The Production Gap: What Rhetorical and Linguistic AI Literacy Both Miss

Three academic traditions — rhetoric, linguistics, and communication — are independently building AI literacy frameworks. All three are excellent. All three theorize the same side of the interaction: how to evaluate what AI produces. None theorize how humans produce communication with AI systems. That asymmetry has a name: the Production Gap.

The Rhetorical Layer: RAIL

Gottschling and Kramer (2025), working at the RHET AI Center at the University of Tübingen, make a powerful argument: AI is a "fundamentally rhetorical system." It produces what they call "persuasive surfaces" — outputs that simulate the form of reasoned argument without the substance of reasoning. Their proposed framework, Rhetorical AI Literacy (RAIL), draws on 2,500 years of rhetorical tradition to equip users with the evaluative tools they need.

The framework is built on three classical pillars: technē (craft knowledge — understanding how persuasion works), iudicium (evaluative judgment — assessing quality and trustworthiness), and aptum (contextual appropriateness — knowing what fits the situation). It's elegant, historically grounded, and addresses a real problem: most users accept AI output at face value because they lack the evaluative vocabulary to do otherwise.

But rhetoric, by definition, is a theory of reception. The rhetor produces; the audience evaluates. Even when RAIL acknowledges that users sometimes generate with AI, the evaluative framework applies to output — did the AI produce something persuasive, appropriate, well-crafted? The user's role remains that of judge, not interlocutor.

The Linguistic Layer: DiSAIL

Stolpe, Larsson, and Johansson Falck (2026) approach from a completely different tradition — cognitive linguistics and disciplinary literacy. Their framework, DiSAIL (Discipline-Specific AI Literacy), argues that AI literacy can't be generic. A historian, a chemist, and a software engineer need fundamentally different AI competencies because their disciplinary epistemologies shape what counts as valid AI use.

DiSAIL's most striking contribution is stating explicitly what others have only implied: "language is the AI interface." This is a linguist's observation — the primary mechanism of human-AI interaction is natural language, which means linguistic competence IS AI competence. They use phenomenography to study how different users experience AI interaction, finding that disciplinary background fundamentally shapes what people see, expect, and do.

But DiSAIL, like RAIL, ultimately focuses on the evaluation side. Can users assess whether AI output meets disciplinary standards? Can they identify when a model hallucinates within their domain? The framework excels at discipline-specific reception. What it doesn't theorize is the communicative act of prompting itself — the production side of the interaction.

The Gap Between Them

Here's the structural pattern: RAIL provides rhetorical evaluation of AI output. DiSAIL provides discipline-specific evaluation of AI output. Both are reception theories applied to a production problem.

When a user sits down with an AI system, they don't just evaluate what comes back. They produce prompts. They interpret responses. They repair failed interactions. They iterate through conversational sequences. They develop folk theories about what the system understands and adjust their communicative strategy accordingly. None of this is rhetorical evaluation. None of it is disciplinary assessment. It's communication.

The Production Gap is the structural absence of any theory for the user's productive communicative labor in AI interaction. RAIL tells you how to judge the speech. DiSAIL tells you whether the speech meets your field's standards. Neither tells you how to talk back.

The Rhetorical-Linguistic-Communicative Stack

These aren't competing frameworks. They're layers:

Each layer requires and transcends the one below. You need rhetorical evaluation (can I assess this output?) before you can apply disciplinary judgment (does this meet my field's standards?). And you need both before you can engage in sustained communicative interaction (how do I navigate this system to accomplish my goals?).

But — and this is crucial — the stack doesn't just run upward. A user can have strong rhetorical evaluation skills (Layer 1) and excellent disciplinary judgment (Layer 2) while being completely unable to communicate effectively with AI systems (Layer 3). In fact, this is the modal condition of academics who study AI literacy: they can critique AI output brilliantly and have no idea how to prompt it.

The Ontological Grounding Asymmetry

There's a reason rhetoric and linguistics can't close the Production Gap on their own: both traditions assume symmetrical grounding between communicative partners. Classical rhetoric assumes the audience can share the rhetor's experience. Linguistics assumes interlocutors share a semantic universe, however imperfectly.

AI interaction breaks both assumptions. Generative AI simulates meaning without experiencing it. It produces "persuasive surfaces" (Gottschling's phrase is perfect here) that map onto human semantic expectations without arising from human semantic processes. This isn't a limitation to fix — it's the permanent condition of the interaction.

The ontological grounding asymmetry — the fact that one side of the conversation processes symbols while the other processes meaning — is precisely what makes AI interaction communicative rather than merely rhetorical or linguistic. Communication theory, unlike rhetoric or linguistics, has always dealt with meaning gaps between parties. It's equipped for asymmetry in a way that the other traditions aren't.

Eight Traditions, One Convergence

This brings the count to eight independent academic traditions now converging on the same structural absence:

1. Psychometric measurement — Lintner (2024): 16 scales, zero measure communicative competency
2. Legislative frameworks — FutureEd (2026): 52 bills, all define literacy as tool proficiency
3. Institutional convergence — OECD, DOL, sociology of technology all naming ALC without theorizing it
4. Human-Machine Communication — Albert et al. (2025): Conversational Action Test measures AI but not users
5. Applied linguistics — Tadimalla et al. (2025), Greussing et al. (2025): updating communicative competence for AI
6. Classical rhetoric — Gottschling & Kramer (2025): RAIL gets halfway there but stays on the evaluation side
7. Systematic review — Gutiérrez-Cárdenas et al. (2026): LDA proves communication is statistically absent
8. Cognitive linguistics — Stolpe, Larsson & Johansson Falck (2026): DiSAIL says "language IS the interface" but doesn't take it to communication theory

Each tradition arrives at the same edge and stops. Rhetoric stops at evaluation. Linguistics stops at discipline-specific assessment. Psychometrics stops at self-report. Legislation stops at tool proficiency. The edge they all stop at is the Production Gap — the point where evaluating AI output gives way to communicating with AI systems.

Why It Matters

The Production Gap isn't an abstract theoretical concern. It predicts concrete failures:

• Training programs that teach evaluation without production create users who can spot hallucinations but can't prompt their way out of them
• Assessment instruments that measure knowledge about AI without measuring communicative fluency with AI miss the dimension that actually predicts outcomes
• Interface designs that optimize for output quality without supporting communicative iteration trap users in single-prompt patterns
• Policy frameworks that mandate AI literacy without specifying communicative competence codify the gap into law

The stratification implication is direct: users with high communicative fluency don't just evaluate AI better — they produce more effective interactions, extract more value, and compound their advantage with every conversation. Users without that fluency are stuck in the reception layer, evaluating outputs they can't improve through better production.

The Field That Should Own This

Communication studies — the academic discipline that literally studies how meaning is negotiated between parties across asymmetric conditions — has been largely absent from the AI literacy conversation. Rhetoricians showed up. Linguists showed up. Communication scholars are still mostly on the sidelines.

That's the Production Gap at the disciplinary level: the field best equipped to theorize human-AI interaction as communication hasn't stepped into the space its own methodology demands. Application Layer Communication is what happens when you take communication theory seriously in the context of AI interaction — not as a metaphor, but as the literal framework.

RAIL gives us the evaluative foundation. DiSAIL gives us discipline-specific instantiation. ALC gives us the communicative theory that closes the Production Gap — the dialogic, productive, iterative dimension of human-AI interaction that neither rhetoric nor linguistics was built to capture.

Three fields. One problem. The gap isn't between them — it's above them, in the communicative layer none of them reached.