BRENNEN

Industries / Government, education and community / Lobbying

Grassroots campaigns

Grassroots campaigns get members of the public to contact lawmakers or regulators, through petitions, letters, calls, e-mails and comments on proposed rules.

7

Milestones achieved, from 1788 to 2023.

3

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1788

    Mass petitions reach Parliament

    Campaigners against the slave trade, and the trade's defenders, petitioned both Houses of the British Parliament "on a scale not seen before", in the words of Parliament's own history; Thomas Clarkson toured Britain to organise support, and the London Committee rented a house in Downing Street to be close to MPs. The trade was abolished by an act of 1807, which passed the Commons by 283 votes to 16 after repeated failures.

    LessonPetitions gave people who could not vote a way to be counted, but winning took about twenty years of repeated campaigns.

    parliament.ukparliament.uk

    Credit

    Thomas Clarkson; the London Committee; William Wilberforce

  2. 1935

    Telegrams traced to a lobbyist

    When letters and telegrams flooded Congress in 1935 against a bill to break up public utility holding companies, Senator Hugo Black investigated, and Western Union officials testified that the 816 telegrams one representative received had all been dictated and paid for by a lobbyist for the Associated Gas and Electric Company. Congress then required utility company agents to register, and added registration duties to the Merchant Marine Act of 1936 and the Foreign Agents Registration Act of 1938.

    LessonCampaigns that hide who is behind them invite disclosure rules.

    senate.gov

    Credit

    Senator Hugo Black

  3. 2003

    Commenting on rules moves online

    The US government launched Regulations.gov in January 2003, which GAO describes as the first website to let the public both find proposed federal rules and comment on them across government.

    LessonVolume became the problem: one 2017 proceeding drew more than 22 million comments.

    gao.gov

    Credit

    Office of Management and Budget, with the Environmental Protection Agency as lead agency

  4. 2012

    Websites go dark against bills

    English Wikipedia blacked out for 24 hours on 18 January 2012 to protest two US anti-piracy bills, SOPA in the House and PIPA in the Senate, after more than 1,800 editors took part in the decision; five days later the Senate withdrew the motion to begin debate on PIPA.

    LessonLarge websites, not only interest groups, can call the public to contact lawmakers.

    diff.wikimedia.orgcongress.gov

    Credit

    the Wikipedia community; the Wikimedia Foundation

  5. 2013

    First successful citizens' initiative

    The European Citizens' Initiative, open since April 2012, lets one million EU citizens from at least seven member states ask the European Commission to propose a law. Right2Water, submitted on 20 December 2013 with more than 1.6 million supporters, was the first to meet the requirements, by the Commission's account; the Commission answered in 2014 with a public consultation rather than a new law.

    LessonA successful initiative obliges the Commission to answer, not to legislate.

    eur-lex.europa.eucitizens-initiative.europa.eu

    Credit

    European Commission; the Right2Water organisers

  6. 2021

    Fake comments counted

    The New York Attorney General found that nearly 18 million of the more than 22 million comments in the US Federal Communications Commission's 2017 net neutrality proceeding were fake, on both sides: more than 8.5 million impersonated real people in a campaign funded by broadband companies through lead-generation firms, and 9.3 million supporting the other side used made-up identities, most of them from one 19-year-old student. Three lead-generation firms agreed to pay more than $4.4 million, and the Administrative Conference of the United States (ACUS), which advises federal agencies on procedure, recommended that agencies remove duplicates, flag computer-generated comments and let people report comments filed in their name.

    LessonPaying firms by the comment produced made-up comments, and fake grassroots campaigns, called astroturf, can come from any side.

    ag.ny.govacus.gov

    Credit

    New York Attorney General Letitia James's office; ACUS

  7. 2023

    AI letters draw replies

    Sarah Kreps and Douglas Kriner sent 32,398 e-mails, some written by people and some by the AI model GPT-3, to 7,132 US state legislators; legislators answered the AI-written ones 2 percentage points less often, a difference the authors call statistically significant but substantively small.

    doi.orgapi.openalex.org

    Credit

    Sarah Kreps and Douglas Kriner, New Media & Society, March 2023

Watch for

  1. AI-generated comments on air rules

    More than 20,000 public comments against proposed rules on gas appliances at Southern California's South Coast Air Quality Management District were generated through CiviClick, a platform that calls itself AI-powered, according to public records reviewed by the Los Angeles Times; district staff contacted a small sample of the named senders, and three of the five who replied said they knew nothing of the messages. The board dropped the rules in June 2025, and in March 2026 22 California officials asked the state Attorney General and four district attorneys to investigate.

    Achieved whenAn investigator, regulator or court confirms how many comments were written by AI or sent without the named person's consent.

    govtech.comcleantechnica.com

    Claimed by

    Los Angeles Times, February 2026

  2. Checks on human commenters

    H.R. 9643, the Comment Integrity and Management Act, would require US agencies to verify, as far as possible, "whether the comment has been submitted by a human being", to identify mass comment campaigns and to publish policies on computer-generated comments. It was introduced on 13 July 2026 and approved by the House Oversight Committee 24 to 19 later that month; a 2024 version passed the House but was not enacted.

    Achieved whenThe bill becomes law and agencies publish their policies and label mass comments.

    govinfo.govlegis1.com

    Claimed by

    Representative Clay Higgins, July 2026

  3. AI reads consultation replies

    The UK Department for Transport's Consultation Analysis Tool, built with the Alan Turing Institute, has analysed 200,000 replies to public consultations, scored 0.75 against human coding in a blind test and 0.93 in live pilots on a standard accuracy measure (the F score), and saves "around 50-70%" of cost and time, by the department's own evaluation.

    Achieved whenAn independent body audits the tool's accuracy, or its routine use is documented outside the department.

    gov.ukassets.publishing.service.gov.uk

    Claimed by

    Department for Transport and the Alan Turing Institute, December 2025

Open

  1. 01

    Test set for AI comments

    A public, labelled collection of comments on proposed rules, some written by people before AI writing tools were common and some written or edited by AI models, used to score the tools that claim to detect AI text, above all on how often they wrongly flag a person's comment. Nothing is submitted to a live proceeding.

    The brief
    Credit · unclaimed
  2. 02

    Count mass comment campaigns

    A public tally, for every US federal proceeding that drew 1,000 or more comments in 2026, of how many comments were identical, how many were near-copies of a form letter, how many were unique, and how large each campaign was.

    The brief
    Credit · unclaimed

Open 01 · the brief

Test set for AI comments

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public, labelled collection of comments on proposed rules, some written by people before AI writing tools were common and some written or edited by AI models, used to score the tools that claim to detect AI text, above all on how often they wrongly flag a person's comment. Nothing is submitted to a live proceeding.

Why it is still open

In 2019 Max Weiss showed that people asked to sort AI-written comments from real ones did no better than chance. A February 2026 check of 144 comments in two Federal Communications Commission proceedings by the Technology Policy Institute relied on one commercial detector and warned that it "may miss some AI-generated text" and may flag text merely edited with AI; it named no shared test set, and this research found none.

techscience.orgtechpolicyinstitute.org

Built with what exists

Comments from 2015 to 2021 downloaded through the Regulations.gov API; several AI models to write new comments and to edit real ones; commercial and open-source AI-text detectors; a spreadsheet for two reviewers.

  1. First step

    Download 500 comments from five proceedings of 2018 and produce 500 AI-written counterparts on the same topics.

  2. Done looks like

    5,000 human-written and 5,000 AI-written or AI-edited comments across 20 proceedings; every detector scored on false alarms and misses, with false alarms on human comments reported separately; data and scores published.

Open 02 · the brief

Count mass comment campaigns

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public tally, for every US federal proceeding that drew 1,000 or more comments in 2026, of how many comments were identical, how many were near-copies of a form letter, how many were unique, and how large each campaign was.

Why it is still open

GAO found in 2019 that agencies are not required to verify commenters' identity and that their handling of duplicate comments varies; ACUS asked the General Services Administration in 2021 to provide a common tool to remove duplicates, and the 2026 bill above would require agencies to report campaign sizes. An open-source tool built for the US Department of Agriculture groups similar comments, but one proceeding at a time; this research found no tally across all proceedings.

gao.govacus.govgithub.com

Built with what exists

The Regulations.gov API; near-duplicate detection such as MinHash in the open-source datasketch library; an AI model to group paraphrased form letters; the public-comment-analysis code above as a starting point; a spreadsheet for two reviewers.

  1. First step

    Download the comments for ten proceedings that closed in 2026 and test the grouping on them.

  2. Done looks like

    Every proceeding with 1,000 or more comments that closed in 2026 counted; two reviewers confirm at least 95% of groupings in a sample of 300; counts per proceeding published and updated monthly.