Scott Goodstein published something in USA Today last week that should have stopped every Democratic digital director cold. The man who ran external online for Obama in 2008 argued that the most dangerous political misinformation of 2026 will not go viral. You may never see it at all.
He calls them social media missiles. Short videos, matched-list texts, tightly targeted ads built to land on a small set of persuadable voters in one district and nowhere else. An attack that reaches 25,000 of the right people is worth more than one that reaches ten million of the wrong ones. And because it never trends, no reporter flags it, no fact-checker rates it, and no opposing campaign ever gets a copy. AI search compounds the problem. Once false information enters the online ecosystem, AI tools can repeat it as an authoritative answer - giving a targeted lie a much larger audience and a much longer life.
Political operatives launch social media missiles - often before we even know they exist. By the time we detect them, the damage is already done.
Scott’s proposed fix includes better social media listening. Social intelligence, faster detection, real fact-checking capacity in the places PolitiFact never visits - and getting facts to voters before the damage is done. He’s right, and we should build all of it. Several progressive technologists, including the co-author of this post, Tim Chambers, have been working on exactly that sort of advanced social listening.
But detection is a race we start from behind. By the time you have found the ad, coded the narrative, written the correction and bought the media to deliver it, the voters who saw the lie have already moved on - and you are paying to reach them a second time. In Ohio in 2023, Goodstein watched more than $5 million in false ads run against a ballot measure while he was still on the phone looking for someone to fact-check them.
There is a second line of defense defined in this new paper, and it doesn’t require seeing the missile first. It requires getting there before it launches.
What the evidence actually says now
Tim just finished a review of the last four years of research on inoculation - aka prebunking - for Innovating for the Public Good. Two years ago this was a promising lab finding. It isn’t anymore.
It runs at real scale, cheap. Ahead of the 2024 European Parliament elections,Google/Jigsaw put prebunking videos in front of more than 120 million YouTube users. In a separate paid campaign across five countries, the videos were distributed across YouTube and Meta. Researchers then tested national-language versions in 13 preregistered surveys with nearly 20,000 participants in 12 EU countries. Every participant was 45 or older, the audience everyone assumes is a lost cause. The videos worked anyway.
It is affordable enough for a state party. Researcher Sander van der Linden’s and team ran a 19-second prebunking video as a paid Instagram Story to 375,597 UK users. People who saw it identified emotional manipulation 59.6% of the time against 38.2% for the control group. That 21-point gap was still there five months later. The cost was roughly $8.25 CPM. About $825 to reach 100,000 people.
It doesn’t require knowing the specific lie. This is the part that answers Goodstein directly. The strongest results come from technique-based prebunking - teaching people to recognize scapegoating, fake experts, false dichotomies, manufactured outrage - not from pre-rebutting a particular claim. Researchers found cross-protection: learning to spot one tactic improved detection of unrelated ones. You cannot pre-refute an ad you’ve never seen. You can teach a voter to recognize what it’s doing to them.
And it holds up under scrutiny. A 2026 reanalysis of 33 experiments with 37,025 participants found inoculation improves people’s ability to tell reliable content from unreliable, without just making them cynical about everything.
Now the part most people skip
We are not going to sell this as a silver bullet, because the same research is brutally clear about where it breaks.
It dies in a noisy feed. Across five preregistered studies with nearly 4,800 participants, researchers put a validated inoculation video inside a simulated scroll feed. Mixed in with ordinary posts, the effect vanished. Two field studies on X found no meaningful change in real posting behavior. Passive organic exposure is not a delivery plan. Organic prebunking campaigns by trusted messengers in places like Facebook Groups, paid placement of prebunking messages in unskippable pre-roll, placement through trusted creator networks, and paid placement in direct digital channels are.
It decays fast without rehearsal. In one study, participants who watched an inoculation video and then simply went about their day showed zero difference from the control group 48 hours later. Zero. The people who retained it were the ones who did something immediately after - a quiz, a question, any active recall. A booster at day 10 held a medium effect at day 30. The mechanism is memory, not motivation. One-shot campaigns should expect a one-shot result.
Not every version works. In one trial the video teaching “moving the goalposts” made viewers more suspicious of legitimate content too. Researchers flagged it as potentially harmful. Test before you scale.
Lastly, it doesn’t work unless it is a planned, intentional and central part of your digital campaign. This needs focus, attention and planning. It is not a digital “side-quest” to experiment on after your list building and fundraising “real work” is done.This is not a digital side project.
It is part of the work of winning elections. Give it a budget. Make somebody responsible for it. Build repetition and trusted messengers into the plan from the beginning. List building and fundraising matter. But raising money to reach voters while neglecting the lies shaping what they believe is getting the priorities backward. Defending against disinformation is the real work, not something we get to after the real work is done.
What to do in the next eight weeks
Teach tactics, not villains. “Here’s what a manufactured outrage ad looks like” travels further than “the other side is lying to you,” and it doesn’t collapse the second the messenger is coded as partisan.
Buy placement, or target users outside of a noisy feed via earned media and influencers, don’t hope for organic reach. Pre-roll and paid social with verified exposure. If it lives or dies on organic pickup, it dies.
Plan the second and third touch before you buy the first. Boosters and an immediate interactive beat one great video every time.
Use AI for speed, with a human on the trigger. A 2026 study of 4,293 registered voters found AI-generated prebunks built from a carefully human-designed template were not detectably worse than human-reviewed versions. Never skip human review, btw. Build the template once, adapt it in hours. Review every output.
Deliver through people, not just placements. A Stanford evaluation of digital media literacy workshops found that only the participants recruited through trusted community organizations improved at actually telling true news from false, even though both groups understood the material equally well. This is the work Will has been doing for two years: training local messengers and producing content that sounds like it came from the community it’s aimed at. Placement buys you exposure. Trusted delivery is what makes it stick.
Keep it non-partisan on its face. The effects hold across party lines. That’s an asset, and it’s the first thing lost when this becomes a partisan product.
Start now anyway. Eight weeks is not enough to build the ideal program. It is enough to run one. And begin planning NOW for this as central to your 2028 plans.
Goodstein is right that we spent two decades building better missiles and forgot to build radar.
We’d add one thing: We need better monitoring but we also need voters who are better prepared for the attacks our monitoring will miss. Teach people how manipulation works. Reinforce those lessons. Deliver them through people they trust.
And don’t build another program that disappears the morning after the election.
The disinformation won’t stop. Neither should the work of preparing people to recognize it. Start now, keep building, and make this part of how we communicate—not just how we respond to a crisis.
When the rules change, the strategy must change. We cannot keep waiting for the next lie to teach us that lesson.
Social media monitoring tells you what already launched. Prebunking is the only defense that works on the attack you never see. And it’s the only one you can build before you know what’s coming.
We should be doing both. We’re barely doing either.
Scott Goodstein can be reached at Catalyst Campaigns
This is a very helpful analysis. Using this approach when looking at our deteriorating health care system I think we have to continuously and in innovative ways highlight how millions more uninsured is leading to increased death and deprivation.
Smart and important advice as always from Will and something every campaign, issue org, party committee, funder, and more should be thinking about.