ISSUE 02 · OCTOBER 2026 · PEOPLE / TOMORROW
What Is Your
Attention Worth?
Political money buys a chance to be seen. Outrage can win a reaction. Neither is the same as earning trust.
A sourced journey through money, messages, synthetic media and the way a democracy disagrees.
Begin with the disagreementSTOP. THE QUESTION IS BIGGER THAN THE NOISE.
DEMOCRACY DOES NOT ASK US TO AGREE.
IT ASKS US TO LIVE WITH DISAGREEMENT.
Disagreement is part of the work. The question is what happens when a system rewards winning our attention faster than it rewards earning our confidence.
ADIMPACT · JUNE 11, 2026 FORECAST
$11.6BILLION
Projected 2026 political advertising.
Not completed expenditure.
What is all that money trying to buy?
A chance to reach
your attention.
The full-cycle projection describes advertising, not every dollar spent on elections. It cannot tell us how persuasive an ad is, whether its claims hold up or how many minds it changes. [1]
Travel inside the numberWHAT DOES IT BUY? · SAME FORECAST, DIFFERENT CHANNELS
Different screens.
One finite resource.
Each bar uses the $11.6B total as its scale. Digital covers the platforms named by AdImpact: Facebook, Google, Snapchat and X. The $1.7B remainder is TMLPT’s arithmetic residual, not a separately reported category. Source figures are rounded. [1]
Exact values and the calculation
| Channel | Projection |
|---|---|
| Broadcast television | 5.6 |
| Connected television | 2.7 |
| Named digital platforms | 1.6 |
| Other / rounding residual | 1.7 |
11.6 − 5.6 − 2.7 − 1.6 = 1.7. This is a forecast decomposition, not measured spending or a measure of negative advertising.
WHO PAYS? WHO GETS PAID?
The message has a sponsor.
The placement has a seller.
A candidate committee, a party committee and an independent organization are different speakers. An ad’s payer and authorization disclaimer help identify that distinction. They do not settle the truth of its argument. FEC rules address federal communications and vary by format and sponsor; a state or local ad needs its own jurisdictional check. [6]
Political advertisers purchase distribution from media businesses. Editorial judgment, campaign strategy and ad sales are different decisions, even when they occupy the same screen. Asking who profits is legitimate. Assuming that every seller chose a message’s content, or that every journalist serves an advertiser, is not evidence.
The next time a political spot interrupts a program, the smallest type may open the largest question: who paid for this, who authorized it, and what record would let us inspect the purchase?
The political ad your TV station may not be allowed to editTHE ATTENTION INCENTIVE
WHO PROFITS
FROM OUR ANGER?
A reaction is a visible reward.
Rathje and colleagues examined 2.7 million posts from U.S. congressional and news accounts on Facebook and Twitter. References to the political out-group predicted more sharing; that language was a stronger engagement predictor than the other measured emotional categories. This was an observational analysis. It did not establish that an algorithm caused hostility or that a shared post changed a vote. [7]
A reward is not the whole explanation.
An angry response might express solidarity, criticism, fear, disagreement or amusement. Sharing is not endorsement. The economic question is what an engagement reward encourages publishers and campaigns to repeat. The democratic question is whether the repeated message helps us understand an opponent—or only recognize an enemy.
TMLPT interpretation: when reactions are easy to count and trust is harder to observe, the visible reward can become a tempting substitute for the public purpose.
THE COUNTEREVIDENCE BELONGS IN THE STORY
An algorithm is influential.
That does not settle its effects.
A randomized Facebook/Instagram study during the 2020 U.S. election changed consenting participants’ feeds to chronological order. Their platform experience changed, but the researchers did not detect significant changes in key political attitudes over the three-month study. The paper’s correction record includes December 2024 and March 2026 notices, both linked below; no content-exposure estimates are reproduced here. [9][20][26]
This finding does not certify every ranking system as harmless. It also prevents a shortcut: moving from “a feed changes what people see” to “this feed caused today’s political polarization.” A defined experiment, a particular population and a limited observation period answer a narrower question.
A separate study ran 59 randomized experiments with ads from the 2016 presidential campaign and found small average persuasion effects in that setting. That is not a claim that advertising never matters, especially in a close contest, or a prediction about 2026. [27]
We need to distinguish distribution, belief, emotion and behavior. Treating them as one thing makes a simple story. Keeping them separate makes a more useful investigation.
CLAIM → EVIDENCE → WHAT WE KNOW → WHAT WE DON'T KNOW
What if the thing demanding
your attention isn't true?
Before sharing, identify what kind of statement you are being asked to accept. A factual assertion can be tested. An allegation needs support. A prediction concerns an outcome that has not happened. An opinion can be argued without becoming a measured fact.
CLAIM UNDER EXAMINATION · NOT TMLPT'S ASSERTION
“$11.6 billion has been spent.”
- EVIDENCE
- AdImpact’s June 11 release calls $11.6B a full-cycle projection. [1]
- WHAT WE KNOW
- The forecast exists. Its status is projected.
- WHAT WE DON'T KNOW
- This source does not establish the final expenditure or the latest amount actually aired.
CLAIM UNDER EXAMINATION · NOT TMLPT'S ASSERTION
“A station aired it, so it checked it.”
These are illustrative errors about this investigation’s evidence, not quotations attributed to a candidate. We have not manufactured paired partisan examples. The standard travels with the claim, whoever makes it.
THE DIFFERENCE THAT MATTERS
ATTENTION IS EASY TO DEMAND.
TRUST HAS TO BE EARNED.
Trust has a question and a denominator.
Pew’s September 2025 poll put trust in Washington to do what is right always or most of the time at 17%. That is a defined survey response, not a 2026 measure of all trust or proof that campaign ads caused the result. The historical series combines organizations and, for much of its span, a moving average. [10]
Incivility has a measured—but limited—signal.
A meta-analysis of 35 experiments found a small average negative effect on political trust: Hedges’ g −0.19, with a 95% interval of −0.30 to −0.09. The participation estimate was near zero, g −0.02 (−0.11 to 0.08). Publication-bias indications and varying study designs limit confidence. These are standardized effects, not percentage-point changes. [8]
Study estimates, intervals and limits
| Outcome | Hedges’ g | 95% confidence interval |
|---|---|---|
| Political trust | −0.19 | −0.30 to −0.09 |
| Political participation | −0.02 | −0.11 to 0.08 |
24 manuscripts, 35 studies. A near-zero pooled estimate does not mean no individual is affected. Engagement, trust and participation are different outcomes. No universal claim about every political message follows.
OLD BEHAVIOR. DIFFERENT MACHINERY.
We did not invent vicious politics.
The Library of Congress documents the bitter 1800 campaign and partisan press. Its surviving letters and publications make a useful correction to a comforting story: disagreement was not universally gracious before social media. Historical documents do not provide a comparable national civility score. [11]
What changes is not just the temperament of a speaker. It is the cost of making a message, the speed of distributing it, the incentives around it and the difficulty of verifying what looks like evidence. We should investigate those changes without announcing that this is history’s uniquely uncivil year.
TOMORROW · THE VERIFICATION QUESTION
REAL?
ALTERED?
GENERATED?
A fabricated clip and an authentic clip stripped of context can both mislead. A realistic appearance is not a chain of evidence.
NIST examines provenance, watermarks and detection as different tools with different limits. None makes an image’s political argument automatically true. A missing label is not proof of authenticity, and a detector score should not become a verdict on its own. [12]
Travel through synthetic media, disclosure and the rulesA QUIETER STOP · MORAL CONSISTENCY
WHAT DOES OUR FAITH
REQUIRE OF OUR POLITICS?
In Matthew’s Sermon on the Mount, the teachings about anger, truthful speech and love of enemies disturb an easy division between deserving insiders and disposable outsiders. Applying them to contemporary civic conduct is interpretation; claiming that they endorse a party is something else. [16]
Does the standard change
when it is our side?
ACCOUNTABILITY WITHOUT DEGRADATION
CIVILITY IS NOT SURRENDER.
A firm accusation can be warranted. A protest can be urgent. A question can be uncomfortable. Respect for a person’s humanity does not require agreement with their claim, acceptance of their conduct or equal treatment of unequal evidence.
Ask where a message does its work. An argument identifies a proposition. An insult degrades a person. Evidence can support criticism; insinuation invites belief while avoiding the burden of support. A sharp exchange may contain more than one of these. Classify the material, not merely the speaker’s team.
PEACE IS NOT THE ABSENCE OF DIFFICULT QUESTIONS.
Negative campaigning is not automatically incivility. Criticizing an opponent’s record can provide information; humiliating a person does different work. A 2007 research synthesis found no reliable evidence that negative campaigning depresses turnout. Its older, varied studies do not settle the effect of a particular 2026 ad. [25]
The Capitol Police reported investigating 14,938 concerning statements, behaviors and communications in 2025. That administrative count includes activity directed at lawmakers, families, staff and the Capitol complex; it is not 14,938 attacks. Reporting and assessment practices affect the count. [24]
Threats and intimidation deserve their own evidentiary record. This investigation does not attribute a violent incident to a particular advertisement or infer a causal chain from adjacent headlines. The distance between rhetoric and violence must be researched, not filled with certainty.
FROM BEING ACTED UPON TO HAVING AGENCY
BE A PARTICIPANT.
Democracy is something people practice. Voting is one part. Reading, questioning, listening, challenging your own side, joining a public meeting and organizing around a documented need are parts too.
- Name the claim. Separate its factual assertion from opinion, prediction and allegation.
- Open the record. Read the original source, date and relevant surrounding material.
- Check the denominator. Ask whether a number is observed, modeled, projected or estimated.
- Try the same standard. Would the evidence satisfy you if the speaker belonged to another group?
- Leave room to revise. State what would change your mind; correct a shared error where you shared it.
- Take part. Speak, listen, learn, organize and vote according to your own considered judgment.
These are TMLPT editorial practices, not a validated cure for polarization. They preserve your room to think. We do not tell you whom or what to support.
WE'RE ALL ON A JOURNEY.
Different histories. Different convictions. Different conclusions. A shared obligation to live with disagreement.
WE PUSH FOR PEACE.
THE STANDARD ALSO APPLIES TO THIS PUBLICATION
PUSH THE QUESTION.
DON'T PUSH THE ANSWER.
INVITE PEOPLE INTO INQUIRY.
DO NOT RECRUIT THEM INTO CONCLUSIONS.
TMLPT competes for attention too. We will not lie to get it, manufacture certainty, humiliate someone or disguise allegation as fact. The evidence, design and question should earn the spark. If a claim here fails that standard, send a correction to the Train.
FOLLOW THE EVIDENCE · REVIEWED 5 OCTOBER 2026
Trust needs an open record.
Sources establish specific findings, not every connection in the narrative. Questions, participation practices and the argument linking incentives to public purpose are editorial analysis. No stakeholder interviews, original polling or comprehensive ad-content audit is implied.
- AdImpact · updated 2025–2026 projection
11 June 2026 · company forecast: $11.6B total; $5.6B broadcast, $2.7B CTV, $1.6B named digital platforms. Rounded estimates, not audited final spending.
- U.S. Code · 47 USC §315
Statutory text · candidate use, no censorship, news exceptions and political-file records. Context matters; not a general exemption from law.
- FCC regulation · 47 CFR §73.1941
Current regulation reviewed 5 October 2026 · candidate appearance by voice or picture; specified bona fide news exceptions.
- Supreme Court · Farmers Union v. WDAY
360 U.S. 525 · 1959 · station immunity for defamatory material in covered candidate broadcasts; not immunity for the candidate.
- FCC regulation · sponsorship identification
47 CFR §73.1212 · paid broadcast sponsorship and reasonable diligence. Sponsor disclosure is not a truth certification.
- FEC · advertising and disclaimers
Current federal guidance reviewed 5 October 2026 · authorization and payer disclosures; requirements differ by communication and sponsor.
- Rathje, Van Bavel & van der Linden · out-group engagement
PNAS · 2021 · observational analysis of 2.7 million Facebook/Twitter posts from congressional and news accounts. Association, not proof of algorithmic causation or changed votes.
- Van ’t Riet & Van Stekelenburg · incivility meta-analysis
Human Communication Research · 2022 · 24 manuscripts, 35 experiments; trust g −0.19, 95% CI −0.30 to −0.09; participation g −0.02, CI −0.11 to 0.08. Small average trust effect; publication-bias indication.
- Guess et al. · chronological-feed experiment
Science · 2023 · U.S. 2020-election Facebook/Instagram experiment; specific platforms, consenting participants and three months. Corrections dated December 2024 and March 2026 are linked below; no content-exposure estimates reproduced here.
- Pew Research Center · public trust, 1958–2025
4 December 2025 · September 2025 poll: 17% trust Washington always/most of the time. Historical series combines surveys and smoothing; not a 2026 poll or causal test.
- Library of Congress · the 1800 election
Curated historical primary documents and institutional context · not a quantitative civility comparison with 2026.
- NIST · synthetic-content transparency
20 November 2024 · technical overview of provenance, watermarking and detection; limitations remain. Not a universal authenticity certification.
- FEC · AI and fraudulent campaign authority
27 September 2024 · technology-neutral interpretation of 52 USC §30124; not a comprehensive AI-ad ban. Current guidance directory still lists this interpretation.
- Minnesota · election deepfakes
Current §609.771 reviewed 5 October 2026 · consent, intent, knowledge/recklessness and election timing conditions; federally required candidate broadcasts excepted. No case-specific legality judgment.
- Google · political-content advertising policy
Live policy reviewed 5 October 2026 · consequential synthetic/altered election-ad content requires disclosure; inconsequential alterations excluded. Policy is not evidence of uniform enforcement.
- USCCB · Matthew 5
Matthew 5:21–26, 33–37, 43–48 · Sermon on the Mount; primary scriptural text and Catholic study notes. Not a modern party platform.
- USCCB · Luke 10
Luke 10:25–37 · Good Samaritan and the neighbor question; primary text and contextual notes. Civic application here is editorial interpretation.
- Associated Press · AI ads and disclosure
25 September 2026 · contemporary secondary reporting. Documents AI-ad/disclosure examples; no unsupported estimate of election effects.
- FCC · AI voices and the TCPA
2024 rulemaking discussion recounts February declaratory ruling: synthetic voices fall within artificial/prerecorded-call restrictions. Consent, exemptions and context matter; not a blanket ban on all AI audio.
- Science · feed-study correction
5 December 2024 · erratum linked to the 2023 experiment. Retained visibly for correction transparency; no corrected table values plotted.
- Iowa Legislative Services Agency · AI laws
June 2026 · nonpartisan legal-background survey; references 2027 Code and warns against case-specific legal conclusions. Proposed election-media bills are not treated as enacted law.
- Reuters · synthetic campaign depictions
28 March 2026 · secondary reporting documents synthetic political depictions, including Loudoun County Republican Committee videos. Examples do not measure election effects.
- Eighth Circuit · Kohls v. Ellison
9 February 2026 · No. 25-1300 · standing and preliminary-injunction decision; did not resolve the statute’s constitutionality on the merits.
- U.S. Capitol Police · threat assessment cases
27 January 2026 · 14,938 concerning statements, behaviors and communications investigated in 2025. Administrative case count, not a count of attacks or proof of advertising causation.
- Lau, Sigelman & Rovner · negative campaigns
Journal of Politics · 2007 · research synthesis found no reliable evidence of depressed turnout. Negative campaigning differs from incivility; older heterogeneous evidence does not settle every 2026 message.
- Science · March 2026 feed-study correction
19 March 2026 · second erratum identified through publisher Crossmark and PubMed 41855352. No corrected content-exposure values reproduced; current correction record remains part of the source trail.
- Coppock, Hill & Vavreck · advertising experiments
Science Advances · 2 September 2020 · 59 randomized experiments using 2016 presidential ads; small average persuasion effects in this setting. Not a total effect of all campaign activity or a prediction of the 2026 result.