Won’t bots and fake accounts flood a participation platform?

Bots, fake identities, and insincere respondents can flood any open online participation system, and documented cases show the scale: the New York Attorney General found that nearly 18 million of more than 22 million comments in the Federal Communications Commission’s 2017 net neutrality proceeding were fake. My Vote Forward’s demo uses simulated data and does not defend against this in any real sense. A real deployment of the framework would need one verified account per eligible person, published methods for detecting coordinated activity, and results that state how participants were verified; even then, detection is never complete.

The objection in its strongest form

The objection to My Vote Forward on bots is that open online participation has been tested at scale and failed. Comment systems, petition sites, and online polls have all been gamed by automated submissions, impersonation, and paid respondents. Large language models now make it cheap to generate plausible, varied text and to operate many accounts that behave like people. Any published ranking of public priorities becomes a target worth attacking: move the ranking, and you move the headline. The critic concludes that the more consequential the results become, the more certain it is that they will be manipulated, and that a project with no enforcement power and no legal identity infrastructure cannot keep up.

A further version targets the defense. Strong identity checks would stop most fakes, but they bring their own costs: they exclude people without documents or devices, they create a database linking real people to their participation, and they move the system toward the kind of surveillance the framework says it opposes. So the critic offers a dilemma: either the system is open and fake, or it is verified and exclusionary.

Why the objection matters

Fake participation matters for My Vote Forward because the framework’s value depends on results meaning what they say. A ranking of priorities that is 10% fabricated is not 90% useful; readers cannot tell which 10% is fake, so the whole result loses credibility. The objection also matters for fairness. If one side of a question is more willing or able to fabricate participation, fabrication does not add noise, it adds bias.

Defined terms

• Bot: software that performs actions, such as submitting forms or posting, automatically and often at scale.
• Sybil attack: one actor operating many fake identities to gain disproportionate influence in a system that assumes one identity per person.
• Bogus respondent: Pew Research Center’s term for survey participants who do not answer questions sincerely, often to collect rewards with minimal effort.⁠²
• Identity proofing: confirming that a claimed identity belongs to a real person. See identity proofing.
• Authentication: confirming that the person returning is the same one who enrolled. See authentication.

Evidence

The clearest documented case is the FCC’s 2017 net neutrality rulemaking. The New York Attorney General’s office reported on May 6, 2021 that “nearly 18 million of the more than 22 million comments” submitted were fake.⁠¹ The office found that the nation’s largest broadband companies funded a campaign that generated “more than 8.5 million fake comments that impersonated real people,” and that most of 9.3 million fictitious comments on the other side came from “a 19-year old college student using automated software”.⁠¹ Agreements with three companies required them to pay more than $4.4 million in penalties and disgorgement.⁠¹ Fabrication came from both directions of the debate, which is a reminder that this risk is not partisan.

Online opt-in polls show a quieter version of the problem. Pew Research Center reported in 2024 that an opt-in poll found 20% of United States adults under 30 agreeing that the Holocaust is a myth; Pew’s probability-based panel found 3%.⁠² In a 2022 experiment, 12% of opt-in respondents under 30 claimed to be licensed to operate a nuclear submarine, a qualification whose true prevalence Pew says “rounds to 0%”.⁠² Pew attributes these results to bogus respondents and reports that overreporting is concentrated among adults under 30 and Hispanic adults.⁠² In a 2026 methods report, Pew describes an online opt-in survey of 11,114 adults conducted November 14 to 19, 2024, of whom 9,151 passed trap questions designed to catch insincere answers;⁠³ the report’s title is “No Easy Fix for Bogus Respondents in Online Opt-In Polls.”

Fraud is not limited to online systems. In January 2026, the Swiss Federal Council reported that about 30,000 signatures collected for roughly twenty federal popular initiatives had been reported as suspected forgeries since October 2022, and that federal prosecutors had searched several signature-collection companies.⁠⁴ The release does not state that any initiative qualified only because of forgery, and the presumption of innocence applies.⁠⁴ The case shows that paid collection creates incentives for fabrication on paper as well as online.

How comparable systems have handled it

Systems that face fake participation use a mix of identity, detection, and labeling.

• Identity tied to an official registry. Estonia’s internet voting, discussed in the Estonia case, uses the national electronic identity system. That stops casual fakes but depends on a national ID infrastructure most countries do not have.
• Checking and prosecution after collection. In the Swiss case, most criminal complaints about suspect signatures came from the Federal Chancellery, and prosecutors then searched collection firms.⁠⁴ Checking after the fact catches some fraud, but only after it has entered the process.
• Probability samples instead of open participation. Pew’s own response to bogus respondents is to rely on panels recruited by random sampling rather than open opt-in.⁠² That trades openness for reliability.
• Federal identity guidance. NIST’s Digital Identity Guidelines, finalized as Revision 4 in July 2025, separate identity proofing (volume 800-63A), authentication (800-63B), and federation (800-63C).⁠⁵ They are guidelines for federal systems, not an approval of any project, but they give a vocabulary for stating what level of assurance a system actually has.

Competing interpretations

One view holds that strong identity is the only real defense. If each participant is proofed against an official record, fabrication at the FCC scale becomes impossible, and the remaining problem is persuasion, which is legitimate. The cost is exclusion and the creation of a sensitive linkage between people and participation.

A second view holds that detection is enough for advisory results. Coordinated campaigns leave statistical traces, such as bursts of identical or near-identical submissions, and methods for detecting them can be published. On this view, results can be released with an estimate of excluded activity and a stated method, as survey researchers do with trap questions.⁠³

A third view holds that open online participation should never be presented as a measure of public opinion at all, regardless of defenses, because self-selection and fabrication together make it unreliable. On this view a system like My Vote Forward should present results only as “what participants said,” never as “what the public thinks.”

What the framework answers

My Vote Forward’s position combines all three views and labels the limits.

• The demo uses simulated data. Nothing in the demo is real participation, so nothing there can be flooded in a way that matters. That is not a defense; it is a statement of status. See is this an official voting service?.
• One account per verified person in any real deployment. The framework separates identity proofing from eligibility and from authentication. See identity and registration and does a fingerprint prove eligibility?.
• Published detection methods. Methods for flagging coordinated activity would be published with each release, along with the count of excluded submissions and the reason category.
• Results labeled by verification level. A release would state how participants were verified and would never present open, unverified participation as a measure of the public.
• No reward for participation volume. The framework does not pay per response, which removes the incentive Pew identifies behind many bogus respondents.⁠²

Hypothetical example. A pilot release finds that 2,000 submissions arrived within ten minutes from accounts created the same day, all ranking the same subject first. The release excludes them under a published rule, reports the count and the rule, and shows results with and without the exclusion so readers can judge the effect.

Where the objection still wins

The objection wins on the dilemma. Strong identity reduces fabrication but excludes people and creates sensitive data; open participation includes people but invites fabrication. The framework chooses verified participation for any real deployment and accepts the exclusion cost, which is a real cost. See what about people without internet?. The objection also wins on detection: no method catches every coordinated campaign, and publishing detection rules helps attackers learn to evade them. The framework has no tested detection system, no real data, and no adversarial evaluation.

Unresolved questions

• What level of identity assurance is proportionate for advisory priority-setting, as opposed to official voting?
• Should detection rules be published in full, in summary, or only to independent auditors?
• How should a release treat genuine but organized participation, such as a union or church urging members to take part?
• Can privacy-preserving credentials prove “one real eligible person” without revealing who that person is?

Related

• Measurement
• Identity and registration
• Can wealthy groups dominate?
• Opt-in sample
• Hard questions