Industries / Government, education and community / Lobbying
Firms
Firms are the professional lobbyists: consultancies and law firms hired by clients, and the in-house government-affairs staff of companies and groups.
6
Milestones achieved, from 1875 to 2025.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1875
The King of the Lobby
Sam Ward, called the "King of the Lobby", was the best-known hired lobbyist after the US Civil War; first hired by Treasury Secretary Hugh McCulloch, he worked members of Congress over fine dinners for many clients. In 1875 he was called before a congressional investigation of subsidies to the Pacific Mail Steamship Company, after newspapers charged that he had spent more than $100,000 on bribes.
LessonAccess sold through hospitality produced the scandals that led to calls for registration.
Credit
Sam Ward; history by Senator Robert C. Byrd
1976
Earmark lobbying becomes a business
Tufts University hired the new firm Schlossberg-Cassidy & Associates in 1976 to win money written directly into federal spending bills, known as earmarks; Tufts' own history says it received $39 million in special appropriations or authorisations in 1977. By 1987 the firm took credit for $165 million of buildings for university clients, The Scientist reported.
LessonMoney that bypassed peer review drew a response from universities themselves, whose presidents in the Association of American Universities agreed to stop seeking earmarked building funds.
Credit
Gerald Cassidy and Kenneth Schlossberg; Tufts president Jean Mayer
1994
UK firms regulate themselves
After the cash-for-questions affair, in which MPs were paid to ask questions in Parliament, five UK consultancies founded the Association of Professional Political Consultants with a code of conduct, a public register of clients and a ban on paying politicians. A Commons committee found in 2009 that three bodies ran differing codes, one of which did not require members to name clients, and that most public affairs staff worked in-house for single organisations rather than for firms with many clients.
LessonVoluntary codes without penalties did not win the committee's trust, and Parliament passed a statutory register in 2014.
Credit
the founding consultancies; the House of Commons Public Administration Committee
2007
Registered lobbyists peak
The number of lobbyists named on US federal filings peaked at 14,844 in 2007 and fell to 11,171 in 2016, while reported spending rose from $2.87 billion to $3.16 billion, by OpenSecrets' count. Researchers Timothy LaPira and Herschel Thomas estimated that 52% of people paid to influence federal policy in 2012 did not report lobbying, partly because people who spend less than 20% of their time lobbying for a client need not register.
LessonTighter rules for registered lobbyists can move work into unregistered "strategic advice"; the causes of the fall are still disputed.
Credit
OpenSecrets; Timothy LaPira and Herschel Thomas
2025
In-house lobbying outnumbers firms
Of the 16,474 organisations on the EU Transparency Register at the end of 2025, 714 were professional consultancies, 111 law firms and 193 self-employed consultants, against 4,922 companies, 2,911 trade and business associations and 4,444 non-governmental organisations, by the register's own count.
LessonHired firms are a small share of registered lobbying; most of it is done by organisations for themselves.
Credit
Joint Transparency Register Secretariat
2025
Record US lobbying spending
Reported US federal lobbying reached $5.13 billion in 2025, up 15.4% on 2024 and the highest since OpenSecrets' series began in 1998, with 16,897 clients and 13,709 named lobbyists, still fewer than in 2007. The firm with the most lobbying income was Ballard Partners, with $88.1 million from 345 clients.
Credit
OpenSecrets, from filings with the House and Senate
Watch for
Agents that triage bills
Quorum, a software company for government-affairs teams, launched AI bill tracking that it says "reads every bill in every state" in August 2025, and in July 2026 a Bill Tracker Agent that recommends "a stance, priority level, and issue classification" for each bill.
Achieved whenAn independent party publishes accuracy or adoption figures for these tools.
Claimed by
Quorum, August 2025 and July 2026
AI in everyday lobbying work
The Public Affairs Council, a US association of public affairs professionals, reported in September 2025 that 89% of teams are "exploring using AI", restating a survey by FiscalNote, a company that sells AI tools to the same teams, which found 57% of about 1,000 professionals open to or using AI against 14% a year earlier.
Achieved whenA survey by a body that does not sell AI tools, or by a regulator, measures how lobbyists use AI.
Claimed by
Public Affairs Council and FiscalNote, 2025
Open
01
Test the bill-tracking agents
An independent test of tools that sort bills for lobbying teams: given a public description of an organisation, does the tool find the state bills that matter to it and classify their topic correctly. Tools tested include vendors' agents that agree to take part and general-purpose models.
The briefCredit · unclaimed02
Count the unregistered advocates
A repeatable yearly estimate of how many people are paid to influence US federal policy without registering as lobbyists, published as counts and shares with error ranges and never as lists of names.
The briefCredit · unclaimed
Open 01 · the brief
Test the bill-tracking agents
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An independent test of tools that sort bills for lobbying teams: given a public description of an organisation, does the tool find the state bills that matter to it and classify their topic correctly. Tools tested include vendors' agents that agree to take part and general-purpose models.
Why it is still open
The vendors' accuracy claims are their own. The one public benchmark found, the "corporate lobbying" task in LegalBench, built by John Nay, has 500 hand-labelled examples that pair a federal bill's title and summary with a company's description; it covers neither state bills, nor full bill text, nor topic classification.
Built with what exists
State bill text from the Open States and LegiScan APIs; public descriptions of ten varied organisations, such as a hospital group, a union, a city and an environmental charity; general-purpose models and any commercial agents whose makers agree to be tested; a model that scores many yes-or-no relevance questions in one pass (see the Note "A thousand questions, one pass"); a spreadsheet for two reviewers.
First step
Label 50 bills from one state's 2026 session for two of the organisations and run one model against them.
Done looks like
500 bills from 2026 sessions labelled for relevance and topic for each of the ten organisations, with the two reviewers agreeing on at least 85% of labels; precision and recall published for every tool; labels and code released.
Open 02 · the brief
Count the unregistered advocates
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A repeatable yearly estimate of how many people are paid to influence US federal policy without registering as lobbyists, published as counts and shares with error ranges and never as lists of names.
Why it is still open
The best-known estimate, LaPira and Thomas's figure of 52%, rests on a sample of people from 2012, and its authors stress that it relies on broad assumptions; CRS calls estimates of unregistered lobbyists "inexact".
Built with what exists
The LDA.gov API for who is registered; public staff pages of Washington firms, associations and companies; a web agent to collect those pages (see the Note "Deleting 80% to start over"); an AI model to classify job descriptions as policy advocacy or not; a spreadsheet for two reviewers.
First step
Collect the public staff pages of 50 Washington firms and associations and classify 100 people by hand to set the rules.
Done looks like
A published method and a 2025 estimate with a 95% confidence interval; two reviewers agree on at least 85% of 300 sampled classifications; the method rerun on 2026 data.