Paid ads for real estate agents: the rules and the math

Less often than the category implies, and the honest number is worth knowing before you set a budget. The National Association of REALTORS® runs the only well-powered survey programme on this question: for the 2025 Profile of Home Buyers and Sellers it mailed 173,250 questionnaires in July 2025 and received 6,103 usable responses, an adjusted response rate of 3.5% with a stated 95% confidence interval of plus or minus 1.25%. That is unusually well documented for a trade-body study, and it is why numbers from this programme deserve more weight than the unsourced benchmark tables that dominate this topic.

In NAR's 2025 Home Buyers and Sellers Generational Trends Report, 40% of buyers reported being referred by a friend, neighbour or relative, and 17% had used the agent before. Everything an ad budget can plausibly reach adds up to roughly 15%: 7% inquired about a specific property viewed online, 6% arrived through a website without a specific reference, 1% saw the agent's social media page without a prior connection, and 1% came through a mobile or tablet application.

None of that is an argument against advertising. NAR's 2025 Profile of Home Buyers and Sellers found 88% of buyers purchased through an agent or broker and 91% of sellers sold with an agent's assistance, with for-sale-by-owner at an all-time low of 5%. The demand is there and nearly all of it routes through somebody. The argument is about job description. For most agents the real work of paid media is manufacturing the first conversation and feeding the referral engine, not replacing it. A budget sized as though ads were the primary channel is sized against a 15% slice.

The competitive field is thinning while that stays true. The California Department of Real Estate counted 408,393 active licensees in June 2026 — 116,784 brokers and 291,609 salespersons — down from 424,188 a year earlier and 433,510 two years earlier, a 5.8% contraction in 24 months. That is a licence count rather than an auction report — DRE does not publish who is advertising — but a thinning field is at least not a headwind. It does not change who reads the ads.

It applies, and there is no listing-size exemption. Meta's Special Ad Category is a mandatory classification for ads about housing, employment or credit, and a single promoted listing sits squarely inside it. Declaring it is not a compliance gesture that leaves your campaign intact. It removes controls.

Meta's Marketing API documentation states what goes away for housing ads in the US and Canada:

• Location targeting must use a radius of at least 15 miles (25 kilometres). ZIP codes, sub-city areas, neighbourhoods and metro areas are unsupported location types.
• Age is fixed at 18 through 65+, and selecting a specific gender is forbidden.
• Location exclusion is disabled, so you cannot carve a district out of your radius.
• Behaviour and demographic targeting are blocked, as are interest and detailed-targeting exclusions.
• Lookalike audiences are, in Meta's own words, "unavailable for housing, employment, and financial products and services ads."

Read that list next to a map and the campaign-structure consequence is obvious. A 15-mile radius drawn from downtown Salinas reaches Castroville, Prunedale and much of the Toro area. From Hollister it takes in San Juan Bautista and a long stretch of open county in every direction. You cannot narrow it and you cannot exclude out of it, so every bit of narrowing has to move downstream into the creative and the landing page. The ad copy names the neighbourhood. The landing page answers one question about one property. The targeting panel is no longer where relevance is built.

One warning about the guidance still circulating. Explainers that still rank for this topic recommend Special Ad Audiences as the compliant replacement for lookalikes. Meta's current documentation says plainly that lookalike audiences are unavailable for housing ads. Check the documentation rather than the guide, because those explainers mostly do not link the source document at all, and a page written against a 2019 snapshot of this policy is not merely stale.

Yes, and treating Fair Housing as a Facebook-only problem is the most common expensive misunderstanding in this category. Google's restricted targeting policy for personalized advertising states that in the United States and Canada, housing, employment and consumer-finance ads cannot target audiences using gender, age, parental status, marital status or ZIP codes. That is the same protected-attribute set Meta blocks, arrived at independently, and it applies whether or not you ever open Ads Manager.

The scope is wider than most agents assume. Google's policy FAQ confirms the housing designation covers all ad formats — text, display and video — across Search, Display and Video, and applies in Google Ads, Search Ads 360 and Display & Video 360. Advertisers who do not self-identify within the 60-day acceptance period find that, in Google's phrasing, "the ads that are labeled will stop serving," and that new campaigns can no longer be created in the account. That second consequence is account-wide, not campaign-level, which is what makes it worth handling before it happens rather than after.

One asymmetry in that table is worth the whole comparison. Google permits radius targeting down to a stated minimum of 1 kilometre; Meta's floor is 15 miles, roughly 24 kilometres. That is a twenty-fourth of the radius and something on the order of one five-hundred-eightieth of the ground area. For an agent working Pacific Grove, downtown Hollister or a single Salinas neighbourhood, that is the strongest structural argument for putting search budget ahead of social budget. Not creative preference, not audience quality. Geometry. It is why Google Ads and paid social get scoped differently for housing clients.

Because the delivery system decides who sees the ad, and it reads your creative before it reads your intentions. This is the one part of the topic with genuine experimental evidence behind it. In a controlled field experiment published at CSCW, Ali and colleagues (2019) ran housing ads on Facebook with identical targeting, identical bidding strategy and simultaneous scheduling. The estimated share of white users in the delivered audience still ranged from 27% for luxury rental ads to 49% for cheap house-purchase ads.

The same researchers isolated the cause. They stripped ads down to a bare link and added components back one at a time, and the image was what moved delivery. A bodybuilding ad reached a 91% male audience while a cosmetics ad reached a 5% male audience, despite identical targeting, bidding and budget. The headline and body text barely shifted delivery at all. A separate suite of real housing ads produced the same mechanism, with some ads delivering to audiences over 72% Black under identical targeting.

Budget moved composition too. Across daily budgets from $1 to $50, the fraction of men reached fell as the budget rose, with a Pearson correlation of −0.88 (p < 10⁻⁵) when targeting all US users and −0.73 (p < 10⁻³) for custom audiences. In that setting, the same creative at two different daily spends reached measurably different populations.

Meta has responded to this. The company reported launching its Variance Reduction System for US housing ads in January 2023, having an independent reviewer verify in July 2023 that the system met the compliance standards agreed with the Department of Justice, and extending it to employment and credit ads in October 2023. Those are the vendor's own statements about the vendor's own remediation, and they describe a correction applied after delivery starts rather than a guarantee that skew cannot occur.

The practical conclusion is short. Your photo selection is a targeting decision. The people in your listing images and the lifestyle framing of a carousel are inputs to an optimisation system that will find the audience most likely to engage with them, and Fair Housing law reaches the outcome regardless of what you typed into the targeting panel. Vary the creative deliberately, and treat a single hero image running unchanged for the life of a campaign as a risk rather than a simplification.

Substantially more than most national guides account for. The federal Fair Housing Act protects seven classes: race, colour, national origin, religion, sex, familial status and disability. That is the list every compliance checklist repeats, and for an agent working the Central Coast it is roughly a third of the exposure.

The California Civil Rights Department's fair housing fact sheet lists 20 protected characteristics in housing: the seven federal classes plus ancestry, gender, gender identity, gender expression, sexual orientation, marital status, military or veteran status, source of income, genetic information, age, citizenship, primary language and immigration status. Several of those are attributes a well-meaning listing ad describes without thinking — the language a rental listing is written in, the age framing of a 55-plus community, the phrasing around who a property "suits."

The advertising exposure is explicit rather than implied. CRD names "advertising or stating a preference for or against tenants with certain sources of income, such as: 'No section 8'" as an example of illegal housing discrimination. Source of income is not a federal protected class, which is precisely the trap: ad copy that clears review in Texas can be actionable in Monterey County, and platform approval is not a legal defence. Meta and Google review ads against their own policies, not against California's statute.

The drafting discipline that survives both layers is simple enough to apply without a lawyer on every ad. Describe the property, never the occupant. Say "three bedrooms, walk to the wharf" rather than "perfect for a young family." Name features, schools and distances as facts rather than as signals about who belongs there. Every claim in the ad should be checkable against the listing sheet.

Nobody has published a defensible one, and the sooner you accept that, the better your budget decisions get. The load-bearing citation for this entire category is a single marketing agency's annual benchmark table, whose stated methodology is one sentence about data from thousands of its own customers' campaigns. No account count, no campaign count, no date range, and no acknowledgement that the sample is self-selected toward accounts large enough to hire an agency. Article after article repeats its real estate row to the cent without disclosing that it has no published sample size and no year-over-year comparability. A number carried that far without a methodology is decoration, not evidence.

What you can anchor on is arithmetic you own. NAR's 2025 Member Profile, based on a March 2025 survey of 4,947 members, reported a median gross income of $58,100 for the typical REALTOR® in 2024 across a median of 10 transactions. That is the base rate any ad budget proposal has to sit against, and it is the number that makes most published return-on-ad-spend examples fall apart on inspection: they assume lead-to-close rates several times higher than anything NAR's data on how buyers find agents can support.

The other half of the arithmetic is local, and on the Central Coast it is unusually favourable. The California Association of REALTORS® reported June 2026 median existing single-family home prices of $1,350,000 in Santa Cruz County, $967,500 in San Luis Obispo County, $930,000 in Monterey County and $815,000 in San Benito County, against a statewide median of $904,640.

Run the calculation in that order and the lead-price question mostly answers itself. Take your commission on a median transaction in your county, apply your own honest lead-to-appointment and appointment-to-close rates from your own CRM, and you have a maximum defensible cost per lead. A price only means something against those two numbers. It also matters what you are counting: a downloaded home-value estimate and a booked listing appointment both get called a lead, and most published benchmarks never say which they mean.

On the auction itself, the direction of travel is mildly reassuring. Meta's 2025 results reported ad impressions delivered across its Family of Apps up 12% year over year while the average price per ad rose 9%, on total revenue of $200.97 billion. Inventory grew faster than price. That is not a promise about your account, but it does argue against the assumption that small advertisers are being priced out of the auction outright.

Start by noticing that nobody can answer this with a price, because Zillow does not publish per-lead pricing anywhere public. The head-to-head comparisons that rank for it quote a range for Zillow leads and a range for Google leads without citing anything at all. Those are anecdotes presented as benchmarks, and they are especially unreliable for a product priced by share of voice, where an agent buys a percentage of the impressions shown alongside listings in a defined area at a rate that floats with local demand. Two agents in adjacent markets can pay very different effective prices for the same conversation.

What is public is the financial filing. Zillow Group reported $2.583 billion of 2025 revenue, of which $1.704 billion was Residential — the segment that contains Premier Agent — alongside 221 million average monthly unique users and 2.1 billion visits in the fourth quarter alone. That is the only defensible way to reason about portal economics from outside the company: a reported segment revenue line and a reported audience number. Everything between them is unpublished by design.

The structural difference matters more than the price difference anyway. Portal advertising is a rental. The audience, the brand relationship and the data stay with the portal, and access ends the month the payments do. An ad account you run yourself against your own landing pages produces a first-party list, a remarketing pool, pages that keep ranking, and conversion data that stays yours. It also produces fewer conversations per dollar in the short run, because you are building the demand rather than standing next to inventory that already has it.

So the reasonable posture is not loyalty to either model. Portals buy volume at a price you do not control and cannot audit. Your own campaigns buy fewer conversations plus an asset that compounds. Which deserves the marginal dollar depends on whether you need transactions this quarter or a pipeline next year, and most working agents need both. We size that split on the real estate side of our practice against a client's own closed-transaction data, not against a published benchmark.

It looks like a funnel with a deadline attached. NAR's generational data shows buyers searching for a median of 10 weeks and viewing a median of seven homes, but contacting an agent after a median of just two weeks. The window in which paid advertising can reach an unrepresented buyer is roughly a fortnight wide. After that, the buyer has an agent, and your impressions are being served to someone who already made the decision your ad exists to influence.

The deadline gets sharper further down. NAR's 2025 Generational Trends Report found 75% of buyers interviewed only one real estate agent before deciding who to work with (Exhibit 4-6), and the median buyer contacted that agent just once before getting a response. NAR's later 2025 Profile of Home Buyers and Sellers reports the same behaviour split by cohort rather than as a single figure — 76% of repeat buyers and 67% of first-time buyers — so expect to see both quoted. Being first is usually the same thing as being chosen. Being fourth is usually the same thing as not existing. Every routing decision in the response layer — who gets the notification, what happens after hours, whether a missed call triggers a text — is worth more than another round of headline testing.

Be careful which speed statistic you quote for this, though. The famous "21 times more likely to qualify" figure traces to a 2007 InsideSales.com/MIT lead response study covering three years of data from six companies, over 15,000 leads and over 100,000 call attempts. It found the odds of contacting a lead were associated with a 100-fold drop, and the odds of qualifying one with a 21-fold drop, between a five-minute and a thirty-minute response. It is observational, nearly twenty years old, and routinely misattributed to Harvard Business Review. Its direction is almost certainly right. NAR's current, survey-backed numbers make the same argument without any of the baggage.

The last stage is the one most accounts skip. Both platforms bid toward whatever you tell them to value, so if a form fill is the conversion, you will reliably be sold cheap form fills. The fix is offline conversion measurement: capture the Google Click Identifier on the landing page, store it in the CRM alongside the enquiry, and send the later events back — appointment held, listing agreement signed, transaction closed — with a value attached. Google's documentation reports that advertisers using first-party data such as email addresses and phone numbers alongside imported GCLIDs saw a median 10% increase in conversions. That is a vendor figure with no published methodology or sample size. The mechanism stands on its own: a bidding system optimising toward closings behaves differently from one optimising toward whoever fills in a form most cheaply, and wiring it up is the highest-value hour in a real estate paid ads account.