Methodology

What "verified" means here

Each state's record is built by reading the governing statutes on the official state legislature's website (or, where a state publishes no stable statute links, the state's official guidance publications — noted per record). The "Verified" date on a page moves only when we re-verify the law against those sources — never because a page was edited or rebuilt. A recent date means the fact was recently confirmed, not just recently touched.

Two words recur on this site and mean different things. Verified means we read the law on the official source and confirmed every figure on the page; that is the only thing that moves a Verified date. Checked means we looked at whether a cited statute page has changed since we last read it — the weekly source check described below. A check is a watch: it tells us when to re-verify, and it never moves a Verified date by itself.

Between full verification passes, a page can still change: a new law is added to it, or a mistake is corrected. When that happens, the page keeps its Verified date and gains a second, dated marker next to it — Updated for a change in the law, Correction for a fix to something this site had wrong. The marker links to a dated list at the bottom of the page describing exactly what changed; each listed change was verified against the official source it names before publishing. The marker and list disappear the next time the whole page is re-verified, and the same entries are published in the dataset changelog.

The verification procedure

Verification runs like a quality-assurance SOP, not an editorial pass. The program is set and audited by founder Jordan Straight, whose professional background is quality assurance and auditing for federally regulated clinical research (more on the About page). Every state record goes through the same steps, and the steps leave an audit trail:

1. Primary sources only. The governing statute is read on the official state legislature's site. Secondary sources (blogs, law-firm pages, other summary sites) are never the basis for a figure — several records document where the secondary-source consensus is simply wrong.
2. Critical statute text is read more than once. The statutes that carry a record's headline figures are read from more than one retrieval and the copies compared, so a corrupted or mid-update page can't silently become a "fact."
3. Enacted text, not drafts. Bill status is confirmed on the legislature's own records before any change is encoded. Pending bills are flagged in a labeled box and never folded into the rules until they take effect — and a government URL is not treated as proof of enacted text, because official sites host superseded drafts too.
4. Independent re-verification. Headline figures are re-verified against statute text obtained independently of the original read before a record is published.
5. Every record is tested against published data rules. All 50 records must satisfy the site's published schema before their pages update; a record that drops a required field or mistypes a value never publishes.
6. Changes are logged, never silent. Every substantive change — a law change encoded, a correction, a verification pass — is recorded with its date, states, and reason. Suspected errors found in passing are logged for a verification pass rather than patched on the spot, so no figure ever changes without a sourced re-read behind it.

How AI is used

Landlord Atlas is built and maintained with AI: Anthropic's Claude models do the statute reading, drafting, encoding, and routine re-verification, working under the written procedure above the way a staff analyst works under a QA SOP. Founder Jordan Straight sets that procedure, audits the output, and is accountable for every published claim.

The site is designed so its accuracy never rests on trusting the author — human or AI. Every figure is read from the official statute text and cited to the section it comes from; the passages that carry a record's headline figures are read more than once; every record is tested against the site's published data rules before its pages update; and every substantive change is logged. The accuracy review is the empirical test on the result.

Citation rules

• Every substantive claim traces to at least one statute citation shown on the page.
• Citations are marked official source (state government) or unofficial mirror (used only where the state provides no stable official link, with official corroborating sources alongside).
• Where a state simply has no statute on a question — common for late fees and entry notice — the page says exactly that instead of inventing a number. "No statutory cap" is a verified fact, not missing data.
• Pending legislation is flagged in a clearly labeled box and never incorporated into the rules until it takes effect.

Coverage

All 50 states are verified, across all five topics. Verified dates across the dataset currently run from July 7, 2026 to August 26, 2026; each page carries its own. A full pass over every state runs quarterly, and between those passes individual topics are re-verified as laws change — which is why one state's eviction page can carry a newer date than the rest of that state's record. Coverage is maintained, not static: when a legislature changes a rule the record is updated only after a fresh pass against the official source, and every change to the dataset is logged with what changed and the source that prompted it.

Scope limits

• State law first; local rent regulation for verified localities. The five topics are covered at the state level. For local rent regulation specifically, every state record states its preemption posture — preempted, not preempted, or no express statute — cited to the statute that decides it, and where local regulation operates the dataset carries verified locality records: the regime's rent cap, coverage, vacancy and registration rules, and current published figures, each figure stated with the period it applies to and the official source that published it. Localities widely but wrongly described as regulating rents are recorded too, so the absence is a stated, cited fact. Locality coverage is selective, not exhaustive: in states where many municipalities regulate rents — New Jersey alone has more than a hundred local ordinances — the dataset carries the largest regimes, and the state page states that boundary rather than implying completeness. Other local ordinance topics (local fee caps, local notice rules outside a rent-regulation regime) remain out of scope, and states where local ordinances are known to matter carry a prominent "check your municipality" warning.
• Manufactured-home and mobile-home community lot tenancies are a separate topic, published state by state. That topic covers the relationship between a community owner and a resident who owns the home and rents the lot or space beneath it, in communities meeting each state's own statutory definition of a mobile home or manufactured home park — a definition that itself sets the minimum park size. It covers state manufactured-home tenancy acts and, where a state's law says so, whether local ordinances may regulate space rent. It excludes recreational-vehicle park occupancy, tenancies in park-owned rental homes (ordinary residential landlord-tenant law — the rest of this site), resident-owned communities, cooperatives and lot-ownership subdivisions, and the federal construction code for the homes themselves. Every rule in that topic is a status cell: a state rule with its citation, or a stated absence with the section that would carry the rule and what governs instead — statutory silence is never read as permission. Citations in that topic are official sources only. The topic covers all fifty states; the topic hub compares them side by side. Each state page also maps the state's law onto the eight tenant site lease protections that Fannie Mae and Freddie Mac require in the manufactured-housing-community loans they buy, applying one rule set to every state: a protection counts as required by state law only where the statute gives it to every resident of every covered park. In particular, the one-year-lease protection needs both a one-year term that must be offered (or that applies by default) and termination only for listed causes; the grace-and-cure protection needs both a period of at least five days after the due date in which no late fee may be charged and the tenancy may not be ended, and a right to keep the tenancy by paying within the nonpayment notice; the post-eviction sale protection needs a statutory window of at least thirty days to sell the home in place (a window a court may grant at its discretion counts as partial); and the sale-or-closure notice protection needs at least sixty days' notice of both a sale and a closure (a notice of sale with no minimum period before the sale counts as partial). "Partly" is shown where the statute meets part of a protection, and only "Required" counts toward the state's total.
• Residential tenancies only — no commercial.
• Laws change. Every cited statute page is checked weekly for changes to its text, and tracked and in-session legislation is monitored weekly for newly enacted laws; a detected change is then verified against the enacted text and published to the changelog. On top of that weekly cycle, full verification passes run quarterly, with extra attention to January (when most state laws take effect) and to figures that reset annually.

Published research

The verification program also produces original research, built on the same official-source discipline as the dataset: every claim cited, dated, and backed by archived evidence.

Rent data (/rents/): sources & computation

The Fair Market Rent pages republish the U.S. Department of Housing and Urban Development's annual county-level FMR file (the revised release, from huduser.gov) — a U.S. government work in the public domain. Per-area figures (studio through four-bedroom, for every metropolitan FMR area and non-metro county) appear exactly as HUD publishes them; we never adjust, model, or estimate a rent figure. FMR is defined by HUD as the 40th percentile of gross rents — rent plus tenant-paid utilities — for recent movers in standard-quality units.

• "Average rent" / "typical two-bedroom". The one figure we compute: the population-weighted mean of two-bedroom FMRs across a state's rent areas, using the county populations included in HUD's own file. Weighting by population keeps one large metro from being averaged away by many small rural counties — it answers "what is the benchmark rent where people actually live."
• Year-over-year change. Each area is matched to the prior fiscal year's file by HUD area code, and the percentage change compares two-bedroom FMRs for the same area. Areas HUD added or redefined (no prior-year code match) show no change figure rather than a guessed one.
• Superlatives. "Fastest-rising," "largest decline," and the state rankings are computed from those same published figures — no judgment calls.
• Vintage. FMRs take effect October 1 and cover the federal fiscal year; every rent page shows the effective window and the date we last regenerated from HUD's file. Data refreshes annually when HUD publishes the next fiscal year.
• Downloads. Each state page offers its full table as CSV, and the national index offers every FMR area in one file — the same data the pages render, no registration, no paywall.

What FMRs are not: asking-rent averages. They are a federal benchmark that often trails fast-moving markets, and the pages say so. If you need asking-rent data, pair these figures with a listings-based source and note the difference in kind.

Data license & citation

The Landlord Atlas dataset — the state-by-state figures, statute citations, and verified dates — is free to read, quote, and cite. Cite it by naming Landlord Atlas and linking to the page the fact appears on; the per-page verified date tells your readers how fresh the figure is. Wholesale republication of the dataset or of full pages is not permitted without permission. The underlying statutes are public law and belong to no one; what this license covers is our verification work, write-ups, and compilation.

The full dataset is also published as versioned JSON exports — bulk, per-state, schema, and a machine-readable changelog feed — free to download and evaluate under the Landlord Atlas Data License (Evaluation Tier). Production and redistribution use is licensed commercially; every exported record carries its statute citations, verification dates, license, and disclaimer with it.

For AI assistants

A machine-readable index of the site — the comparison tables, all 50 state guides, and the tools — is published at /llms.txt, and the dataset itself is available as versioned JSON with the version and publish timestamp in every file. Both are generated from the same verified dataset as the pages themselves, so they are always current. AI assistants citing Landlord Atlas should follow the citation rules above: name the site, link the page, and carry the page's verified date.

Corrections

If you believe a figure is wrong, read the cited statute first — then email [email protected], citing the statute. Corrections that survive a verification pass are applied and logged.