Parcel Data Options: County-Direct, Aggregators, and Platforms

A parcel record is the smallest unit of real property analysis, and it is also the least standardized. Every commercial site screen, every land metric, every zoning check and every collateral map resolves to a polygon and a row that some county or town produced for the purpose of levying tax. The question this piece answers is not what a parcel record contains. It is where a lending team should get one, on what terms, and how to tell whether the source it is being offered can support the use it has in mind.

There are three procurement paths and no fourth. A team can go to the producing jurisdictions directly, county by county or town by town. It can license a national layer from an aggregator that has already collected those jurisdictions and normalized them into one schema. Or it can consume parcels inside an analytics or mapping product, where the parcel layer is an input to something else rather than the deliverable. Each path answers a different question well and a different question badly, and the choice is rarely permanent: most working stacks end up mixing two of the three.

The evidence in this piece is deliberately narrow and deliberately checkable. Nine public parcel sources were read on August 24, 2026, six statewide programs and three counties, and scored against a ten test checklist that a credit team can rerun. Six of those sources published live attribute services whose field lists were read the same day. Four commercial pages were read for what the vendors themselves say about coverage, lineage and refresh, and every such statement below is presented as the vendor's claim rather than as verified fact. National jurisdiction counts come from the U.S. Census Bureau and from the National States Geographic Information Council's 2025 Geospatial Maturity Assessment, published on August 14, 2026.

The counterintuitive result is about what aggregation is actually worth. The normalization an aggregator sells is worth most exactly where the underlying jurisdictions are already good, and worth least where a lender most needs help. Across the states whose parcel programs apply a verified, quality controlled standard, 73.9 percent of jurisdictions are publicly reachable; across the states whose programs take county data as received, the figure is 46.6 percent (National States Geographic Information Council, 2025; jurisdiction sums by MMCG). And 764 jurisdictions inside states that run a program publish nothing at all, which no aggregator can license into existence. The category's own coverage claims are denominated in population rather than in jurisdictions, which is exactly why a page can accurately state coverage of roughly 99 percent of Americans while several hundred assessing jurisdictions remain unobtainable at any commitment.

What follows sits inside the wider question of choosing a CRE analytics stack. It assumes the diagnostic work has been done elsewhere: the companion survey of openness and quality of parcel records by state covers who publishes what and why, and the study of land metrics computed from parcels covers what breaks once the records are in hand. This piece is about the purchase.

What a parcel source actually is

Three separate products have to exist before a parcel row can be sold, licensed or downloaded, and they are made by different offices on different cycles.

The first is the parcel polygon, compiled by a county mapping, recorder or land information office from plats, deeds and survey control. The second is the assessment roll, maintained by an assessor or appraisal district for the purpose of levying tax, which carries the parcel identifier, a land use or property class code, assessed values, a land area and, in some jurisdictions, building attributes. The third is the join between them, which is usually the responsibility of whoever publishes the combined layer and is almost never documented as a step.

Every source described in this piece is a different answer to the question of who performs that join and who stands behind it. Texas states the arrangement plainly: the parcel files it publishes "are created by county appraisal districts or their third party vendor," and the state office "will attempt to refresh annually from each appraisal district or their third party vendor," with a refresh rate that "will vary across the state" (Texas Geographic Information Office, StratMap Land Parcels, read August 24, 2026). Two sentences on a public page, and they contain the entire lineage problem: the producing office may itself be a contractor, the state is an aggregator of counties, and the cadence is an intention rather than a schedule.

Utah goes further and carries the lineage in the data. Every record in the Land Information Records layer published through the state geospatial program has an assessor source field and a boundary source field, so the two offices behind the row are named on the row itself (Utah Geospatial Resource Center, Land Information Records parcel service, queried August 24, 2026). North Carolina takes a third approach and standardizes centrally: its Integrated Cadastral Data Exchange project "transformed source datasets from county data producers to create a standardized dataset with consistent attributes (fields)," then aggregated the standardized county datasets into one statewide file covering all 100 counties plus the lands of the Eastern Band of Cherokee Indians (NC Center for Geographic Information and Analysis, North Carolina Parcels metadata, read August 24, 2026).

These three are all public, all free and all structurally different. That is the first thing to understand about parcel procurement: the variance inside the free layer is larger than the variance between the free layer and the paid one.

Path one: county-direct, and the arithmetic it commits you to

Going to the source is the only path that gives a lending team full control of vintage, full sight of lineage and, in most cases, an unambiguous right to keep what it has taken. It is also the path whose cost is a headcount problem rather than a contract problem, and the size of that problem is a matter of public record.

The U.S. Census Bureau's 2025 Gazetteer file counts 3,144 county and county equivalent units across the 50 states and the District of Columbia, plus 78 municipios in Puerto Rico. The 2022 Census of Governments counts 3,031 county governments, 19,491 municipal governments and 16,214 township governments, for 38,736 general purpose local governments in total (U.S. Census Bureau, 2022 Census of Governments, Table 2, released 2023). Those two numbers are not in conflict; they count different things, and the gap between them is where a procurement plan goes wrong.

Concentration makes the problem worse rather than better. Texas alone holds 254 county equivalents, Georgia 159, Virginia 133 and Kentucky 120 (U.S. Census Bureau, 2025 Gazetteer). A ten state footprint drawn across the South is a several hundred office project before anyone has written a single request. And in New England the county is not the assessing unit at all: Maine has 16 county equivalents and 484 subcounty general purpose governments, Massachusetts has 5 and 351, Connecticut and Rhode Island have no county governments whatsoever (U.S. Census Bureau, 2022 Census of Governments, released 2023). A team that budgets a New England expansion in counties will be out by an order of magnitude.

The work itself is not uniform either. Some jurisdictions publish a parcel layer with a documented schema, a REST service and a set of bulk formats. Some publish a viewer and nothing else. Some publish nothing and answer records requests. Some answer records requests with a paper map. The aggregator category exists precisely because that distribution is unmanageable at national scale, and the honest way to evaluate the aggregator category is to first understand what the distribution looks like.

The openness reality a county-direct plan meets

The National States Geographic Information Council's 2025 Geospatial Maturity Assessment, published on August 14, 2026 from results collected in the second half of 2025, is the only national instrument that measures parcel availability jurisdiction by jurisdiction across all 50 states and the District of Columbia. Its cadastre theme reports that 73 percent of states, 36 and the District of Columbia, run statewide parcel map programs; that 82 percent, 41 states and the District of Columbia, have 100 percent coverage of digital parcel data; and that 43 percent, 21 states and the District of Columbia, have 100 percent publicly available parcel data by feature API or download.

Summing the survey's own jurisdiction columns turns those percentages into a work plan. The 51 state rows account for 3,699 parcel producing jurisdictions. Of those, 2,231 sit inside a statewide program, and 1,467 of the 2,231, or 65.8 percent, are publicly reachable. That leaves 764 jurisdictions inside program states publishing nothing (jurisdiction sums by MMCG from the 2025 assessment tables). The companion survey of parcel record availability across the country works that distribution state by state; what matters here is what it does to a procurement decision.

Grouped by what it actually takes to obtain them, the 3,699 jurisdictions fall into six populations. In 20 states and the District of Columbia every jurisdiction is publicly reachable through the program, which is 922 jurisdictions available on one contact each. In 10 more states the program is partial: 545 jurisdictions reachable and 272 withheld. In six states with a program, Georgia, Kansas, Nebraska, New Mexico, Oregon and South Dakota, not one of their 492 jurisdictions is recorded as publicly reachable. Among the 14 states with no program, seven put 90 percent or more of their 483 jurisdictions within reach by some route, and seven do not, covering 985 jurisdictions including Maine's 530 and Kentucky's 120.

The standard a program applies is where the aggregation argument turns. Of the 37 program rows, 20 report a standard that includes verification and quality control, nine a standard without verification, five a best effort standard and three that take county data as received (National States Geographic Information Council, 2025). Weighted by jurisdictions rather than by states, the states applying a verified standard hold 938 jurisdictions of which 693, or 73.9 percent, are publicly reachable. The states that take data as received hold 298 jurisdictions of which 139, or 46.6 percent, are reachable. Standardization and openness travel together, and both of them travel away from the places where a rural lender is trying to work.

Path two: aggregators, and what normalization is worth

An aggregator collects jurisdiction files, normalizes them into a single schema, and licenses the result as a national layer. The category includes firms such as Regrid, ATTOM and LightBox, and it is worth being precise about what each of them says it does, because the category's own documentation is more informative than any third party description of it.

On lineage, Regrid's parcel data onboarding documentation states that the company sources "directly from counties, states, municipalities and their designated vendors wherever possible," works "with trusted partners" and will "even digitize paper maps when needed" (Regrid, Parcel Data Onboarding FAQ, read August 24, 2026). That sentence describes the category honestly. A national parcel layer is not one pipeline; it is a portfolio of acquisition methods whose composition varies by county, and a buyer who needs to know which method produced a given county's records has to ask.

On schema, the same page states that the company is at "version 16 of our Regrid Parcel Schema," that it "standardizes the column names for over 100 county provided data columns, and over 40 Regrid provided data columns," and that "this schema is applied to 100% of our dataset." That is the aggregator's real product. The polygons exist without the aggregator; the single column name across three thousand jurisdictions does not.

On refresh, the same page states that the company targets "over 500 counties that are generally fast growing and populous for updated each quarter," that those counties "account for over 50% of the total parcels and population of the United States," that it refreshes "county data directly for 200 - 400 counties per month," and that "on average 94% of our parcels have been refreshed in the last 12 months." ATTOM's public page on parcel boundary data states that the company holds "more than 155 million parcel boundaries in the United States, representing roughly 99% of the population," and that its team "updates its parcel boundaries every six months" (ATTOM, Where To Find Parcel Boundary Data, published January 21, 2025). LightBox's parcel data page states "100% nationwide" parcel coverage, "300+ property attributes" and ownership information "updated daily" (LightBox, Parcel Data, read August 24, 2026).

Those are the vendors' own statements, reproduced as claims and not as findings. Read them together and a structural pattern appears that none of them is hiding. Refresh effort concentrates on fast growing, populous counties. Coverage is expressed against population. Both choices are commercially rational and both of them point away from the 764 jurisdictions that publish nothing, because those jurisdictions are disproportionately rural and low population. An aggregator can buy, digitize or request what a jurisdiction will release. It cannot publish what a jurisdiction withholds, and a coverage claim denominated in people is structurally incapable of telling a buyer which is which.

None of that makes the category a poor purchase. It makes the category a purchase whose value has to be located precisely. Normalization across a hundred column names, a single identifier convention, one delivery format and one support channel are real goods with real internal cost avoided. They are simply not the same good as coverage, and they are strongest in exactly the states where a county-direct plan would also have worked.

Path three: platforms, where the parcel layer is a means

The third path buys parcels as a component of something else. A mapping or analytics product carries a parcel layer because its users need to click a site, not because the parcel file is the deliverable. The commercial arrangement behind that layer is worth understanding, because it is frequently the aggregator path with an extra party in it.

Esri's own parcel data page states that its parcel content comes "from our partners at Regrid," and describes "over 158 million land parcel records that provide 100 percent parcel coverage across the United States and Canada" sourced "from all 3,143 US counties," delivered as tile layers in ArcGIS Online, ArcGIS Pro and ArcGIS Living Atlas (Esri, Parcel Data in GIS, read August 24, 2026). A buyer evaluating that layer is evaluating an aggregator's collection, an aggregator's schema and an aggregator's refresh policy, wrapped in a platform's delivery and terms. The lineage question does not disappear when the invoice comes from a different company; it simply moves one step further from the assessor. The distinction between the two product categories is worked through in the comparison of GIS platforms against CRE analytics platforms.

The platform path is the right answer when the parcel layer is not the point. A credit analyst who needs to see a parcel boundary under a flood zone, a traffic count and a demographic ring is not buying parcels; they are buying the join. That is the case for a map-first analytics workflow for a lending team, and it is a genuinely different purchase from a bulk parcel licence, because the deliverable is an answer rather than a file.

MMCG Analytics sits in this third category. It is a map-first commercial real estate analytics platform for lenders and investors, built by MMCG Invest, LLC, San Francisco, and it is built on federal, state and public data sources with source and vintage provenance carried on displayed values. Its analytical layers include parcels alongside flood, wetlands, wind risk, terrain, traffic and demographics, all sourced from public records. The platform provides data and analytics; the credit decision rests with the lender. Those are the only statements this piece makes about it, and they are made here because a category guide that omits its own author's category would be less useful, not more.

The evaluation method: ten tests before a parcel source enters a credit file

The three paths are not evaluated by the same questions in different words. They are evaluated by the same ten questions, and the interesting result is that the answers cluster by test rather than by path. The checklist below is written so that a credit team can run it against any source, public or commercial, and so that the result is reproducible by someone else on a later date.

One, the producing office. Does the source name the office that produced each record, or only the office that published it? A statewide file that names its counties is a different instrument from one that does not.

Two, geometry lineage. Does the source state where the boundary came from and what it is not? The correct answer usually contains a survey caveat. Wisconsin publishes one directly: "the boundaries depicted on this map do not represent the legal ownership boundaries of any property" and "this map is not a survey of the actual boundary of any property this map depicts" (Wisconsin State Cartographer's Office and Wisconsin Land Information Program, read August 24, 2026).

Three, vintage per jurisdiction. Not a single publication date for the file, but a date per producing office. Utah publishes a last update month per county. Texas puts the year and month in each county file name as YYYYMM. Most sources publish neither.

Four, refresh interval. An interval a buyer can plan against, not a promise of currency. Montana publishes cadastral data monthly for each county. New York states that "these standardized parcel datasets are updated once a year." North Carolina's metadata records its update frequency as "as needed," which is honest and unusable in the same breath.

Five, the attribute dictionary. Is there a published field list, and does it match the fields a credit screen needs? This test is worked in full in the next section, because it is the one most often assumed and least often checked.

Six, the jurisdiction denominator. Which jurisdictions are in the file, out of how many, and where is the list? New York answers precisely: 38 of 62 counties in the standardized polygon file, centroids for all 62. Texas answers imprecisely and openly: "not all counties are available for download."

Seven, the licence text. Is there published text that governs use, or only a disclaimer that governs liability? These are different documents doing different jobs and they are routinely confused.

Eight, machine access. Is there a documented API or feature service alongside a bulk download? A source that offers only a viewer has not published data, whatever its openness grade says.

Nine, redistribution and resale. Does anything address whether the recipient may pass the data on, embed it in a product, or sell an output derived from it?

Ten, retention after access ends. Does anything address what the recipient may keep and continue to use once the licence, the subscription or the relationship ends?

Running that checklist across nine public sources, six statewide programs and three counties, on August 24, 2026, produced 61 answers out of 90 possible. The scoring rule was strict and mechanical: a test counts as answered only where the source's own published page or open data catalogue record answers it, because that is the evidence a buyer holds before making contact.

The distribution is more instructive than the total. Every one of the nine names its producing office. Eight of nine carry a geometry caveat. Seven publish an attribute dictionary, seven state a jurisdiction denominator, seven publish licence or use constraint text and seven offer machine access beside a download. Then the floor drops. Two of nine address redistribution or resale. Two of nine address retention after access ends. The two tests that matter most once parcel data enters a lender's own derived work are the two tests the public layer answers least often, and that is an argument for reading contracts carefully on every path rather than an argument for any particular path.

The split by path is equally useful. County publishers answered every producing office, geometry lineage, vintage and denominator test, because a single county has no denominator problem and its catalogue record carries a modification date by construction. Statewide programs answered every attribute dictionary test and five of six refresh interval tests, because standardization is what a program is for. Neither group answered the last two tests with any consistency. The tests a source answers well are the tests its institutional form makes easy, which is precisely why a single source is rarely enough.

The attribute dictionary test, run on live services

Field counts are the most quoted and least useful number in parcel procurement. A layer with seventy fields and a layer with twenty five fields can be equally useless for a credit screen if the twenty five include the ones needed and the seventy do not. The test that matters is coverage of a named list.

Six public parcel services were read on August 24, 2026, four statewide and two county, taking the attribute field list the service itself reports and excluding object identifiers and geometry measures. New York's statewide public tax parcel service publishes 71 fields. North Carolina's statewide parcels layer publishes 67. King County, Washington's public parcel and address layer publishes 66. Montana's statewide cadastral framework publishes 41. Utah's Land Information Records layer publishes 27. Fulton County, Georgia's tax parcel service publishes 25.

Against those six, take the ten fields a commercial credit screen actually needs before it can size a site, classify its use and read its assessed value without going to a second source: parcel identifier, owner name, site address, land use or property class code, land area, building area, year built, assessed land value, assessed improvement or total value, and last sale date. Of 60 possible answers, 44 are present.

Four of the ten are near universal. All six services publish a parcel identifier, a site address, a land use or class code and a land area. After that the record thins out fast. Five publish an assessed land value and five an assessed improvement or total value. Four publish an owner name, and the two exceptions are instructive: Utah's Land Information Records schema carries no owner name field at all, and King County's public parcel and address layer publishes taxpayer mailing fields without an owner name field. Three publish a year built. Two publish a building area. One publishes a sale date.

That last line deserves its own sentence. The field most useful for comparable evidence is the field least often standardized in a public parcel layer, and the reason is legal rather than technical: sale price disclosure is a matter of state law, and several states restrict it. Any screen that assumes a sale date and price travel with the parcel record is a screen that will work in North Carolina and fail across most of the country. The same asymmetry runs through the wider census of building attributes recorded on assessor rolls, where building area and year built behave the same way.

The practical instruction is short. Write down the fields the screen needs before looking at any source. Score every candidate source, free or licensed, against that written list rather than against its own field count. A vendor that standardizes a hundred column names has done real work, and it has not conjured a sale date into a jurisdiction that does not publish one. Field standardization is a naming exercise; field availability is a legal and administrative fact about the jurisdiction, and no schema version changes it.

The licence test, run on the same sources

Free to download and free to use are different properties, and the gap between them is where derived work gets built on sand. The same nine public sources were read for what their licence text actually grants, scoring a term as present only where the published text states it and recording silence as silence.

Seven of nine publish a warranty disclaimer. Six publish a survey or legal use caveat. Two grant explicit permission to copy or distribute. Two request or require attribution. One addresses resale or commercial use. One applies a named open licence. One addresses derived works.

The two affirmative grants are worth reading closely because they are the exception. Utah's parcels page states that "there are no constraints or warranties with regard to the use of this dataset," and that "users are encouraged to attribute content to: State of Utah, SGID." That is a complete answer to tests seven, nine and ten in two sentences. King County, Washington's terms state that "to the extent copyright in said information is held by King County, you are hereby permitted by King County to copy, distribute, and otherwise use the information," while adding that "no one is permitted to sell this information except in accordance with a written agreement with King County," and that the data is provided on an "AS IS," "AS AVAILABLE," and "WITH ALL FAULTS" basis (King County GIS Center, terms, conditions and copyrights, read August 24, 2026). A lender building an internal screen is fine. A lender embedding the layer in something it sells has just found a clause it needs to route to counsel.

The sharpest finding in this section came from inside a single county's own portal. The Fulton County, Georgia open data portal carries 17 datasets with a parcel title, published by four different offices: 12 by the county's own GIS and five by cities inside the county. They do not carry the same licence. Ten of the county's twelve carry an as is disclaimer with no licence grant, including the current Tax Parcels layer, last modified August 6, 2026. The other two, the 2023 and 2024 tax parcel vintages, carry Creative Commons Attribution ShareAlike 4.0. Of the five city datasets, two carry Creative Commons Attribution 4.0 and three carry no licence field at all (Fulton County, Georgia GIS open data portal, read August 24, 2026).

A share alike obligation on a parcel vintage is not a formality. It is a condition on derivative works, and a lender that builds an internal collateral layer on the 2024 vintage and on the current vintage has taken on two different obligations from the same county office in the same portal. Nobody at the county did anything wrong; catalogue metadata accumulates over years and across staff. The lesson is procedural: the licence attaches to the dataset record, not to the publisher, and a source list that records publishers rather than dataset records has not captured the terms. That is the same discipline the provenance standard for analytics applies to figures, extended to rights.

Four questions belong in every parcel licence review, on all three paths. May the data be used internally without restriction. May derived work be created, and does any obligation attach to it. May outputs be shown to third parties, including borrowers and credit committees, and may they be sold. And what may be retained and used after the relationship ends. A source that answers all four in writing is rarer than the openness statistics suggest, and a source that answers none of them is not free, it is unpriced risk.

The denominator question

Every coverage claim in this category is a fraction, and the numerator is almost always the part being discussed. The denominator is where the claim gets its meaning, and there is no single agreed denominator for the United States.

Four published counts answer the question of how many local units there are. The Census Bureau's 2025 Gazetteer counts 3,144 county and county equivalent units in the 50 states and the District of Columbia. The 2022 Census of Governments counts 3,031 county governments, because Connecticut and Rhode Island have none and several counties are consolidated with cities. The 2025 Geospatial Maturity Assessment counts 3,699 parcel producing jurisdictions, because it counts municipalities in the states where municipalities assess. And the same Census of Governments counts 38,736 general purpose local governments, which is the number that matters for zoning and permitting even though it is far too large for parcels.

Inside a single state the divergence gets worse. Georgia has 159 county equivalents on the Gazetteer, 152 county governments in the Census of Governments, 159 parcel producing jurisdictions in the 2025 assessment, and zero jurisdictions recorded as publicly reachable. Maine has 16 county equivalents, 16 county governments, 530 parcel producing jurisdictions and 378 jurisdictions holding a digital parcel file. A coverage percentage computed on Maine counties and a coverage percentage computed on Maine assessing towns are different by a factor of thirty three, and both are defensible arithmetic.

This is why a claim expressed against population is not comparable with a claim expressed against counties, and why neither is comparable with a claim expressed against jurisdictions. When a vendor page states coverage across "all 3,143 US counties," that count sits one unit below the Gazetteer's 3,144 and a hundred and twelve above the Census of Governments' 3,031, and it says nothing at all about the towns of Maine or the 764 withheld jurisdictions. The fix is not to distrust the number. The fix is to convert every candidate source's claim into the buyer's own denominator, which is the list of jurisdictions in the lending footprint, and to ask for the covered subset of that list by name.

One more caution belongs here, and it cuts against the survey rather than the vendors. The 2025 assessment records Georgia at zero jurisdictions with public parcel access. Yet Fulton County, Georgia publishes a tax parcel layer through its own open data portal, with a REST service, CSV, shapefile, GeoJSON and KML distributions, last modified August 6, 2026. The survey is a state reported instrument on a two year cycle, and its authors say so: the report notes that some grades were reduced after "further evaluation found that these areas should have been classified as 'internal only'." A national survey is the right tool for a national picture and the wrong tool for a five county footprint. Verify at the jurisdiction level for the jurisdictions that matter.

When county-direct beats an aggregator, and when it cannot

The choice is not ideological and it is not permanent. Four conditions decide it, and all four can be checked before any commitment.

County-direct wins when the footprint is small and named. A lender working ten counties in three states has ten sources, not three thousand, and those ten sources hand it full lineage, full vintage control and an unambiguous licence position for a cost that is measured in analyst days once and maintenance hours thereafter. It wins again when the required attribute is rare. If the screen needs a sale date, a building area or a field that only the assessor publishes, the aggregator's schema cannot manufacture it, and the county's own extract may carry it.

County-direct wins a third time when timing is the binding constraint. Montana publishes cadastral data monthly for each county. Utah refreshes basic parcels for its five most populous counties monthly and its rural counties on a rotating quarterly to annual schedule. A team that pulls those files directly is on the publisher's clock. A team consuming a national layer is on the aggregator's clock, and the aggregator's own documentation describes a rolling schedule weighted to populous counties, which means the rural county in the footprint may sit at the back of the queue.

And county-direct wins when the licence position has to be clean. An internal derived product built on a source with no written grant is an exposure; an internal derived product built on Utah's stated absence of constraints is not. Public sources that publish an affirmative grant are scarce, but where they exist they are the strongest position available on any path.

County-direct cannot win in three situations, and no amount of effort changes them. It cannot win when the jurisdiction withholds: 764 jurisdictions inside program states publish nothing, and in six states not one jurisdiction is recorded as reachable. A records request may still succeed, and in some states it will meet a statutory restriction rather than a policy one. It cannot win when the footprint is national or unpredictable, because the marginal cost of the next jurisdiction never falls and the coordination cost rises. And it cannot win when the requirement is a single schema across many states within a fixed timeline, because building the crosswalk is the aggregator's product and rebuilding it internally is the definition of a build against buy decision that should be taken deliberately rather than by drift.

Hybrid patterns that hold up

Most working parcel stacks are mixtures, and three mixtures recur because they are structurally sound.

The first is a national layer for screening and county-direct for the file. A licensed national layer answers the first question in a site screen, which is usually whether the parcel exists, how large it is and what it is classed as. When a deal advances, the analyst pulls the producing jurisdiction's own record for the credit file, with the office named and the date recorded. The national layer carries the breadth; the county record carries the evidentiary weight. This is the pattern that sits behind the 30 minute pre-term-sheet site screen, where speed matters at the front and provenance matters at the back.

The second is county-direct for the core footprint and a national layer for the tail. A community lender with a defined market and occasional out of area participations maintains a handful of jurisdictions properly and buys the rest. The maintained set is the one where vintage discipline and attribute completeness pay for themselves, and the purchased set is the one where they would not.

The third is a platform for the join and county-direct for the exceptions. When the analytical question is a join of parcels to flood, terrain, traffic and demographics, the parcel layer is an input and the value is in the assembly. In that case the platform path is the efficient purchase, and the exceptions worth pulling directly are the jurisdictions where the platform's underlying source is thin. The same logic governs how zoning data is read by parcel and how buildable area is estimated from a parcel, where the parcel is the key and never the answer.

What does not hold up is a stack whose provenance changes silently. If the screening layer and the file layer disagree about a parcel's area or class, someone has to be able to say which source is which and why they differ. That capability is a documentation practice, not a purchasing decision, and it is the reason the source list in a credit file should name dataset records rather than companies.

The exit question and the public floor

The last test is the one most often skipped, and it is the one that determines whether a parcel purchase is an operating expense or a dependency. What happens to derived work when access ends.

A derived analytical product built on a licensed parcel layer usually contains three separable things: the raw records, the derived values computed from them, and the analysis presented to a committee. A contract may treat those three identically or differently, and silence is the worst outcome because it leaves the position to be argued later. The four questions from the licence section apply with particular force here: internal use, derivative works, third party display and post termination retention. Ask them in writing, on every path, and record the answers in the same place the figures are recorded.

The public floor is what makes that conversation negotiable. A lending team that knows what the free layer already carries is negotiating from information rather than from need. That floor is substantial. Twenty states and the District of Columbia publish every one of their 922 jurisdictions. Six statewide programs publish documented schemas, live services and bulk downloads at no licence cost. Two of the nine public sources examined here publish an affirmative permission to copy and distribute. The federal coordination layer, through the Federal Geographic Data Committee's cadastral subcommittee, works with public domain states on standardized statewide datasets. And the parcel adjacent layers a screen needs, structures, flood, wetlands, terrain and traffic, are federal and free in their own right. The full accounting of that floor sits in the audit of free federal data against paid subscriptions, and it belongs in any procurement file before a commercial conversation starts.

The floor also sets the honest test for a paid layer. A commercial parcel product should be evaluated on what it adds above the free layer in the buyer's own footprint: jurisdictions that are otherwise unreachable, a schema the team would otherwise build, a refresh cadence the team could not maintain, and a written licence position the public sources do not offer. Where a candidate adds all four, the case is easy. Where it adds one, the case is a conversation. Where it adds none, the free layer was the product all along. The wider vetting discipline, applied to any data category rather than to parcels alone, is set out in the twenty questions to ask any CRE data vendor, and the same tests apply when the subject is demographic vendors against Census direct data or the public CMBS tape against commercial services.

Frequently asked questions

Who are the main parcel data providers?

There are three categories rather than a ranking. Producing jurisdictions publish their own parcels, and 1,467 of the 2,231 jurisdictions inside statewide programs are publicly reachable by download or feature API (National States Geographic Information Council, 2025). Aggregators, a category that includes Regrid, ATTOM and LightBox, collect those jurisdictions and normalize them into one schema. Platforms embed a parcel layer inside an analytics or mapping product, frequently licensed from an aggregator: Esri's own parcel data page describes its content as coming "from our partners at Regrid" (read August 24, 2026). The right question is not which name is best but which category matches the footprint, the required fields and the licence position.

Can I get parcel data for free directly from counties?

Often, and the share depends heavily on the state. The 2025 Geospatial Maturity Assessment reports 21 states and the District of Columbia with 100 percent publicly available parcel data by feature API or download, while 764 jurisdictions inside states that run a parcel program publish nothing. Six states with programs, Georgia, Kansas, Nebraska, New Mexico, Oregon and South Dakota, record no publicly reachable jurisdiction at state level, although individual counties in those states may still publish directly. Verify jurisdiction by jurisdiction for the jurisdictions in the footprint rather than inferring from a state grade.

How current is parcel data from a national provider?

Currency is a distribution, not a date, and the vendors describe it that way in their own documentation. Regrid's onboarding page states that it targets over 500 fast growing and populous counties for quarterly updates, refreshes 200 to 400 counties per month, and that on average 94 percent of parcels have been refreshed in the last 12 months. ATTOM's public page states a six month boundary update cycle. Read those as statements about a distribution weighted to populous counties. If the footprint contains a rural county, ask for that county's own last refresh date rather than the national average.

What is the difference between a parcel aggregator and a GIS platform?

The aggregator sells the collection and the schema. The platform sells the delivery, the join and the workflow, and often licenses the collection from an aggregator. Esri's parcel data page states coverage of "over 158 million land parcel records" sourced "from all 3,143 US counties" and names its supplying partner. A buyer choosing the platform is still buying the aggregator's lineage, refresh policy and coverage, with an additional party in the contract chain, so the lineage questions apply to both.

What licence terms should a lender check before using parcel data?

Four, in writing, on every path: internal use without restriction, creation of derivative works and any obligation attaching to them, display or sale of outputs to third parties, and retention and continued use after the relationship ends. Public sources answer these less often than their openness suggests. Of nine public parcel sources read on August 24, 2026, seven published a warranty disclaimer but only two granted explicit permission to copy or distribute and only one addressed derived works. Licences also attach to dataset records rather than to publishers: one county open data portal carried four different licence treatments across 17 parcel datasets from four publishers.

How many counties are there for parcel data purposes?

There is no single number, which is why coverage claims need their denominator stated. The Census Bureau's 2025 Gazetteer counts 3,144 county and county equivalent units in the 50 states and the District of Columbia. The 2022 Census of Governments counts 3,031 county governments. The 2025 Geospatial Maturity Assessment counts 3,699 parcel producing jurisdictions, because it counts municipalities where municipalities assess: Maine alone contributes 530. Convert any vendor claim into the buyer's own list of jurisdictions before comparing it with anything.

Do I actually need national parcel coverage?

Rarely, and the question is worth asking before the first commercial conversation. National coverage is worth its cost when the footprint is unpredictable, when a single schema across many states is required on a fixed timeline, or when the footprint includes jurisdictions that do not publish. When the footprint is a named set of counties that publish directly, national coverage is buying breadth that will not be used, and the free layer plus a documented pull schedule delivers better lineage and better vintage control. Score the requirement first, then score the sources against it.

Sources

1. National States Geographic Information Council, 2025 Geospatial Maturity Assessment, Full Report, published August 14, 2026 from results collected in the second half of 2025; Cadastre (Parcels) theme, the state ranking tables for the 36 states and the District of Columbia with parcel programs and the 14 states without, the sub grade tables and the theme summary statistics. https://nsgic.org/wp-content/uploads/2026/08/2025-GMA-Full-Report-20260814.pdf
2. National States Geographic Information Council, Geospatial Maturity Assessment initiative page, listing the 2025 Full Report and the assessment history, read August 2026. https://nsgic.org/initiatives/geospatial-maturity-assessment/
3. U.S. Census Bureau, 2025 Gazetteer Files, county and county equivalent national file: 3,144 units across the 50 states and the District of Columbia and 78 municipios in Puerto Rico, read August 24, 2026. https://www2.census.gov/geo/docs/maps-data/data/gazetteer/2025_Gazetteer/2025_Gaz_counties_national.zip
4. U.S. Census Bureau, 2022 Census of Governments, Organization component, Table 2, Local Governments by Type and State, released 2023: 90,837 local governments, 38,736 general purpose, 3,031 county, 19,491 municipal and 16,214 township governments, with state detail. https://www.census.gov/data/tables/2022/econ/gus/2022-governments.html
5. U.S. Census Bureau, 2022 Census of Governments data file CG2200ORG02, local government counts by type and state, released 2023. https://www2.census.gov/programs-surveys/gus/tables/2022/cog2022_cg2200org02.zip
6. New York State GIS Program Office, Statewide Parcel Data: program created 2014, standardized tax parcel polygons for 38 of 62 counties updated once a year, statewide parcel centroids, the Standardized Tax Parcel Data Dictionary and the fitness for use statement, page read August 24, 2026. https://gis.ny.gov/parcels
7. New York State Information Technology Services, NYS Tax Parcels Public feature service, attribute field list read August 24, 2026 (71 published attribute fields). https://gisservices.its.ny.gov/arcgis/rest/services/NYS_Tax_Parcels_Public/FeatureServer
8. Utah Geospatial Resource Center, Utah Parcels, SGID cadastre category: 29 counties, per county last update months, the basic and Land Information Records attribute lists, the monthly to annual refresh statement and the use statement that there are no constraints or warranties, page read August 24, 2026. https://gis.utah.gov/products/sgid/cadastre/parcels/
9. Utah Geospatial Resource Center, Land Information Records parcel feature service, Salt Lake County, attribute field list including the assessor source and boundary source fields, queried August 24, 2026 (27 published attribute fields). https://services1.arcgis.com/99lidPhWCzftIe9K/arcgis/rest/services/Parcels_SaltLake_LIR/FeatureServer/0
10. Wisconsin State Cartographer's Office and Wisconsin Land Information Program, Statewide Parcel Map Initiative: version 12 with 3.56 million records collected from counties in the first half of 2026, version 13 tentatively scheduled for June 30, 2027, the schema documentation and the warranty and survey statements, page read August 24, 2026. https://www.sco.wisc.edu/parcels/data/
11. Montana State Library, Montana Spatial Data Infrastructure cadastral framework: monthly county cadastral downloads, Department of Revenue Orion computer assisted mass appraisal data, the schema and data dictionary documentation and the cadastral REST endpoint, page read August 24, 2026. https://msl.mt.gov/geoinfo/msdi/cadastral/
12. Montana State Library, Montana Cadastral Framework feature service, Montana Parcels layer, attribute field list queried August 24, 2026 (41 published attribute fields). https://services.arcgis.com/qnjIrwR8z5Izc0ij/arcgis/rest/services/Montana_Cadastral_Framework/FeatureServer/1
13. Texas Geographic Information Office, Texas Water Development Board, StratMap Land Parcels: the statement that files are created by county appraisal districts or their third party vendor, the annual refresh intention and its stated variability, the YYYYMM file naming convention, the standardized schema and the statement that the data is not survey grade, page read August 24, 2026. https://geographic.texas.gov/stratmap/land-parcels.html
14. NC Center for Geographic Information and Analysis, North Carolina Parcels metadata, NC OneMap: the Integrated Cadastral Data Exchange transformation, all 100 counties plus the lands of the Eastern Band of Cherokee Indians, the enumerated core cadastral attributes, the as needed update frequency and the responsibility statement, read August 24, 2026. https://www.nconemap.gov/datasets
15. NC Center for Geographic Information and Analysis, North Carolina Parcels (Polygons) map service, attribute field list read August 24, 2026 (67 published attribute fields). https://services.nconemap.gov/secure/rest/services/NC1Map_Parcels/MapServer/1
16. Fulton County, Georgia GIS, open data portal: 17 datasets with a parcel title from four publishers, carrying four distinct licence treatments, the Tax Parcels dataset last modified August 6, 2026 with REST service, CSV, shapefile, GeoJSON and KML distributions, and the as is disclaimer text, catalogue read August 24, 2026. https://gisdata.fultoncountyga.gov/
17. Fulton County, Georgia GIS, Tax Parcels feature service, attribute field list read August 24, 2026 (25 published attribute fields). https://services1.arcgis.com/AQDHTHDrZzfsFsB5/arcgis/rest/services/Tax_Parcels/FeatureServer/0
18. Maricopa County Enterprise GIS, open data catalogue, Parcels and Parcel Points datasets with the Assessor's disclaimer text including the statements that boundaries are for illustrative purposes only and that the data is not the equivalent of a title report or a real estate survey, catalogue read August 24, 2026. https://data-maricopa.opendata.arcgis.com/
19. King County GIS Center, Washington, open data catalogue and the Parcels with Address Property and Ownership Information (Public) feature service, attribute field list read August 24, 2026 (66 published attribute fields). https://gis-kingcounty.opendata.arcgis.com/
20. King County, Washington, GIS Center terms, conditions and copyrights: the permission to copy, distribute and otherwise use the information, the restriction on sale except under written agreement, and the as is, as available and with all faults disclaimer, page read August 24, 2026. https://kingcounty.gov/en/dept/kcit/data-information-services/gis-center/about/terms-conditions-copyrights
21. Regrid, Parcel Data Onboarding FAQ: the statements on sourcing directly from counties, states, municipalities and their designated vendors, on version 16 of the company's parcel schema standardizing over 100 county provided and over 40 company provided columns, and on the rolling refresh of 200 to 400 counties a month with 94 percent of parcels refreshed in the prior 12 months, read August 24, 2026 and reported as the company's own claims. https://support.regrid.com/parcel-data/parcel-data-faq
22. Regrid, Nationwide Land Parcel Data Licensing page: the stated 100 percent United States land parcel coverage and the statement that parcel data is continuously refreshed, read August 24, 2026 and reported as the company's own claims. https://regrid.com/nationwide-parcels
23. ATTOM, Where To Find Parcel Boundary Data, published January 21, 2025: the stated holding of more than 155 million parcel boundaries representing roughly 99 percent of the population and the stated six month boundary update cycle, read August 24, 2026 and reported as the company's own claims. https://www.attomdata.com/news/most-recent/where-to-find-parcel-boundary-data/
24. LightBox, LightBox Parcel Data page: the stated 100 percent nationwide parcel coverage, more than 300 property attributes and daily ownership updates, read August 24, 2026 and reported as the company's own claims. https://www.lightboxre.com/data/lightbox-parcel-data/
25. Esri, Parcel Data in GIS: the statement that parcel content comes from the company's partners at Regrid, the stated coverage of over 158 million land parcel records across all 3,143 US counties, and the delivery through ArcGIS Online, ArcGIS Pro and ArcGIS Living Atlas, read August 24, 2026 and reported as the company's own claims. https://www.esri.com/en-us/arcgis/products/arcgis-data/explore/parcel-data
26. Federal Geographic Data Committee, Cadastral Subcommittee: coordination of cadastral data among federal, state, tribal and local governments and standardization work with public domain states, page read August 2026. https://www.fgdc.gov/organization/working-groups-subcommittees/cadastral/index_html
27. MMCG Research, Parcel Procurement Series: jurisdiction sums and acquisition route groupings from the 2025 Geospatial Maturity Assessment state tables; jurisdictions and public reach weighted by the parcel standard each program applies; jurisdictions grouped by state cadastre grade; the ten test checklist scored across nine public parcel sources; published field counts and required field coverage across six live parcel services; and licence feature tabulations across the same nine sources and one county catalogue. Computed August 24, 2026 from sources 1, 3, 4, 5, 6 to 20; MMCG database, 2026. https://mmcganalytics.com/methodology/