Google’s machine learning is impressive. It’s also… confidently wrong, surprisingly often. If you let it “get on with it,” it’ll happily expand your reach into places you never meant to be — and bill you for the privilege.
Case in point from our podcast:
- A Building/Property Management company (they manage large blocks for landowners) repeatedly misread by Google as an Estate Agency. Two very different businesses. Result: clicks from people looking to buy or sell homes — useless for a block management provider.
- Home Air Conditioning companies showing for Car Air Con Re‑Gas searches. Not even the same category. Same credit card charge.
This isn’t malice. It’s how intent-expansion works: Google groups “themes,” “close variants,” and “similar intent” to find more inventory. Sometimes it nails it. Sometimes it confidently wanders off into the weeds. Your job is to catch it quickly — before your budget bleeds out on the wrong searches.
Why this happens (without the fluff)
- Match types are looser: Exact and Phrase now include “close variants” and “same intent.” Your meticulous keywords don’t fence the machine in like they used to.
- Broad match + Smart Bidding: Great for scale, but it uses signals, not syntax. If your signals are fuzzy (or your niche is nuanced), Google guesses. Guessing costs money.
- Asset and theme expansion: Responsive ads and PMax test wide. Without strong guardrails, they’ll match adjacent categories you never intended.
Translation: Google optimises for clicks/conversions, not brand nuance. If “estate management” and “estate agents” look statistically similar, it’ll blend them — unless you stop it.
The discipline: expand with one hand, exclude with the other
You can have scale and control. But you need a standing operating procedure (SOP) that assumes drift and fixes it fast.
- Build “Always‑On” negative lists
- Jobs & Careers: jobs, vacancies, careers, recruitment, indeed, salary, apprentice
- Training & Accreditation: course, training, qualification, certificate, NVQ, CPD
- DIY/Info/Free: definition, what is, wiki, template, example, free, cheap, DIY
- Category‑Confusers (your bespoke list):
- For Block/Property Management (B2B): “estate agent, lettings, to let, for sale, mortgage, conveyancing, house valuation, landlord services (if not relevant), rightmove, zoopla”
- For Home AC Installers: “car air con, regas, recharge, automotive, garage, MOT”
Apply these as shared lists across all non‑recruitment/non‑training campaigns.
- Ring‑fence your intent themes
Structure by intent, not just keywords:
- Brand Protect: your brand terms. Keep tight control, exclude career/training/blog URLs.
- Core Commercial: the money terms. Use phrase/exact plus carefully curated broad. Layer audiences. Aggressive negatives.
- Adjacent/Pilot: where you test broader themes with capped budgets and stricter SLAs for search term reviews.
- Competitor/Conquest: surgical, with tailored negatives to avoid generic research clicks.
- Tighten your landing page and asset signals
- Use copy that screams your category so Smart Matching has fewer excuses: “Block & Estate Management for Freeholders and RTM Companies” (not “Property Services”).
- Add disqualifiers on-page: “We do not sell or let residential property.” That line can reduce irrelevant Quality Score drag and filter enquiries.
- Site structure: keep “Careers,” “Blog,” and “Resources” out of paid paths. Exclude those URLs in campaign settings.
- Weekly search term triage (non‑negotiable)
- 15–30 minutes, every week. Sort by spend and add negatives at the theme level (not one-offs only).
- Promote repeat offenders to your shared negative lists so the whole account benefits.
- Watch for new look‑alike clusters (e.g., “block management” getting matched to “student halls management” if that’s irrelevant).
- Smart bidding with guardrails
- Start with realistic tCPA/tROAS targets based on actual funnel data (not vibes).
- Use minimum conversion volume per campaign before widening match types.
- Daypart and geo-trim based on real conversion hours/locations; don’t fund late‑night curiosity clicks if they never close.
- Measurement that keeps the machine honest
- Track qualified conversions (SQLs, sales, or weighted values) — not just form fills.
- Import offline conversions (GCLID/GBRAID back into Google Ads) so the algorithm learns what a “good” lead actually looks like.
- If you sell B2B services like block management, assign higher values to the right ICP forms and zero‑value “wrong category” submissions.
Two mini playbooks
Property/Block Management (not an estate agency)
- Campaign naming themes: “Block Management – Freeholders/RTM,” “Property Management – Commercial Blocks,” “Brand.”
- Must‑have negatives: estate agent(s), lettings, to let, for sale, valuer, valuation, mortgage, conveyancing, rental, tenant find, landlord services (if not offered), rightmove, zoopla, house, flat, apartment for sale/let.
- RSA assets: “We Manage Residential Blocks for Freeholders & RTMs,” “Service Charge & Compliance Experts,” “Not Estate Agents. Specialist Block Managers.”
- On-page qualifiers: “We don’t buy/sell/let property.”
Home AC Installers (not automotive AC)
- Must‑have negatives: car, automotive, regas, recharge, refrigerant top‑up, garage, MOT, mechanic, aircon gas, R134a, R1234yf.
- RSA assets: “Home Air Conditioning Supply & Install,” “Domestic Split Systems,” “No Car AC/Re‑Gas Services.”
- Landing page: hero copy and imagery showing domestic installs, not under-bonnet visuals.
Red flags that mean you’re leaking money
- High spend in “search categories” that aren’t your category (check Insights > Search themes).
- Rising CPC with flat or falling CVR after broadening match types.
- Spike in “contact” form fills that sales marks as “wrong service” or “not our ICP.”
- RSAs stuck on low ad strength because assets try to cover multiple categories.
Fix order:
- Block the wrong themes with shared negatives.
- Re‑write RSAs to be more specific and disqualifying.
- Trim geos/time slots that never convert.
- Re‑set targets to sane levels; give the system a week to relearn.
The cheeky bottom line
Google is brilliant at spending your money. It’s less brilliant at understanding your nuance. That’s your job. Don’t abdicate. Set lanes with negatives, reinforce your category in copy and pages, review search terms weekly, and teach the algorithm what a good lead actually looks like.
Let the machine explore — on a leash.

Chris Barnard has spent over 15 years delivering exceptional revenue growth for ambitious businesses in the UK, Europe and North America through his marketing technology business, FeedbackFans.com and as an independent business consultant.
By his mid-20’s he was running digital departments for FTSE100 companies in London, eventually leading to a very successful period in digital customer acquisiton for a well-known brand in his early 30’s generating nine-figure revenues with seven-figure budgets. He now puts his experience, knowledge and ideas into good use, supporting challenger insurgent brands and forward thinking businesses to outperform in their sectors, whilst disrupting and improving the marketing, technology and development sectors.
Feedback Fans provides a unique next-generation managed technology and marketing platform that delivers outstanding and out-sized results for businesses in sectors such as finance, retail, leisure, and professional services.
With our unparalleled expertise in creating cutting-edge solutions and environments, we empower our clients and users to thrive and outperform in the digital age.
Chris Barnard is Managing Director of FeedbackFans.com and producer of the Bear Business Vodcast