Two Front Doors

Most advertisers in a regulated category run one search engine properly, run the second one by pressing import, and never notice that they have just copied a set of targeting assumptions into a platform with different rules.

Real estate is the sharpest version of this because housing is a restricted category, the two engines restrict it differently, and one of those two publishes far less about how.

Start with what is actually written down

The industry's shorthand after the fair housing settlements was that local targeting is over. We went and read Google's policy rather than the commentary, and the shorthand is wrong in a way that matters.

Google's own wording: "Radius targeting requires setting at least 1 km around any given location."

One kilometre. Local precision on search survived. What did not survive, for housing in the US:

ZIP codes. Prohibited. Radius, city and country remain. In Canada the first three characters of the postal code are still permitted.

Gender, age, parental status and marital status. All prohibited as targeting criteria.

Advertiser-curated audiences. Customer Match, your own data segments, audience expansion and lookalike segments are all barred for housing. Your first-party list, the asset everyone spent a decade telling you to build, cannot be pointed at this category.

Compare that with Meta, where the housing floor is fifteen miles. Same regulatory origin, radically different practical outcome. An advertiser who assumes the platforms converged is leaving precision on the table at one door and breaching policy at the other.

The second door is the one nobody reads

Microsoft's position is that an advertiser in housing services may not use postal code to personalise, segment or profile. So the ZIP restriction travels.

What we could not find is a Microsoft housing policy page with the specificity of Google's. Google publishes the radius minimum as a number. Microsoft's guidance on this category is thinner and harder to locate. We are stating that as a research finding rather than an accusation: it is entirely possible the rules are equivalent and simply less well documented.

But the operational consequence is real, and it is the point of this piece. Import is the default workflow. Everyone builds in Google and pushes the campaign across. That import carries keywords, geo settings, audiences and negatives into an environment whose published rules you have not read, because they are harder to read.

Importing a campaign is importing a compliance posture. Nobody thinks of it that way, and it is exactly how a governed account acquires an ungoverned twin.

What the channel now costs

WordStream's 2026 benchmarks, from 13,474 US search campaigns run between April 2025 and March 2026, put real estate at a median $3.22 cost per click, 7.61% click-through, 3.70% conversion and $102.51 per lead.

The figure that should govern the decision is the change rather than the level: real estate CPC rose 27.27% year over year, the largest increase of any industry in the study.

A channel getting that much more expensive that much faster is not a channel to run casually. It is a channel to run with negatives, tight match types and a conversion definition that means something.

Why this is a governance problem, not a media one

Everything above could be handled by a careful person with a checklist. The reason it is not is that the careful person is one holiday away from not being there, and the campaign keeps running.

This is the case we built AEGIS for. A restricted category should not depend on somebody remembering that ZIP targeting is prohibited, that marital status is off the table, that the customer list cannot be attached, and that the second engine has its own rules. Those become preflight constraints on how a campaign can be constructed at all, checked before anything is allowed to spend, on every platform, every time, with the reasoning written down.

The point is not that a machine is more diligent than a person. It is that a machine is diligent on the same Thursday afternoon in August that a person is not.

The playbook

1. Read the primary policy, not the summary

The 1 km figure is published by Google and is widely reported wrong. We found secondary sources quoting fifteen kilometres, and our own internal memo carried that error until this piece corrected it. If a number governs what you are allowed to do, get it from the source.

2. Treat each engine as a separate compliance surface

Same category, different rules, different documentation quality. The import button copies settings, not permission.

3. Turn the restrictions into build-time constraints

A campaign that cannot be built compliantly should fail at construction, not at review. Post-hoc auditing finds the breach after it has served.

4. Rebuild the audience strategy around intent, not identity

When demographics and first-party lists are off the table, what is left is what somebody typed. That is a narrower instrument and a more honest one, and it rewards keyword and negative discipline far more than it rewards clever segmentation.

5. Define the conversion as something that has value

At a hundred dollars a lead and CPCs rising twenty-seven percent a year, optimising toward form fills is how a budget disappears. Optimise toward the appointment.

6. Keep the record

In a category with federal exposure, being able to show what was targeted, when, and who approved it is worth more than any individual campaign. Build the log before you need it.

Our position

Two front doors, two sets of rules, one import button between them. That is the actual shape of search advertising in a restricted category, and almost every operator we see treats the second door as a copy of the first.

The interesting part is not that the rules are strict. It is that they are strict, asymmetric, and in one case not clearly published, which means compliance cannot be a thing somebody remembers. It has to be a property of how campaigns get built.

Go and read the policy. Then go and check what your second engine is actually running.

Sources

Google Advertising Policies, "Housing in personalized advertising" and "Restricted targeting in Personalized advertising", for the 1 km radius minimum quoted verbatim, the ZIP code prohibition, the prohibited demographics including marital status, the prohibition on Customer Match, your data segments, audience expansion and lookalike segments, and the Canadian FSA exception. Both pages checked directly.

Microsoft Advertising policy on personalised advertising for housing and related services, for the postal code restriction. We could not locate a Microsoft housing policy page with the specificity of Google's, and have said so above rather than implying the rules are identical or absent.

WordStream, 2026 Google Ads Benchmarks, for the real estate medians and the year-over-year CPC change. Study covers 13,474 US search advertising campaigns from 1 April 2025 to 31 March 2026, minimum 52 campaigns per subcategory, median values.

A correction we are making in public. Our own internal real estate research memo recorded Google's housing radius minimum as 15 kilometres and flagged it as unverified. Checking the policy directly for this piece established that the figure is 1 kilometre. The memo has been corrected. Fifteen kilometres and one kilometre are the difference between a strategy and a wasted channel, which is why the flag was there.