The qualities AI scoring rewards are the same qualities persuasion research has identified for decades as prerequisites for effective communication with human readers.
Abstract. AI presence optimization is typically framed as a technical challenge: structure content so AI retrieval systems can parse it. This note examines a less-discussed implication — that the editorial disciplines AI presence work requires are also disciplines that make communication more effective for human readers. Drawing on cognitive load theory, the elaboration likelihood model, schema theory, and processing fluency research, it argues that prose which fails AI presence scoring on the dimensions of entity clarity, position explicitness, and factual density tends to fail human persuasion for structurally similar reasons. The connection is not coincidental. Both AI retrieval systems and human working memory are constrained information-processing systems that extract meaning most effectively from content that is explicit, consistent, and well-organized. The paper explores that parallel, notes its limits, and argues that AI presence scoring may function as a useful diagnostic for communication quality more broadly — surfacing failures that degrade effectiveness in every channel simultaneously.
When campaigns are introduced to AI presence optimization, the framing is usually technical: AI retrieval systems have specific requirements for parsing and extracting information, and your content either meets them or it doesn't. Get the structured data right. Name the subject consistently. State the positions explicitly. This is often presented as a constraint — something you do in addition to writing for human readers, or potentially in tension with it.
This paper argues that framing is backwards. The qualities AI presence scoring rewards are not arbitrary machine requirements. They are the same qualities that decades of research in cognitive psychology and persuasion have identified as prerequisites for effective human communication. A candidate's website that scores poorly on AI presence — because its key messages are vague, its positions are implied rather than stated, its subject is named inconsistently across pages — is not just poorly calibrated for AI retrieval. It is poorly calibrated for anyone trying to form a clear impression of the candidate.
"The question is not: how do I write for AI? The question is: what does it mean that AI cannot find my candidate's positions on their own website?"
This is not an argument that AI presence work will improve voter persuasion — we have not conducted the experiments that would establish that. It is an argument that the diagnostic function of AI presence scoring may be more valuable than its technical function: it surfaces communication failures that degrade effectiveness in every channel simultaneously.
The Kyanos AI presence scoring rubric evaluates AI responses across four dimensions. Understanding these dimensions precisely is necessary before drawing any connection to human persuasion research.
Factual accuracy — whether the facts AI surfaces about a candidate or organization are correct. Inaccuracies include wrong policy positions, incorrect biographical details, misattributed endorsements. This dimension measures the information environment, not the content directly.
Subject clarity — whether AI responses clearly and consistently identify who the subject is: role, institutional affiliation, geographic scope, and distinguishing characteristics. Failure appears as AI conflating a state senator with a U.S. senator of the same name, or providing accurate information that belongs to a different organization with a similar name.
Position explicitness — whether AI responses state the subject's positions directly, as specific commitments, rather than implying them through organizational affiliation or general framing. Failure appears as "supports environmental protection" when the subject has taken specific legislative positions that are publicly documented.
Factual density — whether the information AI surfaces is grounded in specific, verifiable facts, or consists of general claims without supporting evidence. Failure appears as high-sounding language with no concrete referents: "a lifelong commitment to working families" rather than specific legislation passed, votes cast, or constituents served.
In practice, low AI presence scores are rarely caused by false information published on a campaign website. They are caused by information that is absent, vague, inconsistently named, or buried in prose structures that AI systems cannot easily parse. The website may contain every relevant fact — and still produce thin, inaccurate, or generic AI responses because the facts are not presented in a form that makes them extractable.
This pattern has a direct analogue in human communication research, which we examine in the following sections.
Cognitive load theory distinguishes between intrinsic load (the inherent complexity of the subject matter), germane load (mental effort that produces learning), and extraneous load (effort caused by poor presentation — effort that consumes working memory without contributing to comprehension). Effective communication minimizes extraneous load to preserve cognitive resources for engagement with the content itself.
John Sweller's cognitive load theory was developed in the context of educational instruction, but its implications extend to any form of information communication. Working memory is limited. Every cognitive demand imposed by the structure of a text — inconsistent terminology, implicit references requiring inferential work, information that must be assembled across multiple locations — subtracts from the capacity available for processing the content itself.
AI retrieval systems are, in the relevant sense, low-tolerance readers. They extract information from explicit statements, consistently named entities, and directly articulated relationships. They do not infer. A webpage that refers to the subject as "she" in one paragraph, "the senator" in another, and "the Congresswoman" in a third imposes no meaningful burden on an attentive human reader, who tracks the referent through context. For AI systems, that inconsistency is not merely inconvenient — it increases ambiguity about which facts belong to which subject. AI presence scoring penalizes this: low subject clarity scores frequently trace to exactly this pattern.
The parallel for human readers is less absolute but structurally similar. Inconsistent naming does not block comprehension — but it adds resolution overhead. "The organization" when the reader might mean LCV, or the campaign, or a coalition partner, requires the reader to resolve the ambiguity before processing the substantive claim. Across many such instances, accumulated extraneous load reduces both comprehension and retention.
The same logic applies to implied rather than stated positions. When a candidate's website says "committed to protecting our environment" rather than "voted for the Climate Action Now Act in 2023 and cosponsored HB 4291," the reader must do inferential work to connect the general claim to the candidate's actual record — work that serves no persuasive function and may not be completed at all in a low-attention reading environment. AI scoring penalizes this as low factual density. Persuasion research identifies it as a contributor to message ineffectiveness for the same underlying reason: the cognitive cost of elaboration falls on the reader when it could have been done by the writer.
Prose that fails AI presence scoring on subject clarity and factual density fails because the mental work of resolving ambiguity and assembling scattered information has been left to the reader — whether that reader is an AI retrieval system or a voter scanning a website before casting a ballot. The editorial discipline that addresses one audience addresses both.
The ELM proposes two routes to persuasion. The central route involves careful scrutiny of the arguments in a message — active engagement with the substance of the claim. The peripheral route relies on heuristics and cues (speaker credibility, social proof, affect) without systematic argument processing. Central route processing produces more durable attitude change that is more resistant to counter-persuasion; peripheral route processing produces shallower, less stable attitude change that is more easily reversed.
For central route processing to occur, the argument must be present, clear, and elaborable — the reader must be able to engage with the substance. A message that asserts a general commitment without specifying what that commitment means in practice cannot be elaborated: there is no claim to think about, no evidence to weigh, no conclusion to accept or reject. The reader who encounters "lifelong commitment to working families" has nothing to elaborate. The message goes through the peripheral route at best — or is simply not processed at all.
This is not a failure of voter motivation. It is a failure of message construction. Petty and Cacioppo demonstrated that even high-involvement, highly motivated readers cannot engage with the central route if the argument structure is absent. You cannot deliberate about a claim that was never made.
"Position explicitness is not an AI requirement. It is a prerequisite for the kind of persuasion that produces durable attitude change."
AI presence scoring's position explicitness dimension is, in ELM terms, a test of whether a message is even eligible for central route processing. A candidate whose website implies positions through organizational endorsements and general language is providing voters with peripheral cues, not arguments. That may still influence behavior — peripheral route effects are real — but it produces shallow persuasion, easily reversed by counter-messaging, and unlikely to generate the committed, active support that competitive campaigns require.
The editorial work of making positions explicit — identifying the specific legislation, the specific votes, the specific commitments — is the work of making messages elaborable. It serves AI retrieval for the same reason it serves human persuasion: both require that the claim be present and specific enough to be evaluated.
| AI Scoring Dimension | ELM Parallel | Failure Mode (Both Audiences) |
|---|---|---|
| Position explicitness | Central route eligibility | Cannot be elaborated; routes to peripheral processing or no processing |
| Factual density | Argument quality | Claim without evidence; peripheral cue rather than argument |
| Subject clarity | Source identification | Credibility and accountability cannot attach to an unclear referent |
| Narrative coherence | Message integration | Contradictory signals; elaboration produces confusion rather than persuasion |
Schema theory, developed in cognitive psychology by Bartlett (1932) and extended by Rumelhart (1980) and others, holds that new information is encoded in memory by connecting it to existing knowledge structures. Information that fits cleanly into a schema — that names the relevant entities precisely and locates them in known categories — is encoded more efficiently and recalled more reliably than information that requires the reader to construct new representational frameworks from scratch.
For political candidates, the relevant schemas are well-established: voters have knowledge structures for legislators, for progressive organizations, for environmental advocacy, for healthcare policy. The entity clarity that AI presence scoring rewards — naming the subject consistently, specifying role, district, and institutional affiliation — is also the information that activates the right schema. "LCV's Legislative Scorecard rates Senator X at 92%" connects immediately to existing knowledge; "the Senator's strong record on environmental issues" provides nothing for the schema to anchor to.
Processing fluency research (Reber, Schwarz & Winkielman, 2004; Alter & Oppenheimer, 2009) demonstrates that the ease with which information is processed influences independent judgments of credibility, truth, and likeability. Text that is more fluent to read — consistent terminology, clear structure, prominent placement of key information — is perceived as more credible and more accurate, independent of its actual content. This is not a rational response; it is a cognitive bias. But it is a well-documented one with significant implications for political communication.
Prose that generates low AI presence scores is typically low-fluency prose: it requires resolution work, tracks the subject through implicit references, and buries its claims in subordinate structures. That same low fluency degrades reader perception of the source's credibility and the content's accuracy — not because the content is wrong, but because the processing experience is effortful.
George Lakoff's work on cognitive linguistics and political communication argues that voters understand issues not through isolated facts but through frames — cognitive structures that organize how information is interpreted. Facts that fit an established frame are retained; facts that don't fit the frame are ignored or explained away. Political communication that activates a consistent, coherent frame is more persuasive and more durable than communication that lists facts without framing them. Lakoff argues that this is not a failure of voter rationality but a feature of how conceptual thought works: abstract ideas are understood through concrete metaphorical mappings.
Lakoff's core critique of progressive political communication is that it has historically relied on facts rather than frames — assuming that voters who receive accurate information will update their beliefs accordingly. This approach fails, he argues, because facts without frames don't activate the cognitive structures that make them meaningful or memorable. A campaign website that lists a candidate's legislative accomplishments without framing them within a coherent narrative about what kind of representative this person is provides facts but no interpretive structure.
This critique does not contradict AI presence scoring; it complements it at a different level. AI scoring operates below the level of frame — it asks whether claims are present, explicit, and specific. Lakoff operates above it — he asks whether those claims are organized into a coherent conceptual structure that gives them meaning and makes them stick. The failure modes are different but often co-occurring: a website that is low on factual density is usually also low on coherent framing, because vague language is the default output of communication that has not made the hard choices about what it's actually trying to say.
The Lakoff connection also illuminates why consistency across pages matters for AI presence. AI systems that encounter contradictory framing — a candidate positioned as a consensus-builder on one page and as a fighter against special interests on another — produce confused or generic responses because there is no coherent frame to reproduce. Voters who encounter the same inconsistency form weaker impressions for the same reason Lakoff identifies: the frame never fully activates because the signals keep canceling each other out.
Robert Entman defines framing as "selecting some aspects of perceived reality and making them more salient." His framework focuses on how intermediaries — journalists, editors, media systems — choose which facts to foreground and which to suppress as information travels from source to audience. The frame is not in the original message; it is in the selection decisions that determine which parts of the message survive the journey to the reader.
Entman's model maps directly onto the AI retrieval pipeline. AI systems perform their own salience selection: they extract from source content what is most prominent, most explicit, and most consistently present, and suppress what is buried, implicit, or inconsistent. A campaign website is the source; the AI answer is the intermediary's output. What AI surfaces is not a neutral summary — it is the result of a selection process that rewards exactly what Entman says good communicators should make salient in the first place.
For practitioners, the Entman implication is this: if you do not control what is salient in your own content, the intermediary will. In the media era, that intermediary was a journalist. Now it is also an AI system. The selection is not malicious in either case — it is structural. Content that was never made salient will not be reproduced. Salience is a choice, and PREMISE is where it gets made.
Walter Fisher argues that humans are fundamentally storytelling beings who evaluate communication not primarily through logical argument but through two narrative tests: narrative coherence (does the story hang together internally — are its parts consistent with each other?) and narrative fidelity (does it ring true against the listener's lived experience?). A message that fails coherence or fidelity is rejected regardless of its logical validity.
Fisher's coherence test is directly relevant to AI presence. A candidate's website that describes them as a consensus-builder on one page and a fighter against special interests on another fails narrative coherence — not just for AI systems, which produce confused output when they encounter contradictory characterizations, but for every voter who encounters both pages. The contradiction does not signal nuance. It signals that no one has made a committed choice about who this person is.
Fidelity matters differently. AI systems do not have lived experience to test claims against. But voters do — and a website that claims a legislative record contradicted by public knowledge, or a biography that doesn't match what local reporters have written, will fail the fidelity test for human readers even when AI reproduces it without question. Fisher's framework reminds us that coherence and fidelity are two different problems: AI primarily enforces coherence; voters apply both tests simultaneously.
Hovland, Janis & Kelley (1953) identified expertise and trustworthiness as the core dimensions of source credibility. Subsequent research has repeatedly confirmed that perceived expertise is strongly influenced by the specificity and factual grounding of a communicator's claims: sources who make precise, verifiable, evidence-based assertions are rated higher on expertise than sources who make general, vague, or unsupported claims — regardless of the actual information content.
AI presence scoring's factual density dimension is, in source credibility terms, a proxy for perceived expertise. Prose that grounds every claim in specific, verifiable facts signals epistemic authority. Prose that relies on general assertions signals its opposite. The effect operates for human readers as a credibility judgment, and for AI retrieval systems as an indexing decision: specific, verifiable facts are more parseable and more likely to be reproduced.
The most common AI presence failure we observe in political content is not false information. It is diffuse information — positions stated as values rather than commitments, biographical facts buried in prose narrative rather than prominently structured, key messages present somewhere on the site but not findable as explicit claims.
This failure is not unique to AI retrieval. It appears in every channel simultaneously. A direct mail piece that talks about "fighting for working families" without specifying what legislation the candidate has actually championed is making the same error as a website that produces low position explicitness scores. A digital ad that identifies the candidate by name but not by district, office, or party affiliation is making the same error as a page that produces low subject clarity scores. The AI presence diagnostic surfaces these failures more precisely than most existing communication audits — but it is identifying problems that predate AI retrieval by decades.
It is sometimes assumed that fact-based, policy-focused political communication has a structural advantage in AI presence scoring over emotional, value-laden, or narrative communication. This assumption is partially correct and partially misleading.
What AI presence scoring rewards is not policy-focus per se — it is specificity. Emotional and narrative communication that names its subjects precisely, states its claims explicitly, and grounds assertions in verifiable facts scores well. The failure mode is not emotional language; it is vague language. "A threat to our democracy" is emotionally charged but referentially empty — AI cannot extract who threatens what, or what the speaker means by democracy. "Voted against the Voting Rights Advancement Act in 2021" is equally urgent and completely parseable.
The communication styles that struggle with AI presence scoring are those that have historically substituted rhetorical intensity for argumentative specificity — relying on affect and social proof rather than explicit claim-making. The solution is not to abandon affect but to ground it: specific claims can carry emotional weight. They can also be elaborated by voters, reproduced by AI systems, and remembered after the ad ends.
For campaigns and advocacy organizations operating in the current information environment, this has a practical implication: the editorial review process that AI presence work requires — reading every page of the site against a scoring rubric that asks "what exactly does this say, and about whom?" — is valuable independent of any AI outcome. It is an audit of communication quality that most organizations have not conducted systematically.
James Zaller's Receive-Accept-Sample (RAS) model (1992) argues that voters do not hold stable opinions — they construct them on the spot by sampling from whatever considerations are most accessible in memory at the moment of judgment. Accessibility is driven by recency, repetition, and salience: the considerations most recently encountered, most frequently reinforced, or most emotionally salient are most likely to be sampled. A voter's answer to "who will you vote for?" is shaped heavily by what happens to be at the top of mind when the question is asked.
AI answers are an accessibility-setting mechanism that Zaller's model could not have anticipated but maps onto cleanly. When a voter asks an AI chatbot about a candidate and receives an answer, that answer becomes the most accessible consideration they will bring to any subsequent judgment about the candidate — including how they vote. The AI answer is not just providing information; it is setting the cognitive agenda for everything that follows. Thin, generic, or inaccurate AI answers load the voter's working consideration set with the wrong material at the moment it matters most.
Jonathan Haidt's moral foundations research (The Righteous Mind, 2012) and Drew Westen's work on political emotion (The Political Brain, 2007) are sometimes cited as evidence that fact-based communication is strategically insufficient — that voters respond to moral frameworks and emotional activation more than to argument quality.
Neither finding contradicts the argument in this paper. Haidt and Westen operate at the level of strategic frame choice — which moral emotions to activate, which narrative to run. AI presence scoring operates below that level: it asks whether the chosen frame is executed with clarity and specificity, not which frame was chosen. A campaign running a "protector" frame and a campaign running a "fighter" frame both benefit from making their positions explicit, naming their subject consistently, and grounding their claims in verifiable facts. The emotional register is a strategic decision made in PREMISE; the editorial discipline that AI scoring rewards applies equally to every register.
The more pointed implication of Haidt and Westen for AI presence work: emotional claims that are vague and unsupported ("a lifelong commitment to protecting families") fail the AI scoring rubric and fail Westen's test of emotionally resonant specificity. The fix is the same in both cases: name the threat, name the victim, name the action taken. Specific emotional claims are more persuasive to voters and more parseable by AI simultaneously.
This paper draws on established theoretical frameworks in cognitive psychology and persuasion research. It does not report original experimental findings. We have not conducted controlled studies measuring whether AI-presence-optimized content produces better voter recall, attitude change, or persuasion outcomes than unoptimized content. The connections drawn here are theoretical inferences, not empirical results.
Readers should treat the argument as a plausible hypothesis and a framework for thinking about AI presence work — not as a proven claim about communication effectiveness.
AI-optimized prose is not maximally persuasive prose. Emotional resonance, narrative structure, authentic voice, and relational authenticity remain important for human readers in ways the AI scoring rubric does not capture. The ELM explicitly allows that peripheral route persuasion — affect, credibility cues, social proof — is real and effective, particularly when audience involvement is low. Not every voter wants to elaborate carefully on a candidate's legislative record. Some will respond to how a candidate makes them feel.
The claim here is narrower: prose that fails AI presence scoring on entity clarity, position explicitness, and factual density tends to be less effective for human readers on the same dimensions. Fixing those failures does not produce maximally persuasive content — it produces content eligible for persuasion. The floor is raised. The ceiling depends on factors this framework does not address.
There is also a category of AI presence work — structured data markup, Wikidata entries, llms.txt files — that is purely technical infrastructure, invisible to human readers and irrelevant to the persuasion frameworks discussed here. This paper's argument applies specifically to the prose writing that AI presence work requires, not to the technical backlot work.
AI presence scoring is new. The failures it surfaces are not.
Campaigns have always struggled to make their communications specific, explicit, and grounded. Time pressure, institutional defensiveness, the pull toward aspirational language, the desire not to say anything that could become a negative ad — these pressures have always produced the vague, implicit, rhetorically intense but argumentatively thin communication that AI scoring penalizes. What is new is the precision with which AI scoring identifies these failures, and the urgency with which the information environment now punishes them.
Voters who research candidates through AI-mediated channels receive a summary of what those channels can extract from what campaigns have published. If campaigns publish vague, inconsistently named, position-implicit content, that is what AI surfaces — not because AI is hostile to the candidate, but because there is nothing more specific to find. The voter who asks "what does this candidate believe about healthcare?" and receives a generic response is not getting propaganda. They are getting the logical output of content that was never written to answer the question directly.
"The AI scoring rubric is not asking campaigns to write differently. It is asking them to notice what they have consistently not done."
Cognitive load theory would predict that vague, implicit content is less well-remembered. The elaboration likelihood model would predict it is less persuasive to the motivated voters who matter most in competitive elections. Processing fluency research would predict it reduces perceived credibility. These are independent predictions from independent theoretical traditions, converging on the same conclusion.
Whether those theoretical predictions hold in the specific context of AI-mediated political communication — whether optimizing content for AI presence actually improves voter persuasion outcomes — is a question that can be tested empirically. The experimental design is straightforward: measure AI presence scores before and after content revision, and measure voter recall and attitude formation for matched content sets. We have not run those studies. That is the research agenda this paper is intended to motivate.
Until that work is done, the argument rests on theoretical inference. But theoretical inference from well-established frameworks is not a weak form of evidence. It is what research programs are built from. The null hypothesis here — that the qualities AI scoring rewards are entirely unrelated to human persuasion effectiveness — is the harder claim to defend.
Theoretical references cited
Alter, A.L. & Oppenheimer, D.M. (2009). Uniting the tribes of fluency to form a metacognitive nation. Personality and Social Psychology Review, 13(3), 219–235.
Bartlett, F.C. (1932). Remembering: A Study in Experimental and Social Psychology. Cambridge University Press.
Hovland, C.I., Janis, I.L., & Kelley, H.H. (1953). Communication and Persuasion. Yale University Press.
Lakoff, G. (1996). Moral Politics: How Liberals and Conservatives Think. University of Chicago Press.
Lakoff, G. (2004). Don't Think of an Elephant! Know Your Values and Frame the Debate. Chelsea Green Publishing.
Lakoff, G. & Johnson, M. (1980). Metaphors We Live By. University of Chicago Press.
Petty, R.E. & Cacioppo, J.T. (1986). Communication and Persuasion: Central and Peripheral Routes to Attitude Change. Springer-Verlag.
Reber, R., Schwarz, N., & Winkielman, P. (2004). Processing fluency and aesthetic pleasure: Is beauty in the perceiver's processing experience? Personality and Social Psychology Review, 8(4), 364–382.
Rumelhart, D.E. (1980). Schemata: The building blocks of cognition. In R.J. Spiro et al. (Eds.), Theoretical Issues in Reading Comprehension. Erlbaum.
Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257–285.
Sweller, J. (1994). Cognitive load theory, learning difficulty, and instructional design. Learning and Instruction, 4(4), 295–312.
Entman, R.M. (1993). Framing: Toward clarification of a fractured paradigm. Journal of Communication, 43(4), 51–58.
Fisher, W.R. (1984). Narration as a human communication paradigm: The case of public moral argument. Communication Monographs, 51(1), 1–22.
Fisher, W.R. (1987). Human Communication as Narration: Toward a Philosophy of Reason, Value, and Action. University of South Carolina Press.
Haidt, J. (2012). The Righteous Mind: Why Good People Are Divided by Politics and Religion. Pantheon Books.
Westen, D. (2007). The Political Brain: The Role of Emotion in Deciding the Fate of the Nation. PublicAffairs.
Zaller, J. (1992). The Nature and Origins of Mass Opinion. Cambridge University Press.
About Kyanos
Kyanos measures AI presence for political candidates and advocacy organizations — specifically, what AI systems say about them when voters ask. This paper represents the authors' current thinking and is shared for discussion. It is not peer-reviewed research.