Paper V: The Last Mile Problem — Kyanos Intelligence

Abstract The traditional channel model — television, mail, canvassing, digital — measures persuasion at delivery. It does not account for what happens when a voter turns to an answer engine to verify what they heard. That moment is the last mile: a persuasion that is confirmed or undone by whatever the AI returns. This paper introduces Effective Persuasion Cost (EPC) as a framework that accounts for this fifth dimension. It maps the exposure campaigns are currently failing to measure, quantifies the persuasion value at risk across three race tiers, identifies the offensive upside of favorable AI representation, and establishes monitoring as a standard budget line alongside earned media tracking and tracking polls.

↑ Contents I. The Incomplete Channel Model

Campaigns buy television to reach voters, mail to remind them, canvassers to persuade them, and digital to target them. The four-channel model has served as the architecture of political communication budgets for two decades. It has a conceptual flaw that has not mattered until recently: it treats persuasion as terminal.

A voter moved by a contact — a door knock, a television spot, a mailer — is assumed to hold that persuasion through election day, interrupted only by counter-advertising, opponent contact, or the passage of time. The model does not account for what the voter does next.

In 2026, what many voters do next is open their phone and ask an AI. They Google the candidate. They open ChatGPT. They check Perplexity or Google AI Overview. What they find in that moment either confirms the contact and hardens the persuasion, or contradicts it and unravels it. The traditional model does not capture this step at all. It is the incomplete model's blind spot.

The campaigns operating on the incomplete model are making budget decisions with an input missing. They are measuring cost per contact, cost per persuasion, and contact rates across four channels — and leaving unmeasured the one moment when a persuasion is most likely to be reversed or reinforced. That moment is the last mile.

↑ Contents II. The Fifth Dimension

AI verification is not a fifth campaign channel in the way television, mail, canvassing, and digital are channels. A campaign cannot buy AI inventory. It cannot target AI against a voter file. What a campaign can do is build the information record that AI draws on — and monitor whether the AI is drawing on it favorably or unfavorably.

That is what makes AI verification the fifth dimension rather than the fifth channel. It does not operate alongside the other four. It operates across all of them simultaneously, either amplifying their effect or diminishing it. When a voter verifies after a TV contact, AI acts on that persuasion. When they verify after a canvass, AI acts on that persuasion. The fifth dimension multiplies every channel the campaign runs.

The exposure arrives through two separate, additive layers.

Layer	How it reaches the voter	Est. congressional volume
Dedicated AI tools ChatGPT, Gemini, Perplexity	Voter deliberately opens an answer engine to research the candidate	~33,00011% × 300K likely voters — CCI 2024
Google AI Overview	Voter Googles candidate name for any reason; AI-generated answer appears before any organic result	~90,000–180,000Est. 30–60% of likely voters conducting at least one candidate name search
Total organic AI encounters		~120,000–210,000

Google AI Overview volume is the author's estimate based on search engine reliance data and the architecture of Google AI Overview as a pre-result layer on informational queries. For a Senate race, multiply by 10–15×.

The Google AI Overview layer is passive. A voter does not decide to "use AI" when they Google a candidate's name. The AI answer appears above every organic result — before the campaign website, before news coverage, before any human-curated source — automatically. Voters reading it often do not know it is AI-generated. They believe they are reading a search result. The 11% figure campaigns reference covers only the intentional layer. The passive layer is likely three to five times larger.

↑ Contents III. AI Adoption and Verification Behavior

52%

of Americans used an AI chatbot in the past week

Edison Research, Infinite Dial 2026. Figure has doubled in two years.

11%

deliberately used AI to research candidates in 2024

Center for Campaign Innovation post-election survey, n=1,500.

1B+

users reached by Google AI Overview globally

Appears above first organic result on all informational queries.

In December 2025, two peer-reviewed studies published simultaneously in Nature and Science — involving real voters in four countries across three national elections — established a finding campaigns should treat as a material fact about the current information environment.¹

The Research Finding A single conversation with a politically oriented answer engine moved likely Trump voters 3.9 points toward Harris on the study's 0–100 candidate-preference scale — an effect roughly four times the measured persuasion effect of a political advertisement in the 2016 and 2020 elections. In Canada and Poland, the same experiments moved opposition voters' attitudes and voting intentions by approximately 10 percentage points. The mechanism is factual density: answer engines generate large numbers of specific claims rapidly, giving them the persuasive weight of a well-briefed expert. When researchers blocked the AI from citing facts, the persuasive effect dropped substantially.

The research also examined how accuracy and persuasiveness relate, and found them in tension: across the models tested, the more persuasive a model was, the less accurate the information it provided. The methods that raised persuasiveness most — post-training and prompting — systematically reduced factual accuracy. Separately, the two U.S. answer engines were not equally effective. The pro-Harris answer engine moved likely Trump voters 3.9 points, while the pro-Trump answer engine moved likely Harris voters 1.51 points, and answer engines instructed to advocate for right-leaning candidates made more inaccurate claims than those advocating for left-leaning candidates in all three countries.²

For campaigns, the implication is narrower than it first appears. Persuasiveness and factual accuracy are distinct properties of a model's output, and the research finds them in tension. Both describe how these systems behave rather than anything a campaign can set. What a campaign can affect is the record those systems draw on, and what it can know is what they are currently saying.

What the Research Does Not Prove The Nature/Science studies used chatbots explicitly instructed to persuade. Organic AI answers — the ones voters encounter when they Google a candidate or ask an unprompted question — are informational, not persuasion-optimized. Their persuasive effect is real but lower than the figures above. A separate Yale study (PNAS Nexus, March 2026) reported that unprompted AI summaries of historical events moved opinions, with modest effects its authors suggest could compound over repeated interaction.⁷ The 4× figure should be read as the upper bound on AI persuasive power; organic exposure operates somewhere below that ceiling.

Evidence Pointing the Other Way Two studies complicate this picture and are worth stating plainly. A UK AI Security Institute study found that political knowledge rose about equally for people researching with answer engines and with search engines, with belief in true information increasing and belief in false information decreasing to the same extent for both.⁸ And a PNAS experiment using a retrieval-grounded bot built on official party platforms raised users' knowledge of party positions by roughly 13 points while barely moving their party evaluations or vote preference.⁹ Neither measures the question this paper is about — how accurately a named candidate's record is represented across engines — but both point the same way. An informational AI encounter is a weaker persuasive event than a persuasion-optimized one, and the reversal rate assumed in Section IV should be read as an upper bound rather than a central estimate.

Favorable answers are doing persuasion work right now. For free. Unmanaged. Most campaigns are getting neither the defensive protection nor the offensive leverage. — Section V · Channels Under EPC

↑ Contents IV. Effective Persuasion Cost

Effective Persuasion Cost extends traditional cost-per-voter analysis to account for the downstream verification step. A $1.50 CPM television impression that moves a voter toward a candidate, followed by an AI encounter that moves them back, has a different effective cost than one that moves a voter and holds. EPC captures the difference — and it is the metric that reveals whether the channel model is working as planned.

The Chain of Loss

For every 1,000 voters a campaign successfully persuades:

Step	Estimated rate	Result per 1,000 persuaded voters
Persuaded voters who go online to search or verify	15–30%	150–300 voters
Those searches that encounter AI content	50–70%	75–210 AI encounters
AI encounters with unfavorable framing or error	15–35%Unknown for unmonitored candidates — this is the number the diagnostic produces	11–74 unfavorable encounters
Voters whose persuasion is reversed	25–50%	3–37 reversals

The wide range reflects genuine uncertainty at each step. The critical variable — the unfavorable answer rate — is unknown for unmonitored candidates. It varies by candidate, platform, and how much structured information exists about the candidate online. Without a diagnostic, this number is a range. With one, it collapses to a specific figure for a specific race.

Value at Risk by Race Tier

A fully-loaded door-knock persuasion costs approximately $570 per durable shifted voter, a figure carried across from turnout field experiments rather than persuasion experiments and used here as an approximation. A television or digital-driven persuasion in a competitive federal race runs $300–600. The table below applies that cost directly to the reversals produced by the chain of loss above.³

Race	Persuasion spend	Persuasion value at risk	As % of spend
Competitive State House	$150K	$400–$5,500	0.3–3.7%
Competitive Congressional	$1.5M	$4,200–$55,000	0.3–3.7%
Competitive Senate / Statewide	$8M	$22,500–$294,000	0.3–3.7%

All figures are the author's estimates derived directly from the chain-of-loss model above. An earlier version of this paper multiplied them by a 2–4× “recovery cost” drawn from the relative persuasive magnitude of answer engines versus advertising; that derivation does not hold and has been removed, which is why these figures are roughly a third of those published previously. The percentage band is identical across tiers because the model is proportional to spend. Actual values depend on the unfavorable answer rate, which is unknown without a diagnostic.

The Key Unknown The unfavorable answer rate — what share of AI responses about an unmonitored candidate contain material errors or unfavorable framing — is the most consequential input in this model. Published research has begun to measure it, though not yet for a US race. An audit of four leading answer engines during two 2026 Irish byelections found wrong answers about who was on the ballot and where to vote, markedly different source selection across systems, and attention concentrated on frontrunners while candidates further down the field were largely ignored.⁶ No equivalent US figure has been published, and the rate varies by candidate, platform, and structured information record. The appropriate response to that uncertainty is not to discount the model; it is to measure. The diagnostic produces this number for a specific candidate. Without it, the value at risk estimates above remain ranges. With it, they become a specific dollar figure for a specific race.

↑ Contents V. Channels Under EPC

The chain-of-loss model is the defensive case — persuasion value lost to unfavorable AI answers. There is an equally important offensive case that most campaigns are missing entirely.

The 120,000–210,000 organic AI encounters in a competitive congressional race include a substantial share of voters the campaign never contacted. They Googled the candidate because they saw a competitor's ad, heard the name in a news story, or received opposition mail. For these voters, the AI encounter is the first impression. The campaign did not pay for it. The candidate either benefits from it or absorbs the damage silently.

Under EPC, channels that perform identically under traditional cost-per-contact metrics can rank very differently once the AI verification step is accounted for. A high-CPM channel whose message is consistently reinforced by accurate, well-sourced AI answers outperforms a low-CPM channel whose message is routinely undermined in verification. The campaigns planning channel mix without EPC data are making budget decisions with a critical input missing.

The Offensive Arithmetic

At a midpoint estimate of 150,000 organic AI encounters in a congressional race: if a candidate's information record is associated with a 10-percentage-point improvement in favorable answer rate, that produces approximately 15,000 additional favorable encounters. At a 2% conversion to durable persuasion, that is 300 additional voters moved — with an equivalent advertising cost of approximately $171,000.

Illustration Only — No Input Here Is Measured Every number in the paragraph above is an assumption chosen to show the scale of the exposure, and none of them is a finding. The 10-percentage-point improvement is assumed, not predicted: Kyanos does not claim that any action produces a given improvement in how an answer engine describes a candidate, and no published research establishes such a figure. The 2% conversion rate from a favorable AI encounter to a durable persuasion is likewise unmeasured, and two studies cited earlier in this paper point below it. The $570 cost per voter is carried across from turnout field experiments. Treat the $171,000 as an order-of-magnitude illustration of what the encounter volume could be worth if these assumptions held, and not as a return anyone should budget against.

What This Is and Is Not This arithmetic does not establish that any specific action will produce any specific persuasion outcome. AI systems are black boxes; their responses cannot be predicted or controlled. It is worth being explicit about what the research does not support. Where the December 2025 studies examined persuasiveness and factual accuracy together, they found the two in tension, with the more persuasive models providing less accurate information. Nothing in that work shows that a better information record makes an AI more persuasive about a candidate. The arithmetic above is conditional on an improvement in favorable answer rate, and the 300-voter figure is illustrative, not a guaranteed return. It is presented to establish the structural scale of the opportunity, not to promise a specific result.

Whether the ambient AI layer is currently working for or against a given candidate is unknown to most campaigns. The diagnostic answers that question. Remediation addresses what the diagnostic finds. Ongoing monitoring tracks for abrupt changes — including the shifts MIT and Carnegie Mellon researchers found occur in AI model responses without public notice.⁴

↑ Contents VI. Monitoring as a Budget Line

Competitive campaigns treat certain intelligence services as standard line items. These services are well established, and their costs are not debated on first principles each cycle:

Service	What it tracks	Typical cost
Earned media monitoringCision, Meltwater, comparable	Press coverage, broadcast mentions, online news	$2,000–$10,000/month
Opposition research	What opponents may say; vulnerability identification	3–8% of total budget
Tracking polls	Voter preference movement over time	$5,000–$15,000/cycle
Digital ad monitoring	Competitor digital creative and spend patterns	$500–$2,000/month

None of these services tracks what AI platforms say about a candidate. That is not a criticism — they predate answer engines becoming primary information intermediaries. It is a gap in coverage at a moment when the December 2025 research suggests AI encounters may be more persuasive per voter than the press coverage those services already track.

Diagnostic — The Entry Point

Every campaign spending meaningful money on federal persuasion should know what AI platforms say about their candidate before significant spend begins. A diagnostic should cover the major platforms where voters encounter AI content: ChatGPT, Google AI Overview, Perplexity, Meta AI, Microsoft Copilot, and Grok. It should test responses across question types — candidate background, policy positions, opponent comparisons, and general searches. A single prompt is not representative of the range of ways voters query a candidate's name.

Rational Diagnostic Budget · Congressional Race

Break-even: prevents how many reversals at $570 each (cost per durable persuasion) 4 reversals

Rational spend range $3,000–$8,000

Comparable reference point One tracking poll cycle

Rational Diagnostic Budget · Senate Race

Break-even: prevents how many reversals at $570 each (cost per durable persuasion) 6 reversals

Rational spend range $6,000–$15,000

As % of persuasion spend <0.2%

Ongoing Monitoring

A one-time diagnostic has a limited shelf life. The MIT and Carnegie Mellon research documented that AI model responses to identical political questions shifted — sometimes abruptly — over the course of the 2024 election cycle, without public notice from model developers. A campaign that runs a diagnostic in August and assumes the results hold through November is taking on unexamined risk. Ongoing monitoring should run at a cadence sufficient to detect changes before they propagate at scale — bi-weekly is a reasonable standard for an active race.

Rational Ongoing Monitoring Budget · Per Month

Congressional (House) $1,000–$3,000/month

Senate / Statewide $2,000–$5,000/month

Comparable reference Earned media monitoring at comparable tier

Recommendation For any competitive federal race spending more than $500,000 on persuasion: commission a diagnostic before significant channel spend begins. Budget $3,000–$15,000 depending on race tier. Evaluate remediation against the specific findings — not as a standard add-on. If the race extends into the fall with active paid media running, add ongoing monitoring at $1,000–$5,000/month. The total investment for a congressional race is unlikely to exceed $25,000, or 1.6% of a $1.5 million persuasion budget.⁵

What Carries That Recommendation, and What Does Not These ranges are offered as market guidance, benchmarked against what campaigns already pay for comparable intelligence. The defensive arithmetic in this paper does not on its own justify them. At the base-case inputs, a congressional program at the top of these ranges costs more than the modeled value at risk, and three of the model's five terms — the post-contact search rate, the reversal rate, and the cost per persuasion — are unmeasured or carried across from a different literature. A reader who accepts only what is measured should read the spend guidance as a market comparison and not as a return calculation.

The Case That Does Not Depend on the Model The argument for measuring does not rest on reversal arithmetic, and it survives every caveat above. Four leading answer engines asked about two 2026 byelections gave wrong answers about who was on the ballot, drew on markedly different sources from one another, and concentrated their attention on frontrunners while largely ignoring candidates further down the field.⁶ For a challenger or a down-ballot candidate, the finding is not unfavorable framing but absence. Analysis of real assistant conversations finds that when people bring politics to an AI they overwhelmingly ask for explanation and analysis — only about 3% of political requests seek advice or a recommendation.¹⁰ They are asking who this person is. And what comes back is shaped by what is crawlable and structured rather than by what is authoritative, in an environment where campaign and party sites remain open to crawlers while independent news organizations increasingly block them.¹¹ None of that requires a persuasion model to matter. It only requires that voters ask, which they do.

Six findings for principals

Campaigns evaluate channels on four dimensions; AI verification is a fifth that multiplies the others. A persuaded voter who turns to an answer engine to verify what they heard either has the persuasion reinforced or undone. The fifth dimension is not additive; it is multiplicative.

A majority of voters now use answer engines, and every Google search for a candidate's name surfaces AI content before organic results. Total exposure to AI-mediated information about candidates is the floor, not the ceiling. Campaigns planning around the 11% explicit-answer engine figure understate actual exposure by three to five times.

Effective Persuasion Cost (EPC) captures the multiplier. A candidate's true cost per persuaded voter must account for downstream answer engine reinforcement or undoing. Channels that look efficient under traditional CPV metrics may be expensive under EPC; channels that look expensive may be efficient.

Under EPC, channels rank differently than under traditional cost-per-contact metrics. High-cost channels with strong AI reinforcement can outperform low-cost channels whose persuasion gets reversed in verification. The ranking inversion is the practical consequence campaigns must plan around.

Answer engine monitoring is not an optional add — it is a budget line. The floor of the monitoring budget is set by what an unmonitored AI encounter costs across the chain of loss. Campaigns spending nothing on monitoring are not saving money; they are accepting an unaccounted tax on every persuasion dollar.

The monitoring budget scales with race tier. Congressional races, statewide races, and Senate cycles have measurably different floors. The race-tier tables in Section IV show the value at risk that sets the floor; the actual budget line should sit above it, not at it.

References

Hause Lin, Gabriela Czarnek, Benjamin Lewis, Joshua P. White, Adam J. Berinsky, Thomas Costello, Gordon Pennycook and David G. Rand, "Persuading voters using human–artificial intelligence dialogues," Nature 648, 394–401, December 4, 2025 (nature.com, subscription). Companion paper: Kobi Hackenburg et al., "The levers of political persuasion with conversational artificial intelligence," Science 390(6777), December 4, 2025 (science.org, subscription; free preprint at arXiv:2507.13919). Accessible summaries: Cornell Chronicle; MIT Technology Review.
Cornell University Chronicle, December 4, 2025. On the study's 0–100 candidate-preference scale, the pro-Harris answer engine moved likely Trump voters 3.9 points; the pro-Trump answer engine moved likely Harris voters 1.51 points. The same release reports that the more persuasive a model was, the less accurate the information it provided, and that models advocating for candidates on the political right made more inaccurate claims in all three countries.
Alan S. Gerber and Gregory A. Huber, "Getting Out the Vote Is Tougher Than You Think," Stanford Social Innovation Review, 2016. Note that this source synthesizes turnout (GOTV) field experiments rather than persuasion experiments; the ~$570 per durable shifted voter figure is derived from Gerber and Green cost-per-vote analyses adjusted for current labor costs, and should be read as an approximation carried across from a turnout literature. Race-tier persuasion spend estimates from Tech for Campaigns, "2024 Digital Ads Report."
Sarah Cen et al., MIT CSAIL, MIT Sloan and MIT LIDS, 2025 (lead author now at Carnegie Mellon University). Nearly daily queries across 12 models on more than 12,000 prompts between July and November 2024, producing over 16 million responses; candidate–trait associations shifted over the period, in some cases tracking news events. MIT CSAIL; reported in Tech Brew, October 6, 2025.
Edison Research with SSRS, February 2026: 52% of Americans 18+ used an AI chatbot weekly (SSRS). Edison's May 2026 wave put the figure at 65%. Center for Campaign Innovation, "2024 Post-Election National Survey," fielded November 3–7, 2024, n=1,500 (11% used ChatGPT or Google Gemini to research candidates). Google AI Overviews: company-disclosed reach, reported at 2 billion monthly users on Alphabet's Q2 2025 earnings call and above 2.5 billion at Google I/O 2026.
Audit of Claude, ChatGPT, Gemini and Grok during the Galway West and Dublin Central byelections, University College Dublin and the University of Strathclyde, May 2026. Approximately 200 election-related questions per system. UCD News; method detail at the Connected_Politics Lab; coverage in The Irish Times.
Matthew Shu et al. (Daniel Karell, senior author), "How latent and prompting biases in AI-generated historical narratives influence opinions," PNAS Nexus 5(3), pgag022, March 2026 (open access); Yale News. n=1,912. The study tested AI-generated accounts of two 20th-century events rather than candidate information; the senior author describes the effects as modest and says they "could compound" with frequent use.
UK AI Security Institute, "Do chatbots inform or misinform voters?," September 30, 2025. Also reports that 32% of answer engine users, or 13% of eligible UK voters, used AI to seek political information during the 2024 election.
Yamil R. Velez, Donald P. Green and Semra Sevi, "Chatbot Voting Advice Applications inform but seldom sway young unaligned voters," PNAS 122(50), December 2025 (pnas.org; free full text at PMC).
Andy Hall and Dan Thompson, "Mapping America's Political Conversations with Claude," Free Systems, August 4, 2026. Analysis of Anthropic Economic Index data covering Claude conversations in April and May 2026. Politics accounted for roughly 0.5% of US conversations, ranking 66th of 189 topic categories; nearly half of political requests produced explanations and just 3% ended in advice or a recommendation. Published on Substack rather than in a peer-reviewed journal.
Andy Hall, "AI Is A Shitty Political Advisor," Free Systems, March 19, 2026, with survey data contributed by Sean Westwood and Justin Grimmer. Hall's stated central finding is that AI political advice "is driven by what is crawlable and structured — not by what is authoritative or accurate," and he documents news organizations restricting AI crawlers while party and campaign sites remain open. Published on Substack rather than in a peer-reviewed journal. The author discloses consulting relationships with a16z crypto, Forum AI and Meta Platforms, and states that his writing is independent of those roles.

About the Author

Ed Forman

Ed Forman is the founder of Raise Presence LLC, which built Kyanos to measure and improve how progressive candidates and causes are represented across AI platforms. About Raise Presence →

Note: AI helped me research and draft this document. I have read every word and it has been processed by my (human) brain — I've checked what AI drafted for me, made the edits I wanted, and I stand behind what it says.