The Room
New Jersey State Senator Nellie Pou sits in her Trenton office reviewing a draft bill about facial recognition in law enforcement. On her desk: three position papers. One from the ACLU. One from a police chiefs association. One from a tech industry coalition that didn’t exist six months ago.
The tech coalition’s paper is eight pages, single-spaced, with footnotes citing peer-reviewed studies. It proposes specific language for three sections of the bill. Senator Pou’s legislative aide has already flagged two of those suggestions as “reasonable compromises.”
This is how lobbying works. Not dramatic. Not corrupt. Just early, specific, and in the room before the public debate starts.
What Lobbyists Actually Do
A lobbyist’s job is to make a legislator’s job easier. Legislators face hundreds of bills per session, covering everything from school funding to water rights to emerging technology they learned about last month. They can’t become experts in all of it. They need information they can trust, fast.
Lobbyists provide that information — but shaped to support a particular outcome. The ACLU’s position paper on facial recognition will cite different studies than the tech coalition’s paper. Both sets of studies are real. Both are peer-reviewed. The difference is which questions got asked and which risks got emphasized.
The tech coalition isn’t lying when it says its proposed language “balances innovation with privacy protections.” The ACLU isn’t lying when it says the same language “creates loopholes that gut enforcement.” They’re highlighting different parts of the same system.
The Timing Advantage
Here’s what makes lobbying powerful: timing.
When a bill is first conceived — often by a single legislator or a small working group — the public isn’t watching yet. There’s no hearing scheduled. No news coverage. Just a legislator who knows something needs to be done about AI in hiring, or algorithmic bias in mortgage lending, or deepfakes in political ads.
At that stage, the legislator’s office is hungry for two things: (1) what other states have tried, and (2) draft language they can adapt. Lobbyists show up with both. They bring binders comparing five states’ approaches. They bring model bills with blank spaces where the state name goes.
By the time the bill reaches a public hearing, the core framework is already set. The hearing is where voters and advocacy groups respond to what’s already on the page. They can push for amendments. They can oppose the whole thing. But the starting position — what counts as the “moderate” option that most legislators will default to — was often shaped weeks earlier by whoever got in the room first.
State Level Stakes
Why statehouses? Why not focus lobbying on Congress?
Three reasons. First, state legislatures move faster. A federal AI regulation bill might take three years and face 50 competing interests. A state bill can pass in one session if the governor supports it. Second, state laws create facts on the ground. If California bans a certain use of AI in employment screening, tech companies build their products to comply — because California is too big a market to ignore. Other states then adopt California’s standard, and the federal government often follows years later.
Third, state legislators are more accessible. A lobbyist for a large tech company can get a meeting with a state senator’s chief of staff in two weeks. Getting equivalent access to a U.S. senator takes months and often requires hiring someone who used to work in that office.
This is why AI companies have hired 30% more state-level lobbyists in the past two years, even as federal AI legislation stalls. The action is in Sacramento and Austin and Albany.
The Information Asymmetry
Most lobbying isn’t about campaign donations. It’s about information asymmetry.
A state legislator might have two hours to prepare for a committee hearing on algorithmic transparency in healthcare. The lobbyist has spent six months studying the issue. The lobbyist has talked to engineers who build the systems. The lobbyist has read the EU’s Digital Services Act and can explain why the European approach won’t work in U.S. markets.
The legislator knows the lobbyist has a motive. But the lobbyist is also the most informed person in the room — more informed than the legislator’s own staff, who are generalists covering ten issues. So the legislator listens. And when the lobbyist suggests language that “threads the needle” between innovation and safety, the legislator often accepts it, because threading needles is the job and the lobbyist just saved them 20 hours of research.
This isn’t corruption. It’s the subsidy of free expertise. But free expertise is never neutral.
What Public Testimony Misses
By the time a bill reaches public comment, the battle is halfway over.
The public sees: a hearing where citizens testify for three minutes each, a committee vote, a floor debate. What the public doesn’t see: the six months before the hearing, when the bill’s framework was built from model language a lobbyist provided, using definitions the lobbyist suggested, with enforcement mechanisms the lobbyist said were “workable.”
A citizen can show up at a hearing and say, “This bill doesn’t go far enough.” That’s valid. But if the bill defines “AI system” in a way that excludes 60% of the tools causing harm, the citizen is arguing over details while the foundation is already set.
The Mechanic Underneath
Lobbying works because legislation requires specificity. You can’t pass a law that says “make AI fair.” You need definitions: What counts as AI? What counts as fair? Who checks compliance? What happens when someone violates it?
Every one of those questions has 10 possible answers. The lobbyist’s job is to make sure the answer that gets written down is the one their client can live with — ideally, one their client already complies with, so the law burdens competitors more than it burdens them.
This is the system. It’s not hidden. Lobbyist meetings are often public record. Position papers are submitted as part of the legislative file. But knowing the system exists and watching it operate in real time are different things.
Senator Pou will read all three position papers. She’ll ask her staff which one’s language is cleanest. And the bill that reaches the floor will be a hybrid — part ACLU language, part police language, part tech coalition language. The coalition won’t get everything. But they’ll get some things. And “some things” is often enough, because it’s “some things” that weren’t in the bill before they showed up.
The Question For The Rest Of Us
If lobbying is early, specific, and armed with binders of research, what’s the counter-move for people who aren’t paid to be in the room?
One option: find the legislative aide before the hearing happens. Aides read email. They take meetings if you represent a group. They’re looking for information too — just usually later than the lobbyists arrive.
Another: track model bills. Organizations like the American Legislative Exchange Council (ALEC) publish model legislation that shows up, nearly word-for-word, in multiple states. If you see the same AI bill language in three states at once, someone’s lobbying at scale. Knowing that changes what you’re responding to.
The structural answer is public-interest lobbying — advocacy groups that do the same work lobbyists do, but for outcomes that don’t have a corporate client. Those groups exist. They’re outspent 10-to-1, but they know the timing game. They show up early too.
The system runs on who’s in the room when the room is still empty. If you’re not in that room, you’re responding to decisions already made.