How Many Comments is a Lot? The Rough Distribution of Comments Published on regulations.gov

This post was co-authored with: Taunton Paine

November 09, 2026

How Many Comments is a Lot? The Rough Distribution of Comments Published on regulations.gov

Public participation in notice and comment on federal rulemaking rarely reaches six figures. When it does happen, and comments submitted to regulations.gov cross the 100K threshold, it can generate a lot of press about the proposed regulatory action. Two recent dockets of rules crossed that mark: the Office of Management and Budget revisions to the Uniform Guidance in 2 CFR 200, which received extensive coverage regarding its high comment count, and the Census Bureau proposal on residency criteria, which drew significant attention and organized response.

What was interesting about these proposed rulemakings is that they were both related to information policy and previous iterations of those regulations were mundane and garnered little attention with few comments. In this Administration, however, proposed changes to OMB’s proposed regulation on how grants for research would be administered in the future — including codifying the primacy of political influence over peer review; Census’ proposed rule on the collection methods for obtaining and using a count of the population; and removing race and ethnicity from decennial and excluding undocumented immigrants and some other non-citizen from apportionment — were dramatic departures from what would typically have been pretty lackluster regulations. The excess of comments received on those notices of proposed rulemaking felt like a major deviation from past precedent, where regulatory dockets on scientific and statistical matters typically draw far smaller, mostly technical, responses.

This got us to thinking: was this a true deviation or has there been a change in the distribution of comments submitted to regulatory dockets over time? Both of us previously served in federal positions where we had to read, consider, and respond to public comments. While high topline counts can drive attention to a proposed rulemaking and can provide a social barometer on overall support or opposition to a rule, those figures may mask the relatively weak signal of the few comments in such dockets that contain substantive information that may be useful to policymakers in improving a rule. Furthermore, the Administrative Procedures Act contains no requirement for agencies to consider the volume of comments received, focusing more on the substance of the unique arguments presented. Nevertheless, it may still be useful to policymakers and observers alike to understand when the number of comments received is truly unusual. To help make these figures broadly accessible, we’re introducing a dashboard to examine the distribution (Abigail Haddad also has a similar tool). We also hope this will provide timely context for those considering responding to the recent RFI by the Administrative Conference of the United States (ACUS) on how agencies process public comments on rulemaking.

In short, we find that the distribution of comments is heavy-tailed and roughly power-law, with the tail growing heavier over time, though a strict power law fails in recent years and a lognormal tail fits about as well. This suggests that the comments are driven by a small number of dockets that attract very large, mobilized responses and that phenomenon is occurring more frequently over time. This is important because it means that dockets with thousands of comments, once a rare occurrence, are becoming more common and that will shape how agencies need to process them. We also conclude with some simple observations on how transparency and utility of data on rulemaking dockets could be improved to support further analyses.

The Docket Distribution

To examine the historical and contemporary distribution of the number of published comments in dockets on regulations.gov we need an efficient source of data. There would be a lot of overhead to pull data directly from regulations.gov via their API to track this over time. Thankfully, a couple of other projects already do that and package the data in a convenient format for analysis. We use a combination of spicy-regs aggregated data and mirrulations mirrored data from regulations.gov. Both projects are free and open-source and do not require an API. In addition to the analysis below, we use the data — along with some help from Claude and automations from github routines — to provide an interactive dashboard of the comment dockets to date. The data on our dashboard may be slightly different than reported in the analysis (we ran the analysis on 11/6/2026 and published the dashboard on 11/9/2026 so any comments or new dockets added between that time will not be reflected in the analysis and any difference should be statistically ignorable).

At the time of our analysis, the Spicy Regs dataset covers 62,432 dockets identified as rulemaking by agencies (subtracting about 900 dockets that appeared to be only Information Collection Requests under the Paperwork Reduction Act or collections of bundled Federal Register notices) and 19.77 million comments recorded for docket dates as old as 1973 through early November 2026 (see below, however, for a note of caution on that long window). These figures represent published comments and not necessarily the total number submitted - agencies redact comments for many reasons and, until recently, the submitted comment topline counter could be gamed to an arbitrary value.

It’s also worth noting that while regulations.gov has been adopted by over 200 US agencies, 129 do not participate (although they may post other documents on regulations.gov), and a few have developed their own docket and comment systems, such as the NIH, FCC, SEC, the Federal Reserve Board, and FDIC, among others. Of these, FCC is a notable outlier for regularly receiving many public comments, in particular the net neutrality proposals which received more than 20 million public comments (which were largely the result of a bot-farm scheme). The public comments maintained by these other systems are not typically cross-posted to regulations.gov, so while regulations.gov is not comprehensive, the range of agencies that do participate makes it the best available cross-section of public comment activity on federal rulemaking–at least until future work consolidates these collections.

Figure 1. Published comments on rulemaking dockets by year (top-panel) and the number of dockets receiving at least 1 published comment by year (lower-panel).

Regulations.gov came online in 2003, and comment volume grew fivefold to sixfold per (rough) decade of operation (earlier comments are from miscellaneous backfill by a few agencies).

Published rulemaking comments averaged approximately 46,000 per year from 2000 to 2007, rose to about 279,000 per year from 2008 to 2015, and reached 1.52 million per year from 2016 to 2025 (Figure 1). Note that the data for earlier decades (before 2007) should be treated with some caution, as agencies migrated their rulemaking to regulations.gov gradually and took different approaches to uploading and dating historical regulations and comments, some going back as far as 1970. About 500 dockets (~0.8%) from 2006 and earlier, accounting for ~247,000 comments (~1.2% of all comments) may be mis-dated due to uploading legacy rules.

Importantly, this volume increase did not occur because agencies opened more dockets that drew public feedback. Rulemaking dockets identified as rulemaking by agencies in regulations.gov with at least one published comment grew from about 540 per year during 2000–2007 to roughly 1,400 during 2008–2015. Since 2011, the annual count of commented dockets has held between 1,300 and 1,850.

The aggregate increase comes from a concentration of comments in a small number of dockets. In a typical year, the ten largest dockets account for roughly two-thirds of all published comments, with a median of 67% and a range between 42% and 91%. From 1994 forward, 43 dockets with 100,000 or more comments contain 8.0 million comments, which is 40% of all published comments.

Figure 2. Rulemaking dockets by order of magnitude of the number of comments published over time.

High-volume dockets have become more frequent over time. Prior to 2012, between zero and ten dockets opened per year accumulated 10,000 or more comments. That count rose to 27–34 per year in 2018–2019 and reached 40 in 2023 .

These individual dockets account for recent annual fluctuations:

• 2023 was the peak year at 2.29 million published comments. The Bureau of Alcohol, Tobacco, Firearms and Explosives rule on firearm dealers accounted for more than 370,000 comments (16%), alongside dockets exceeding 100,000 comments from the Department of Education, Bureau of Land Management, Consumer Financial Protection Bureau, and Fish and Wildlife Service.
• 2024 dropped to about 850,000 comments in part because only two dockets received more than 100,000, and none crossed 150,000, accounting for only 25% of comments received that year.
• 2025 saw a return to a similar dynamic, with over 2 million comments published, where 4 regulations out of 2,062 total each received more than 100,000 comments, accounting for more than 55% of comments received that year driven by extreme interest in the regulatory and deregulatory actions of the Trump Administration on such dockets as the Forest Service roadless-rule rescission (613,926 comments total).
• 2026 has already reached 1.11 million comments through early November, driven by the regulations we discussed earlier, with three 100,000-plus dockets together accounting for 40% of comments received so far.
• The largest docket in the record, however, is the Council on Environmental Quality 2019 National Environmental Policy Act revision, with 720,590 published comments.

Despite higher annual totals, the typical rulemaking docket has not changed. For dockets with at least one comment opened since 2004, the median docket collects between 3 and 7 comments, while the 90th percentile falls between 50 and 500 (Figure 2). Across all 632,43322 rulemaking dockets, about 60% (37,059833 dockets) have zero published comments. That zero-comment share was between 62% and 71% in 2006–2013 and between 44% and 54% after 2019 (Figure 2).

Overall, the distribution appears to be heavy-tailed, has grown in volume over time, but has relatively stable median and stable frequency of dockets with very-low (or zero) published comments. That strongly suggests some wackiness with the variance of the distribution, wackiness that probably follows what’s known as a power law distribution.

Why We Suspect a Power Law

As we mentioned, the difference between a median of five comments and individual proposals drawing hundreds of thousands of comments suggests a heavy-tailed process. Many physical and social processes follow a normal distribution, whereby observations cluster around an average and deviations diminish rapidly, giving rise to a bell-shaped frequency distribution. Human height follows this pattern.

A power law distribution behaves differently because observations span multiple orders of magnitude without a typical scale that can be easily referenced. Earthquake energy is a classic example: minor tremors occur constantly, while rare earthquakes account for the majority of released seismic energy.

Such processes often follow what’s called a power law distribution. Under a power law, the probability of an event decreases as a power of its size ($P(X \geq x) \propto x^{-\alpha}$). Most instances remain small, but large values occur more frequently than an exponential or normal distribution predicts. On a log-log plot, classic power law distributed data appear as a straight, downward-sloping line. There are a few first-principles based reasons to hypothesize that public comments follow this power-law pattern:

1. Preferential Attachment. In notice-and-comment rulemaking, existing attention draws more attention. Sometimes, this is called a “Matthew Effect” (the rich get richer). When a proposed rule receives press coverage, or a major organization sponsors a comment-writing campaign, public awareness of the proposed rulemaking increases and more people comment; the more people comment, the more attention it gets, in a feedback loop.
2. Networked Mobilization. Commenting does not depend solely on individuals finding entries in the Federal Register. Organizations distribute pre-written form letters through email, web portals, and software tools. These networks link central organizations to large memberships, resulting in large submission totals for targeted rules.
3. The Nature of the Internet. With relatively low burden of electronic submission of comments through both email and online forms, a docket can receive ten comments or one million comments with few hurdles. AI is almost certainly making this more challenging and the problem of bot farming comments has long been an issue.

Empirical Distribution and Tail Behavior

Evaluating the data shows that comment volume is heavy-tailed across several orders of magnitude. On a log-log plot, dockets with at least one comment track a straight line over two to four orders of magnitude. For dockets opened since 2013, that linear segment begins around 15 comments and continues through the largest dockets in the record.

Fitting a discrete power-law model using the Clauset–Shalizi–Newman method produces an estimated tail exponent (here, $\alpha$) that declines over time. We use a theoretical reference of $\alpha= 1.51$ for this analysis. The exponent drops from 1.74 for dockets opened in 1994–2007 to 1.60 in 2008–2012, 1.53 in 2013–2017, 1.44 in 2018–2022, and 1.41 in 2023–2026 (Figure 3). A lower scaling exponent indicates a heavier tail, with more total volume concentrated in the largest observations. For dockets opened from 2018 through 2026, the fitted tail accounts for 99.8% of all published comments.

Figure 3. Power-law fit results by docket epoch with reference distribution (left panel) and power-law $\alpha$ exponent over time (right panel).

However, a strict power law does not hold in every period. Bootstrap goodness-of-fit tests reject a pure power law for the two most recent epochs (2018–2022 and 2023–2026), for 2008–2012, and for the pooled data, while failing to reject it for 1994–2007 and 2013–2017. A lognormal tail fits as well as or better than a power law in every epoch, significantly so for 2008–2012, 2023–2026 and the pooled data, while exponential distributions are rejected across all eras. Because the fitted exponents are below 2 (from 1.74 in 1994–2007 to 1.41 in 2023–2026), a pure power law would imply a distribution without a finite theoretical mean, although in practice the tail is bounded (the largest docket has 720,590 comments).

The departure from a power law may reflect the mechanisms described above operating unevenly. Preferential attachment alone, with attention accumulating gradually through press coverage and word of mouth, would produce a straight line on a log-log plot. In the two most recent epochs, however, the data sit above the reference line between a few hundred and 100,000 comments before falling off steeply. It’s possible that preferential attachment dominated in the past; it is clearly ruled out today with such a vastly heavier tail than expected under that process.

The results are, however consistent with a second process layered over the preferential attachment: that is, network mobilization. Here, the opportunity cost for submitting a comment has become arbitrarily cheap - largely resulting from organizations driving form letters and providing automated submission. Until last year, GSA facilitated this by providing a batch upload feature explicitly for “advocacy or membership organizations to efficiently submit large volumes of Comments on behalf of their constituencies” through the regulations.gov POST API. That feature is gone but certain organizations are still able to automate posting through the Federal Register. Together, this moves a targeted docket into the thousands or hundreds of thousands of comments at once, without the gradual cumulative growth that preferential attachment implies. The steep drop beyond about 100,000 comments is consistent with a practical ceiling, as the number of people any one rule can mobilize or that an agency will post is finite. It just so happens that, simultaneously, more dockets have become “targeted” in recent years - compounding the process. The fall in the fitted exponent over time (to 1.41 for 2023–2026) is also consistent with a growing share of dockets reached this way, but these tests are not fully putative in that regard.

Alternatively, a single skewed process, such as multiplicative growth over a finite period, which yields a lognormal tail, cannot be excluded by our analysis (but it is unlikely from a first principles perspective). A direct test would separate campaign-driven dockets from organic ones, for example by duplicate-text share, and refit each group. If the excess in the middle of the distribution is mostly campaign-driven, that would support the mobilization account. However, that’s currently beyond our limited scope here. For analysis on the makeup of comments and commenters in these large submissions, see Devin Judge-Lord’s work on public pressure campaigns.

Improving Transparency and Utility

One observation from this exercise is the difficulty of assembling these figures. While regulations.gov is generally accessible and provides the counts for comments received and posted for each docket, it does not permit sorting dockets by number of comments or downloading search results in tabular form (similar to many NIH websites like PubMed). The bulk data download request option limits requests to agencies, dockets, or documents, and only for individual years, which could require thousands of requests to assemble a picture of the entire collection. Additionally, the API requires a query for each docket to retrieve the comment counts, and usage is limited to 1,000 queries per hour, requiring ~63 hours to retrieve the full figure for the current number of rulemaking dockets (half that if approved for the higher rate).

GSA could facilitate study of public comment behavior by allowing export of search results directly from the website in tabular form, including the posted and received comment counts already shown on each docket page, and by adding those counts as documented, sortable fields in the API’s docket search results. While we wait for GSA to improve the accessibility of its data, check out our dashboard here, which will continue to be updated for at least the near-future.

References

• Clauset, A., Shalizi, C. R., & Newman, M. E. J. (2009). Power-law distributions in empirical data. SIAM Review, 51(4), 661–703. https://doi.org/10.48550/arXiv.0706.1062
• Kaiser, J. (2026). Did 500,000 people comment on controversial U.S. grantmaking rule? Not likely, analysts say. Science. https://www.science.org/content/article/did-500-000-people-comment-controversial-u-s-grantmaking-rule-not-likely-analysts-say
• Lapowsky, I. (2017, November 28). How Bots Broke the FCC’s Public Comment System. Wired. https://www.wired.com/story/bots-broke-fcc-public-comment-system/
• Judge-Lord, D. (n.d.). Public Pressure Campaigns and Bureaucratic Policymaking. Retrieved November 8, 2026, from https://judgelord.github.io/dissertation/index.html
• Marcum, C. S. and Haddad, A. (2026, October 23). GSA closed the comment loophole on Regulations.gov. “GSA Closed the Comment Count Loophole on Regulations.gov.” Open Evidence. https://doi.org/10.59350/nd0y0-agy83 Maury, M., & Marcum, C. (2025, October 17). How You Can (and should) Shape Federal Data Collections. dataindex.us . https://dataindex.us/newsletter/article/6cfecae3-3c89-487e-b8f7-cfcd85ded6c7
• Sherburne, M. (2009, November 17). U-M researcher finds a better power law that predicts earthquakes, blood vessels, bank accounts. Michigan News. https://news.umich.edu/u-m-researcher-finds-a-better-power-law-that-predicts-earthquakes-blood-vessels-bank-accounts/
• Wakefield, J. (2017, October 6). Net neutrality: Millions of fake comments submitted to FCC. BBC News. https://www.bbc.com/news/technology-41497342

Tags: general