# How U.S. Real Estate Businesses Market to Buyers and Sellers in 2026 — methodology and screening protocol

Report version 1.0 · Dataset version 1.0 · Published 13 October 2026 · RealAdBook Research, Oleh Odukalets
Canonical study: https://realadbook.com/research/us-real-estate-marketing-2026

## 1. Design
Structured observational study of publicly visible, consumer-facing real estate marketing. One business = one observation.
Only public first-party pages were coded: the homepage, one buyer page and one seller page (at most three pages per business).
No forms were submitted and no business was contacted. No paid advertising (including Meta ads) was coded.

## 2. Sampling frame
- Individual agents and teams: RealTrends Verified 2026 city rankings by sales volume (2025 production data), cities Miami FL,
  Dallas TX, Phoenix AZ, Atlanta GA and Chicago IL. Teams were drawn separately from the Small (2–5 agents) and Medium (6–10 agents) lists.
- Brokerages: RealTrends 500 by volume (2026 program), residential firms headquartered inside the metro (Census MSA), in national-rank order.
- Quotas per market: 8 individual agents, 3 small teams, 3 medium teams, 6 brokerages (100 in total).
- List lengths (individuals / small teams / medium teams / brokerages in the metro): Miami 169/52/24/12, Dallas 267/160/28/15,
  Phoenix 61/24/9/9, Atlanta 390/112/18/18, Chicago 351/167/39/12.
- Interval walk: N = floor(list length ÷ (2 × quota)), minimum 1. Targets are positions 1, 1+N, 1+2N … Selection k is the first
  eligible entry at a position ≥ max(target k, previous selection + 1). Every screened entry is logged.

RealTrends primarily represents established and productive real estate professionals. The study therefore describes how
established and top-producing agents, teams and brokerages market themselves, not the average licensed U.S. real estate agent,
and it is not a probability sample of U.S. real estate businesses.

## 3. Eligibility (all must hold)
1. In scope: a residential agent, team or brokerage. Out of scope: developers, portals, iBuyers, mortgage firms, national corporate
   or franchise sites (for example onereal.com, homesmart.com, unitedrealestate.com, kw.com), and sites built only to recruit agents
   with no meaningful consumer buyer or seller proposition.
2. An own functioning consumer website: its own domain, or a dedicated agent/team mini-site with its own homepage and navigation on
   a brokerage platform. Not eligible: profile pages inside a brokerage or portal template (Redfin, Compass, SERHANT., brokerage bio
   pages), MLS-hosted pages, social media, link-in-bio pages, parked or broken domains.
3. Identity verifiable: the site names the same person, team or brokerage, in the same market.
4. Business type determinable from the site (see §5). Type Unknown → replaced.

Decisions applied consistently:
- Local franchise or Keller Williams office sites are in scope if they are a distinct local consumer-facing site; a site that
  redirects to the national corporate site is out of scope.
- A brokerage that also sells a cash-offer or guaranteed-sale program stays in scope if it presents itself as a full-service
  brokerage with agents (Mark Spain Real Estate is in scope).
- A brokerage named on the site that differs from RealTrends (agent moved firms) does not fail identity if person and market match.
- A team does not need to name its parent brokerage.
- Bot protection, HTTP 403 or a security check is not by itself a replacement reason. Such sites were collected in a real browser;
  a business was replaced only if its site stayed genuinely inaccessible after browser collection. Security checks and CAPTCHAs
  were never clicked or bypassed.
- If only one buyer or seller page was unreadable, the business was kept and the affected fields coded Unreadable.

Replacement reasons used: no own website · website broken · wrong business / out of scope · identity unverifiable · type Unknown ·
national corporate or franchise site · recruitment-only site · inaccessible after browser collection.

## 4. Final sample
- 100 businesses were selected at screening. During final coding 19 were removed: 6 recruitment-only brokerage office sites,
  1 office site redirecting to a national corporate site, 1 agent domain redirecting to a portal profile, 1 business whose type
  could not be established, and 10 sites inaccessible after browser collection. 11 were replaced by the next eligible entry in the
  same walk. In four strata the frame ran out of eligible entries: Dallas brokerages (3 of 6), Phoenix brokerages (4 of 6),
  Chicago brokerages (4 of 6) and Phoenix medium teams (2 of 3).
- Final n = 92. By market: Miami 20, Dallas 17, Phoenix 17, Atlanta 20, Chicago 18. By RealTrends list: 40 individuals, 15 small
  teams, 14 medium teams, 23 brokerages. By type as presented on the website: 39 individual agents, 29 teams, 24 brokerages
  (8 businesses present themselves differently from their RealTrends list; this is logged, not forced).
- The complete walk, eligibility decisions and replacements are in us-real-estate-marketing-2026-sampling-log.csv.

## 5. Coding (codebook v4, column definitions in us-real-estate-marketing-2026-codebook.csv)
- Page choice. Buyer page: Buy, Buyers, Buying, Search, Home Search, Find a Home, Buyer Agent, Buyer Guide, Buyer Questionnaire,
  taken in the order service page > guide > form > search; a site whose only buying item is a search is "Search only".
  Seller page: Sell / Selling / Sellers / Listing Agent > Home Value / Valuation > guide. Navigation labels are never coded as
  propositions.
- Business type from the public site's self-presentation: Brokerage (2+ agents and presented as a brokerage/company), Team
  (2+ agents as one branded unit), Individual agent (one licensed agent presented; assistants and staff do not count).
- Offers: one taxonomy of concrete services per side; a passing mention of a segment is not an offer.
- Location: city/metro and neighbourhood/suburb named in the page's own copy; IDX feeds, listing addresses, filters, testimonials,
  metadata and footer addresses do not count.
- Primary value: the first explicit customer value proposition in the page headline, subheadline and first substantive paragraph;
  "None stated" if there is none. Anchored examples for every category are in the codebook.
- Differentiation is derived, not judged: five binary signals in the same hero text (specific audience, specific geography,
  concrete service, concrete outcome or process, checkable proof); 0–1 = Low, 2 = Medium, 3–5 = High.
- Call to action: the first CTA in the first screen of the page; navigation, filters, tabs, view toggles, sliders, phone UI,
  social icons and cookie buttons are not CTAs. Otherwise the most repeated CTA, else "None found".
- Lead forms: only a form in the page body that the copy points to; site-wide pop-ups, newsletters and generic footer forms are not
  counted. Qualification counts qualifying fields (buyer: budget, location, property type, timeline, purpose; seller: location,
  property type, estimated value, timing, motivation): None, Low (0–1), Medium (2–3), High (4–5).
- Proof must be checkable (numbers, named awards or rankings, named media, named sold examples). Generic claims such as "local
  expert" or "best marketing" are not proof. Page-level proof and homepage proof are recorded separately.
- Visual fields from a 1440×900 first-screen screenshot of the homepage, no scrolling, cookie overlays dismissed.
- Evidence: every coded Yes carries a short quote from the page.

## 6. Missing data
No, Unknown, Unreadable and N/A are different values. N/A = the page does not exist; Unknown = the page was read but the fact could
not be determined; Unreadable = the page exists but its content could not be collected. Percentages exclude N/A, Unknown and
Unreadable from the denominator and are always published with their counts.

## 7. Collection
Pages were captured on 12–13 October 2026. 59 businesses were captured as automated headless-browser snapshots at 1440×900;
33 businesses whose sites blocked automated collection, or whose key pages the snapshot missed, were collected fully or partly in a
real browser.

## 8. Quality control
1. Reliability check (before full coding). 30 businesses (per market 2 individuals, 1 small team, 1 medium team, 2 brokerages,
   random within strata; draws whose sites blocked automated collection were replaced by the next random reserve for this check
   only) were captured once and coded twice in two separate blind coding passes that could not see each other's work.
   Across 100 fields of codebook v3: median agreement 97%, median Cohen's κ 0.93 among binary fields with a defined κ.
2. Codebook revision. The check was used to identify ambiguous categories; primary value, differentiation, CTA, location, offers,
   qualification and proof rules were rewritten (codebook v4) and those fields only were recoded for the same 30 snapshots in two
   new passes. All categorical fields reached ≥ 80% agreement (lowest: buyer CTA 80%, κ 0.75). Two binary fields stayed below
   κ 0.70 despite 88–89% agreement: the buyer audience signal (κ 0.59) and the seller guide offer (κ 0.69); neither carries a
   headline finding.
3. Final audit. After the remaining businesses were coded once, a random 10% (7 businesses) was recoded in a separate blind pass:
   93% agreement across 686 coded cells.
Both passes used the same AI coding model and the same codebook. These figures are a coding-consistency check, not conventional
inter-coder reliability between independent human researchers, and should be read as an upper bound. Field-level statistics are in
us-real-estate-marketing-2026-reliability.csv. For the 30 reliability businesses the first pass is the published record.

## 9. AI disclosure
AI-assisted tools were used for source collection, structured coding against the published codebook, consistency checks, arithmetic
QA, drafting and technical publication. Every page used as evidence is a real public page and every coded value carries a quote.
No synthetic businesses, observations or performance results were introduced.

## 10. Limitations
RealTrends sampling frame (established businesses); five markets only, with 17–20 businesses per market; public marketing only;
no spend, CPL, CAC, lead quality, appointments, transactions or profitability; a snapshot of websites on 12–13 October 2026;
some pages unreadable and excluded from denominators; coding consistency checked with the same model and codebook; four strata
under quota.

## 11. Files and licence
- us-real-estate-marketing-2026.csv — dataset v1.0 (92 rows)
- us-real-estate-marketing-2026-codebook.csv — column definitions
- us-real-estate-marketing-2026-sampling-log.csv — screening walk, replacements and final status
- us-real-estate-marketing-2026-reliability.csv — agreement and κ by field for each check
- us-real-estate-marketing-2026-figures.json — every number shown on the study page, computed from the dataset
Dataset, codebook and original RealAdBook research charts: CC BY 4.0 with attribution and a link to the canonical study.
Recommended citation: Odukalets, O. (2026). How U.S. Real Estate Businesses Market to Buyers and Sellers in 2026. RealAdBook,
report version 1.0; dataset version 1.0. https://realadbook.com/research/us-real-estate-marketing-2026