Intent capture where the demand already exists. Keyword and negative architecture, landing-page compliance review before launch, and conversion tracking that survives consent mode.
Demand that already exists
Meta finds people who were not looking for you. Google serves people who were. For a clinic, a
law practice or an insurer that difference is decisive: somebody typed the query, and the only
question is whether you are permitted to answer it.
Certification is the first constraint, not the final check
Google gates several of the categories we work in behind verification rather than a ban.
Financial services advertisers — lending, trading, insurance, most of crypto — need clearance, and
the requirements differ per market. Gambling advertisers need country-level certification. Health
advertisers face separate restrictions on what may be promoted at all.
None of this is paperwork you attach at the end. It determines which countries are worth building
campaigns for, which is why it gets resolved before keyword research rather than after the first
disapproval. The landing page is reviewed in the same pass, because Google reads the destination and
a page making impermissible claims takes the account with it.
Keyword and negative architecture
Every category attracts adjacent, worthless traffic, and each attracts its own kind. A recruiter
pays for people looking for salary data. A clinic pays for people researching symptoms with no
intention of booking. A software product pays for the words free, crack and alternative. The
negative list is therefore half the build, and it is maintained on a schedule rather than when
somebody notices the cost per acquisition drifting.
- Exact and phrase where intent is commercial; broad only where there is enough conversion data to
steer it.
- Search terms reviewed on a fixed cadence, with negatives applied at a level that will not also
block something useful.
- Brand and competitor terms handled deliberately — in some categories a competitor bid is the
cheapest conversion available and in others it is the fastest route to a complaint.
Performance Max, used carefully
PMax will spend a budget on placements nobody would have chosen, and its reporting is built to
make that hard to see. We run it where the account has enough conversion history to give the model
something honest to optimise towards, with asset groups split so the reporting still says something,
and exclusions set before launch. Where that history does not exist, we build it on search first. A
model trained on noise produces confident noise.
Tracking that survives consent mode
Consent mode, ad blockers and iOS have between them made naive conversion counting useless.
Server-side tagging, offline conversion import where the sale completes off the site — a booked
consultation, a signed matter, a policy issued — and a documented reconciliation between what Google
reports and what your own system recorded. The point is not a prettier dashboard. Bids are set
automatically against these numbers, and a model bidding on numbers that are wrong will spend
faithfully in the wrong direction.
YouTube
Treated as a paid social channel that happens to live inside Google Ads: creative-led, tested on
openings, useful mainly for demand not yet expressed as a query. Worth running once search is
saturated, rarely worth running before.
