searching for “7 kw screw compressor price” or “compressor for auto repair shop.” The account was spending on terms that only an engineer from the company itself would type.
The solution involved investing in keyword research with search tools but, above all, talking to the sales team to gather how customers asked about products over the phone. Those colloquial phrases became ad groups. In B2B, the customer’s vocabulary rarely matches the manufacturer’s vocabulary. And no tool can replace that conversation.
The mistake of chasing volume: the generic keyword trap
Almost every account I’ve reviewed has a campaign dedicated to the most generic word in the industry. The logic seems obvious: it’s the most searched, so it should bring more customers. In practice, that keyword is usually the most expensive and the least profitable, because it competes with everyone, attracts informational clicks, and usually ends up in broad match with a pile of reactive negatives.
There’s a legitimate exception: when the business is the brand leader in the industry or when the goal is awareness rather than direct conversion. If the account measures return, the generic keyword rarely pays the rent. It’s more honest to say that than to promise that a good negative list will fix it.
How to know if a keyword deserves its own budget
Before creating an ad group, I cross-reference three data points: estimated volume, stated intent (commercial, informational, transactional), and the margin of the product or service. If the intent is informational, I don’t create an ad group in Search; I send it to organic content or a remarketing campaign. If the margin is low, the keyword needs a very contained CPC to avoid eating the profit. This filter eliminates half the keywords that the Keyword Planner tool suggests.
Negative keywords: the work nobody wants to do
The negative list is where you really make money in Ads, but it’s continuous and boring work. Reviewing the terms report every week, adding phrase negatives when a pattern repeats, and creating shared lists across campaigns. A specific case: a furniture store discovered that 30% of spend was going to searches for “second-hand furniture” and “cheap ikea furniture.” They weren’t direct competitors, but the user was comparing and not buying. Adding negatives for “second-hand,” “ikea,” “used,” and “pre-owned” freed up budget for new product keywords that actually converted.
Watch out for excess: over-negativizing can cut off valid searches. You need to review negatives periodically and check that they aren’t blocking terms with previous conversions. No automation can replace that periodic review.
Brand keywords: the territory where almost everyone fails
Bidding on your own brand name is cheap and usually converts well, but it has a nuance that isn’t always explained. If you already rank organically in first position for your brand, many paid clicks will be cannibalization: you’re paying for a click you would have received for free. However, there are cases where it does make sense: protecting yourself from competitors bidding on your brand, controlling the ad message during launch campaigns, or when organic competition is dominated by comparison sites.
What doesn’t work is bidding on brand without measuring the actual increment. A clean test requires pausing the brand campaign for a period and comparing total conversions (paid plus organic). If the total barely drops, you were paying for traffic you already had.
What keyword tools don’t tell you
Keyword research platforms offer volume estimates that, according to Google’s official documentation, are rounded ranges and not exact figures. That means a keyword with “1,000 searches” could have 400 or 2,500. Making budget decisions with that data without cross-referencing against the account’s actual terms report is a common mistake. The most reliable source of keywords for an account that’s been active for a while is its own search terms history, not the Keyword Planner.
There are also limits to keyword matching: Google doesn’t just match by exact words, but by meaning, according to its documentation on match types. That makes broad match increasingly less predictable and means negatives have to be more surgical. If someone promises absolute control over searches with a keyword list, they’re oversimplifying.
A routine that actually works to keep the keyword list alive
A monthly review with four steps prevents most surprises: read the terms report and note unwanted search patterns, review which keywords have increased in CPC without improving conversion, check those that haven’t received impressions in two weeks (possibly bids too low or low quality), and adjust cross-campaign negatives. It’s not glamorous, but it’s what keeps cost per acquisition stable.
And an honest warning: in industries with strong seasonality or with competition that changes bids daily, no keyword list is stable forever. The advantage isn’t in getting the perfect list right, but in iterating with your own data. Anyone selling a closed formula for choosing keywords in Ads is selling smoke.