Building an Attorney Database: Sourcing Bar, License, and Discipline Records Across 50 States

Data teams tend to treat lawyer licensing as a solved public-records job. Every state publishes a search page, the records are official, and the first scrape usually works.

The first real test of an attorney database tends to be a multi-state lawyer: current in one state, suspended in a second, registered under a former name in a third. The product keeps saying active for months, because nothing in the pipeline was built to notice.

If you need to check one lawyer, the licensing authority's own search is the right tool. This article is for teams building attorney data as a product: legal directories and lawyer-profile platforms, marketplaces and referral networks, legal intelligence vendors, background-screening firms and data resellers.

The population is more than 1.3 million resident active attorneys (American Bar Association, 2024 data), and none of them carries a national identifier. What keeps a dataset that size correct is five layers: a source registry, an identity model, status normalization, discipline linkage and change monitoring. We walk through why each layer exists, what breaks without it, and how to decide which ones to build yourself.

What an attorney database has to hold

Start with the unit. The record is the individual lawyer, not the firm. Headcount, practice mix and office footprints are roll-ups computed from lawyer records, and they deserve their own model; our guide to firm-level data covers that side.

The regulators already publish the minimum record. The State Bar of California's certificate of standing verifies "an attorney's name, bar number, admission date, current status, any name or status changes, and any public discipline, as of the date" it is issued. That last phrase is the one most schemas drop. A license status with no date attached is a guess about the present.

The population is large, and it moves. The ABA counted 1,322,649 resident active attorneys for 2024 in its National Lawyer Population Survey, and its 2025 Profile of the Legal Profession reported the first rise in the overall lawyer count since 2020. Every new admission is a new record, and every renewal cycle is a batch of status events.

Here is what one lawyer admitted in two jurisdictions looks like when the model holds up. The values are fictional; the fields map to what licensing authorities publish.

Illustrative attorney record (JSON, fictional values)

{
  "person_id": "atty_000142",
  "name": { "current": "Jane Q. Example", "prior": ["Jane Q. Sample"] },
  "admissions": [
    {
      "jurisdiction": "NY",
      "bar_number": "0000000",
      "admitted_on": "2011-01-19",
      "status_source_label": "CURRENTLY REGISTERED",
      "status_normalized": "authorized",
      "status_effective_from": "2025-03-02",
      "discipline": [],
      "source_url": "https://licensing-authority.example/ny/0000000",
      "retrieved_at": "2026-09-01T06:00:00Z"
    },
    {
      "jurisdiction": "TX",
      "bar_number": "00000000",
      "admitted_on": "2014-11-07",
      "status_source_label": "Inactive",
      "status_normalized": "inactive_voluntary",
      "status_effective_from": "2023-06-01",
      "discipline": [
        { "sanction": "suspension", "ordered_on": "2022-05-10", "stayed": true, "order_ref": "order-0001" }
      ],
      "source_url": "https://licensing-authority.example/tx/00000000",
      "retrieved_at": "2026-09-01T06:00:00Z"
    }
  ]
}

The bar number is a join key that only works inside one state

Every jurisdiction issues its own bar number. A lawyer admitted in New York, New Jersey and Texas holds three, and no national lawyer identifier ties them together. Healthcare has one for providers, which is why healthcare provider data looks similar in shape and behaves better in practice.

So the key is a pair: jurisdiction plus bar number. Connecting three admissions to one person is an entity matching job across multi-source datasets, using name history, admission dates and address signals. Name-only matching is where this goes wrong. Two lawyers with a common name in the same metro area merge, and one of them inherits the other's discipline.

Where do attorney records actually come from?

Knowing what the record holds, the next question is where each field comes from. Fewer places than people expect, and each one is authoritative for different fields.

The system of record for license status is each jurisdiction's licensing authority, which is not always the organization with "bar" in its name. In New York, attorney registration runs through the Office of Court Administration. The New York State Bar Association says plainly that it is not the official bar and is not responsible for licenses or registration, and certificates of good standing come from the Appellate Division where the lawyer was admitted.

Discipline travels a separate route. Disciplinary boards and courts issue orders, and the ABA National Lawyer Regulatory Data Bank, operating since 1968, collects public regulatory actions reported by every state and the District of Columbia. The public can request name searches in writing, for a fee, and bulk searches are offered.

Source	Authoritative for	How you get it	Freshness	Watch-out
State licensing authority search	Current status, bar number, admission date	Web search per jurisdiction; phone in a few	Live, usually without change history	Formats, labels and terms of use differ by state
Discipline orders (boards, courts)	Sanction type, dates, stays	Published orders, opinions, records requests	As issued	Often PDFs; private discipline never appears
ABA National Lawyer Regulatory Data Bank	Public regulatory actions across states	Written request, for a fee; bulk offered	Depends on agency reporting	Not a license-status feed
ABA National Lawyer Population Survey	Headcounts by jurisdiction	Annual publication	Annual; non-responders carried forward	Counts, not records
Certificate of good standing	Formal status on a date	Request per lawyer	Point in time	One lawyer per request
Derived datasets (directories, scraper output)	Coverage and convenience	License or marketplace purchase	As of their last pull	Inherit every gap above, plus their own lag

Access is uneven, and it shapes the design. The Department of Justice's Executive Office for Immigration Review lists telephone numbers, not search pages, for confirming licensure in New Hampshire and South Dakota on its verification page (dated January 2023). Illinois sits at the other end: as of September 2026, the Attorney Registration and Disciplinary Commission (ARDC) lawyer search includes a view of lawyers no longer authorized to practice because of recent registration, continuing-education, discipline or disability changes.

Then come the terms of use. The State Bar of Texas terms of use (accessed September 2026) prohibit capturing site data "through repetitive or automated means", with an exception for search engine crawlers. Terms like these decide which sources you automate, which you license and which you request by hand, so they belong in the source registry next to the URL. For the broader legal picture, see how scraping law treats public data.

This section describes sourcing constraints, not legal advice. Consult qualified counsel about how any site's terms apply to your use.

Why do multi-state attorney databases go stale?

Picking the right source for each field solves the first half of the job, and it is the easier half. We all know how this goes. The pipeline runs, the row counts look right, and the first sign of trouble is a customer who found a suspension the product did not show.

In practice, attorney data goes stale for five reasons. Each one is its own failure domain, and each maps to a layer later in this article.

1. Status labels do not translate. New York publishes nine registration statuses, from "Currently Registered" and "Delinquent" to two different kinds of resignation, one of them for disciplinary reasons. Texas has two membership statuses, active and inactive, and publishes eligibility to practice separately on an ineligible attorney list that it updates daily. Collapse all of that into one "active" flag and the dataset is wrong in two directions at once.

2. Discipline lands on different calendars. Under the ABA's model rule on reciprocal discipline, a lawyer disciplined in one jurisdiction must promptly inform disciplinary counsel in the others, and a second state may impose identical discipline, something different, or defer while a stay runs. Brian Faughnan, an ethics lawyer, describes a lawyer free of probation in Tennessee from August 2024 but still on probation for almost all of 2025 in Minnesota. And in June 2025 the Supreme Court of Texas held in Lane v. Commission for Lawyer Discipline that Texas's four-year limitations period applies to reciprocal discipline and runs from the original misconduct, which ended a Texas case that followed an Illinois suspension. A sanction in one state does not guarantee a matching record in the next.

3. Even the national count is partly carried forward. For 2024, the ABA's population survey reused the prior year's figure for 10 jurisdictions that sent no current data, among them Arkansas, Delaware, Maryland, Minnesota, South Carolina and Vermont.

4. Complaints are not discipline. In fiscal year 2025 the State Bar of California's Office of Chief Trial Counsel opened more than 21,000 cases and obtained discipline for 154 attorneys, including 69 disbarments, down from 229 disciplined attorneys the year before, according to its fiscal year 2025 Annual Discipline Report. The report counts its reprovals as "both public and private", and private ones never reach a public record. A product that treats complaint volume as a risk score is measuring something else.

5. Collectors decay. Search pages get redesigned, fields move, labels get renamed. A scraper that worked in March can return empty rows in June, and unless something compares the output against the last verified pull, nobody notices for weeks. The fix starts with observability for third-party datasets.

The five layers of an attorney database that stays correct

There are five layers every attorney data team ends up building, whether they plan to or not. Each one below comes with a regulator or court already doing a version of it, which is the best evidence we have that the layer is necessary.

1. Source registry

One system of record per jurisdiction, with its access method and terms attached. New York is why this has to be explicit: registration sits with the Office of Court Administration and good standing with the Appellate Division, not with the association most people search first. The registry breaks quietly when a jurisdiction moves its portal, so give it an owner and treat it as data.

2. Identity

Store a person, then attach admissions to that person, keyed on jurisdiction plus bar number. In Texas, state law requires its chief disciplinary counsel to "establish a process to regularly search the National Lawyer Regulatory Data Bank" to catch Texas lawyers disciplined elsewhere. If a regulator has to poll a national repository to connect one lawyer across states, a product cannot assume the connection exists. Name-only merges are the failure to design against.

3. Status normalization

Keep three values for every status: the verbatim source label, a normalized value, and the date it took effect. The normalized value is what the product filters on. The verbatim label is what you show a customer who asks why. The date answers the question that matters in disputes: was this lawyer authorized on the day in question?

Source label (jurisdiction)	Normalized value	Why it needs its own value
Currently Registered (NY)	authorized	Registration filed and current
Due to Re-register Within 30 Days of Birthday (NY)	authorized_renewal_due	Still registered; recheck next cycle
Delinquent (NY)	not_authorized_administrative	Missed filing; may be referred for discipline
Suspended (NY)	suspended	Reason comes from the order, not the label
Disbarred (NY)	disbarred	Struck from the roll of attorneys
Resigned from Bar, Disciplinary Reason (NY)	resigned_disciplinary	Pending discipline at resignation
Resigned (NY)	resigned_voluntary	Non-disciplinary
Deceased / Incapacitated (NY)	deceased / incapacitated + presumed flag	Can be presumed from age
Active (TX)	authorized unless on the ineligible list	Membership and eligibility live on different pages
Inactive (TX)	inactive_voluntary	Not practicing in Texas
On the ineligible attorney list (TX)	not_authorized + reason	Updated daily
Not currently authorized, recent status change (IL)	not_authorized + reason	Registration, continuing education, discipline or disability

Source labels: New York State Unified Court System, Description of Registration Statuses for NYS Attorneys; State Bar of Texas; Illinois ARDC. The normalized values are our proposed model, not any regulator's taxonomy.

We see the same pattern in healthcare provider data, where state boards describe one license state in different words. The fix is the same too: never throw the source label away.

4. Discipline linkage

Link every order to the admission it affects, and record the sanction, its dates and whether it is stayed. The ABA's model rule defers reciprocal discipline while the original sanction is stayed, so a stayed suspension stored as a live one misstates whether the lawyer can practice. Orders often arrive as PDFs or court opinions, which makes this document extraction as much as web collection; the mechanics overlap with legal document processing. For actions taken in other states, the Data Bank is the backstop, which is exactly how Texas uses it.

5. Change monitoring and provenance

Diff, don't re-scrape. Where a source publishes changes, use them: Illinois exposes recent status changes as a search, and Texas refreshes its ineligible list daily. Most jurisdictions publish no delta at all, so in practice you build the diff yourself, comparing each pull with the last verified one and reviewing changed records first. Every value carries its source URL and retrieval time, which turns "is this right?" into a question with an answer. The mechanics match any website change monitoring program.

The layers form a loop. What monitoring finds feeds the registry and the identity model, and that loop, more than any single pull, is the product.

Build it, buy it, or hand it off

With the five layers defined, the practical question is which of them your team should own. The coverage math explains most of that decision. Five jurisdictions (New York, California, Texas, Florida and Illinois) account for 604,057 of the 1,322,649 resident active attorneys the ABA counted for 2024, about 46% by our arithmetic. Covering those five well is within reach of an in-house team. The long tail, with its phone lines, carried-forward counts and one-off formats, is where approaches separate.

Marketplace scrapers lower the cost of a first pull. On Apify's marketplace, one state bar actor ships California only and treats every state as a separate adapter; a multi-state actor covers 14 states and notes its Minnesota records skew toward admissions from about 2009 onward (both accessed September 2026). Neither owns what happens after the pull.

Approach	Fits when	Breaks when	Who owns monitoring
In-house pulls from licensing authorities	One or two jurisdictions, or a one-time study	Coverage outgrows one engineer's maintenance time	You
Marketplace scraper actors	Prototyping, or a state the actor covers	You need uncovered states or status history	You
Licensed directory or legal data provider	A lag of weeks is acceptable	The product promises current status	The provider, on its schedule
Background-screening data	Point-in-time checks on individuals	You need a refreshed feed for a whole population	The screening firm, per request
Managed multi-state pipeline	Every jurisdiction has to stay current in the product	You need one or two states	The partner, to an agreed cadence

If you are shortlisting ready-made datasets instead, our comparison of legal data providers sorts them by vendor class. When two options look even, pick the one that owns change monitoring. It is the layer that fails without an error message, and every other layer depends on it.

Background-screening firms carry an extra consideration. The FTC's guidance for employment screening companies says reports used in hiring decisions can be consumer reports under the FCRA, and that screening companies must follow reasonable procedures to assure "maximum possible accuracy". An undated license status makes that standard harder to show. This is general information, not legal advice. Consult qualified counsel about your obligations.

Where Forage AI fits is narrower than a pitch would suggest, and we would rather say so up front. If a product needs one or two states, build it. If a lag is fine, license a derived dataset. Forage AI earns its place when every jurisdiction has to stay current: managed multi-state sourcing, entity resolution across admissions, and change monitoring run as a service, with selector drift, anti-bot evolution and schema changes handled as part of it. The closest proof shape is licensed-professional data in healthcare, with 99.7% field-level accuracy, client-reported, on healthcare provider data. Every delivery passes a 3x QA team, the client owns the data and Forage AI never resells it, handling and retention rules are agreed per engagement, and the first dataset typically lands in 1-2 weeks.

Keeping an attorney database honest after launch

An attorney database is never finished, and that is the part worth planning for. New admissions arrive every year, statuses change on dates nobody announces, and a sanction in one state takes its own route to the next.

The habit that keeps it true is small. Watch the deltas where regulators publish them, audit a sample of changed records every cycle, and keep the "as of" date visible to your own customers. If you are mapping status labels across jurisdictions right now, we would like to compare notes on the edge cases, especially the ones that fit no normalized value yet.

Frequently asked questions

Is there a national database of all licensed attorneys in the US?

Not a live one. Licensing runs jurisdiction by jurisdiction, so current status has to come from each licensing authority. The ABA's Data Bank covers public regulatory actions rather than status, and its population survey is an annual count, not a register.

Does a lawyer have one bar number or a different one in each state?

One per jurisdiction of admission. A lawyer admitted in three states holds three numbers, and no national identifier links them. That is why an attorney bar number lookup only identifies a lawyer inside the state that issued the number.

Can you scrape state bar directories to build an attorney database?

Sometimes, but the terms matter and so does everything after the first pull. The State Bar of Texas terms of use, for example, prohibit capturing data through repetitive or automated means. Where automation is permitted, a scrape is the start of maintenance, because labels, layouts and statuses keep changing. This is general information, not legal advice.

How often should attorney license data be refreshed?

As often as the product's promise requires, and never less often than the sources change. Status can change on any day, which is why Texas updates its ineligible attorney list daily. Use published deltas where they exist, diff everything else against the last verified pull, and show the retrieval date next to every status.

Is the ABA's lawyer discipline Data Bank public?

Yes, with limits. The ABA runs name searches for the public on written request, for a fee, and offers bulk searches. It covers public regulatory actions only, so private discipline and current status sit outside it. At least one consumer guide ranking for these searches wrongly says it is not publicly accessible.

How do I look up one attorney's license?

Use the licensing authority's search for the state where the lawyer says they are admitted, and match the name, bar number, admission date and status. For a formal record, request a certificate of good standing. That approach works well for a single attorney license lookup and does not scale to a product.