Broad match wastes money in dermatology because one word covers three different buyers. “Mole,” “Botox” and “acne” are medical, cosmetic and do-it-yourself intent at once. Broad match matches on meaning plus account signals, and Smart Bidding spends whatever it is fed. The fix is a weekly search terms review and negatives built per campaign.
Broad match plus conversion-based Smart Bidding is the setup Google recommends, and for many advertisers it is fine. On a dermatology account with high-ticket procedures sitting next to insurance-billed office visits, it is an expensive default — not because the machine is broken, but because the vocabulary is ambiguous in a way most industries’ is not.
What broad match actually does now
Broad match stopped being keyword expansion years ago. Google now describes it as matching searches related to your keyword, including searches that don’t contain the direct meaning of the keyword. The word you typed need not appear in the search at all.
The signals it uses are documented, if briefly. Google says broad match looks at additional signals in your account — landing pages, other keywords in the ad group, previous searches, and more. That last one matters: eligibility is partly a function of the person searching, not only the words typed. Two people type the same three words and only one triggers your ad.
Your own structure feeds the expansion, too. A derm ad group mixing “skin cancer screening” with “laser resurfacing” tells the system both neighborhoods are yours.
Google is explicit that broad match is meant to run with Smart Bidding — the guide calls it critical to use Smart Bidding with broad match, because each query needs its own auction-time bid. That pairing is where the cost shows up. Smart Bidding optimizes toward the conversion action you gave it, and if every form fill counts as one conversion, it cannot know that a Mohs consult and a request for acne cream pricing are different events. It buys the cheaper one.
This is all moving further in one direction. The campaign-level broad match setting is being folded into AI Max, which layers keywordless matching on top of broad match, and those campaigns upgrade automatically starting September 2026. The controls below matter more under that setup, not less.
Why dermatology collides with itself
Most verticals have one axis of ambiguity: buyer versus researcher. Dermatology has several, layered on the same words.
Medical versus cosmetic
A search for “mole removal” is either someone who found something changing and is frightened, or someone who wants a benign spot off their cheek before a wedding. Different department, different payer, different page on your site. Broad match pulls both into whichever campaign it thinks fits, and Smart Bidding chases whichever converts faster on your form — usually not the one with the higher lifetime value.
One brand, two indications
“Botox” is aesthetic injectables and it is a treatment for chronic migraine, and in some searches hyperhidrosis or TMJ. It is also a price search, a training-course search, and a hunt for the nearest med spa charging by the unit. One broad match keyword sits across all of that.
Treatment-seeking versus product-seeking
“Acne treatment” splits between someone who wants a prescription and someone comparing drugstore products. The second group is enormous, searches constantly, and will never book. They do click, which is enough to keep the expansion alive.
The traffic nobody intends to buy
Underneath all of it sits the usual noise, and derm attracts more than most: symptom identification, “is this cancer” questions, students, training programs, ingredient lookups, job searches. Cosmetic terms attract supplier traffic too — the equipment manufacturers use the same words the clinics do.
The twenty-minute search terms audit
Do this weekly on any account running broad match. In order.
- Open the search terms report for the last 30 days, campaign by campaign. Account-wide, nothing stands out.
- Sort by cost, descending. Not clicks, not impressions. Money is concentrated in a short list of terms.
- Read the top fifty rows. For each, answer one question: if this person walked in the door, would the front desk book them. Not “is it related.” Would they book.
- Mark three categories as you go: wrong service, wrong intent (informational, DIY, job seeker, competitor), and right query in the wrong campaign — a cosmetic search landing in the medical campaign, or the reverse.
- Filter to terms with spend and zero conversions over a window long enough to mean something. On a high-ticket list, that is more than a week.
- Add negatives for the first two categories. Wrong service and wrong intent are true waste and belong in a shared list.
- Do not simply negate category three. That is a routing problem. Negate it in the campaign it landed in and add it as a keyword in the campaign it belongs to, so the traffic moves instead of disappearing.
- Check what the report is not showing you. Google omits search terms without enough query activity, for privacy reasons. That unattributed spend is real. If it is large, the answer is tighter match types, not more negatives.
- Log the date and what you added. Negatives compound, and six months on someone needs to know why a term is blocked.
Negative list architecture
Three places to put negatives, and they do different jobs.
Account-level negatives apply across search and shopping inventory in every campaign, with a documented cap of 1,000 terms. Use it only for what is wrong everywhere: jobs, salary, training, free, DIY, wholesale.
Shared negative keyword lists are the working layer. Google allows up to 20 lists of 5,000 keywords each, applied to campaigns. Build them thematically rather than as one blob: informational, competitor and supplier, DIY and OTC, cosmetic, medical.
Campaign-level negatives do the routing — terms valuable to the practice but wrong for this campaign.
Where people get it wrong
- One master list applied to everything. It works until someone adds “cosmetic” to it and quietly kills the aesthetics campaign.
- Assuming negatives behave like keywords. They do not. Google documents that negative keywords account for casing and misspellings but do not match close variants — you have to add singulars and plurals yourself, and synonyms are not excluded automatically.
- Negatives that are too broad. A negative broad match blocks when all its terms appear in any order, so “removal cream” also blocks “cream for removal of scars.”
- Negating a symptom instead of an intent. Blocking “itchy” removes real patients with the noise.
- Forgetting the 16-word rule. Google notes an ad can still show if the negative keyword falls after the sixteenth word of a long search.
When broad match genuinely works
Advertisers who say never use broad match are wrong too. It earns its place when several things are true at once.
- The conversion action is clean and differentiated. Separate conversion actions with separate values for a cosmetic consult and a medical appointment, so the bidding model can tell them apart. This is the load-bearing condition. Without it nothing else saves you.
- There is enough conversion volume for the bidding strategy to learn anything. On a low-volume, high-ticket list, exact and phrase give more control and cost little.
- Campaigns are already split by intent, with their own landing pages, so the account signals feeding the expansion point somewhere specific.
- Someone reads the search terms report every week and can act on it.
- Brand controls are set where relevant. Search campaigns support brand inclusions and exclusions, a cleaner tool than fifty competitor negatives.
One structural detail that makes mixed match types safer than they look: Google prioritizes an eligible keyword identical to the search term across all match types. Exact match terms keep serving on their own queries; broad match picks up what is left. That is the sane version of this setup — broad match as discovery around a tight core, not as the whole keyword list.
The honest caveat
None of this substitutes for deciding what a patient is worth. An account that counts every form fill as one conversion is asking an algorithm to optimize toward a number the practice does not believe. Negatives clean up the edges; conversion values fix the middle. If there is time for only one, do the values first. Broad match then becomes a question of tolerance rather than a leak.