AI Literacy for Citizen Participation Report

Analysis of 976 AI literacy sources—drawn from a corpus of 4,168 this period—reveals a discourse focused on getting people to use AI competently while neglecting their standing to refuse, contest, or govern it. The citizen-as-participant framing appears in fewer than one in ten sources; the overwhelming majority treat literacy as a productivity skill, a thing you acquire to keep up, not a civic capacity you exercise to push back.

The Landscape

Walk through what the word “literacy” is made to carry right now and the center of gravity is unmistakable. The reference texts of the moment are vendor curricula: Microsoft’s Introduction to AI Literacy and its companion AI for educators define competence as fluency with a product line. Alongside them sits a fast-professionalizing skills discourse—prompt engineering reframed as “a new 21st century skill”—that treats the citizen primarily as a more efficient operator of someone else’s system. The framing is not neutral. When the company selling the tool also authors the definition of being literate in it, “literacy” quietly narrows to adoption. What gets marginalized is the other half: knowing when AI is being used on you, and what you are entitled to do about it.

Whose Literacy

The expert-to-citizen ratio in this discourse is lopsided, and the direction of teaching runs one way. Institutions instruct; the public receives. Stanford’s Radical lab racing to study AI’s impact on democracy and UNESCO’s work on inteligencia artificial y desinformación speak with authority about citizens far more often than they transmit anything from them. The rare exceptions are telling: a Latin American study of young people’s own perspectives on AI treats the governed as a source of knowledge rather than a target of correction—and it stands nearly alone. The tide of concern from US parents registers in the press as worry to be managed, not as a constituency to be consulted.

What’s Being Taught

The thematic split is between use and protection, and protection is losing. On the use side: prompting, drafting, “fighting false facts” with AI tools turned against fake news. On the protection side—thinner, but more honest about power—sit the cases that show what citizens are actually up against: an education chatbot company that collapsed and left student data adrift, the L.A. whistleblower account of a chatbot misusing student data, and the surveillance infrastructure quietly embedded in edtech. Notice the asymmetry: the use literature teaches you to operate the tool; the protection literature documents, after the fact, how the tool operated on someone. Almost nothing teaches the move in between—how to see the data extraction while it is happening and decline.

What’s Missing

The competency the discourse most reliably omits is contestation. There is abundant material on evaluating AI outputs—why students and the rest of us fall for AI misinformation—and almost none on contesting AI decisions. When Gothenburg automated school placement or the EU turned to a US-built system to rank its own candidates, the literacy a citizen needed was procedural—appeal, audit, consent—not interpretive. That literacy is barely written. The underserved population is everyone outside the classroom: the discourse still imagines the learner as a student, when the person most in need of it is the applicant, the patient, the welfare claimant being ranked by a model they were never taught they could question.

Core Tensions

The concept of “AI literacy” conceals genuine tensions about what citizens need to know and why. Strip away the consensus rhetoric and the term fractures along at least six fault lines—and the most fundamental is the oldest one: technical skill versus critical understanding. One camp treats literacy as competence, a stack of usable skills. The other treats it as judgment, the capacity to evaluate what a system is doing to you. This isn’t a knowledge gap to fill. It’s contested terrain, and the contest has a winner already favored by the people funding the curricula.

Watch the move. When Microsoft offers an “Introduction to AI Literacy” path, the deliverable is fluency—learning to prompt, to deploy, to integrate Introduction to AI Literacy - Training | Microsoft Learn. The same vendor logic reframes literacy as a “21st century skill,” with prompt engineering elevated to a core competency citizens must acquire to remain employable Frontiers | Prompt engineering as a new 21st century skill. Notice what this definition quietly forecloses: the literate citizen is one who uses the tool well, not one who asks whether the tool should exist, who owns it, or what it extracts. Competence becomes the ceiling. Critique never enters the syllabus, because the syllabus is written by the seller.

The second tension—individual competency versus collective governance—is where the stakes for citizens turn concrete. The dominant framing makes literacy a personal acquisition: you, alone, learn to spot a deepfake, verify a claim, manage your exposure. But the harms that actually require literacy are structural and arrive without your participation. When Göteborg’s schools used an algorithm to assign pupils and produced outcomes families couldn’t contest, no amount of individual skill helped; the injustice was administrative, decided above the level of the person affected Un cas pratique d’injustice algorithmique. When the EU turned to a U.S.-developed system to rank its own candidates, the relevant literacy wasn’t individual—it was the public’s capacity to interrogate a procurement decision L’UE se tourne vers une IA développée aux États-Unis. Selling literacy as self-improvement offloads onto individuals a problem that is, by construction, collective.

Third: protection from versus empowerment with. The empowerment story says: here are the tools, become powerful. The protection story keeps surfacing the reasons that’s naïve. When an education chatbot company collapsed, the student data it had accumulated simply vanished into legal limbo, beyond any user’s control An Education Chatbot Company Collapsed. Where Did the Student Data Go; a Los Angeles whistleblower described that same data being misused as the vendor crumbled Whistleblower: L.A. Schools’ Chatbot Misused Student Data. The surveillance infrastructure beneath these “empowering” tools is itself the thing citizens most need to read Unmasking EdTech’s Surveillance Infrastructure in the Age of AI. A literacy that only teaches use, never refusal, has chosen a side.

Which brings us to the metaphors, because they do the choosing in advance. Across the discourse, AI is overwhelmingly figured as a Tool (304 instances)—a neutral instrument you pick up, and whose effects are entirely a function of your competence. A distant second is Threat (52), the framing that animates the misinformation literacy agenda, where AI is the flood and you are taught to build a levee AI shows promise in the fight against fake news; What Makes Students (and the Rest of Us) Fall for AI Misinformation?. What barely registers is Partner (7 instances)—and that absence is the tell. The Tool frame says the system has no agency; responsibility is yours. The Partner frame would demand reciprocity, accountability, a relationship in which the system and its makers owe you something. UNESCO’s framing of AI and disinformation gestures toward this—treating the problem as institutional rather than merely personal Inteligencia artificial y desinformación. Stanford’s democracy lab is racing to study exactly that systemic level A ‘Radical’ Lab Races to Study AI’s Impact on Democracy.

Here is the test a citizen can apply without any technical training: when someone offers you “AI literacy,” ask whether it makes you better at using the tool, or better at holding its makers to account. Most of what’s on offer does the first and calls it the second.

Power & Agency Analysis

Power in AI literacy operates through definition: whoever decides what citizens “need to know” also decides what stays invisible. Run the week’s evidence through that filter and a pattern emerges in how AI agency gets assigned. Across the corpus, the dominant framing casts AI as a tool — the metaphor surfaces 304 times, roughly six times more often than the threat framing’s 52. That ratio is not neutral. It quietly teaches citizens that the relevant question is how they wield the system, not what the system does to them, and certainly not who built it to do that.

How AI is portrayed

Watch the grammar of agency. When a hiring algorithm or a school-assignment system produces a harmful outcome, the sentence almost always loses its subject. Göteborg’s automated pupil-allocation system did not “decide” to disadvantage children in the coverage — it was described as having done so, which conveniently leaves the officials who procured and deployed it offstage Un cas pratique d’injustice algorithmique. The same evacuation of human agency appears when the EU adopts a US-built system to rank job candidates: the model “classes,” the institution merely “turns to it” L’UE se tourne vers une IA développée aux États-Unis. For citizens, the lesson buried in this syntax is corrosive: outcomes feel weatherlike, beyond appeal. A literate reader learns to restore the missing subject — to ask who chose this system, who profits, who can be sued — because agency that is grammatically erased is also politically unaccountable.

Who defines literacy

Notice who is teaching you what AI literacy means. The most authoritative-looking curricula in the corpus are published by the companies selling the product: Microsoft’s “Introduction to AI Literacy” and “AI for educators” frame literacy as fluency with tools you are simultaneously being onboarded to buy Introduction to AI Literacy, AI for educators. There is real instruction in those modules. There is also a commercial interest in defining “literate” as “comfortable with our defaults” rather than “able to refuse, audit, or regulate.” Compare that to the civic framing in UNESCO’s work on AI and disinformation, which treats literacy as a defense of the public sphere rather than a sales funnel Inteligencia artificial y desinformación. The gap between those two definitions is the power struggle. The vendor wants competent users; the citizen needs competent skeptics.

What metaphors teach

The “tool” metaphor’s dominance is the most consequential framing of all, because a tool implies a neutral instrument fully under its holder’s control — a hammer does not harvest your data or collapse mid-task and lose it. Yet that is precisely what happened when an education chatbot company folded and student records simply went missing An Education Chatbot Company Collapsed. Where Did the Student Data Go?, and what whistleblowers described when a deployed chatbot quietly repurposed the data it collected Whistleblower: L.A. Schools’ Chatbot Misused Student Data. “Tool” obscures the standing surveillance infrastructure beneath the interface Unmasking EdTech’s Surveillance Infrastructure. Meanwhile the rarer “threat” metaphor enables its own distortion — a free-floating dread that, around deepfakes, becomes a liar’s dividend: once people believe anything could be fake, the powerful dismiss real evidence as fabricated Watch out for false claims of deepfakes. Critical metaphor literacy means holding both at once: neither neutral instrument nor autonomous menace, but a system someone owns and operates.

Citizen agency

So what power do citizens actually hold? More than the tool framing suggests, less than the marketing implies. The leverage points are collective, not individual. Documented failure does political work — flawed AI detectors got named and challenged precisely because reporting exposed them Colleges pay millions for AI detectors that are flawed. The captioning lawsuits show accountability arriving through courts rather than user settings The Captioning Lawsuit Cluster. And research consistently finds that what protects people from AI misinformation is not technical wizardry but the older civic muscle of asking who benefits What Makes Students (and the Rest of Us) Fall for AI Misinformation?. Knowledge here is not a productivity skill. It is a form of standing — the capacity to put the human subject back into the sentence, and then hold that subject responsible.

Failure Genealogy

Literacy failures differ from technical failures: they occur when citizens misunderstand what AI is, what it’s doing, or how to evaluate it. The system works exactly as designed; the person using it does not understand the design. Our analysis of this week’s 4,168 sources documents a recurring set of patterns in how that understanding breaks down — and they cluster not around ignorance of code, but around misplaced trust, missed detection, and unwitting exposure.

Where understanding fails

The first failure is calibration. People either trust AI too much or, having been burned, dismiss it wholesale — and both are failures of the same underlying skill. The over-trust case is well documented: research on why students and the rest of us fall for AI-generated misinformation finds that fluent, confident output reads as credible regardless of whether it is true What Makes Students (and the Rest of Us) Fall for AI Misinformation?. The under-trust case is subtler and arguably more corrosive to civic life: once people know convincing fakes exist, they begin dismissing authentic evidence as fabricated. Brookings calls this the “liar’s dividend” — the payoff a bad actor collects simply because deepfakes are now plausible enough that real footage can be waved away Watch out for false claims of deepfakes, and actual deepfakes this election year.

Detection is the second failure. Citizens overestimate their ability to spot synthetic content, and institutions overestimate the machines built to do it for them. Forensic analysis of the 2024 US election found that the most consequential manipulations were often cruder than expected, and that confident human eyeballing was no defense Deepfakes in the 2024 US Presidential Election. The institutional version is worse: colleges have spent millions on AI detectors that are demonstrably unreliable Colleges pay millions for AI detectors that are flawed, and the tools keep getting deployed against people even after their failure rates are known AI detection tools are unreliable. Teachers are using them anyway. A literacy gap, in other words, is not only an individual condition — organizations have it too.

What assumptions mislead

Underneath these failures sit a few load-bearing assumptions. The first is that an AI system answers your question rather than predicting a plausible-sounding string — which is why fluency gets mistaken for accuracy. The second is that “free” or “school-issued” software is therefore safe. It is not. The collapse of an education chatbot company left an open question no one could answer: where did all the student data go? An Education Chatbot Company Collapsed. Where Did the Student Data Go? The L.A. schools case made the pattern explicit, with a whistleblower describing how a chatbot misused student data as the vendor crumbled Whistleblower: L.A. Schools’ Chatbot Misused Student Data. The assumption that an institution vetted the tool on your behalf is precisely the assumption that surveillance-funded edtech relies on Unmasking EdTech’s Surveillance Infrastructure in the Age of AI.

Consequences of gaps

The costs land unevenly. When an automated system assigns children to schools in Göteborg, the family that cannot read the algorithm cannot contest the outcome Un cas pratique d’injustice algorithmique. When the EU adopts US-built software to rank candidates, the people being ranked rarely know the criteria L’UE se tourne vers une IA développée aux États-Unis. The collective cost is a polluted information commons — UNESCO frames AI-driven disinformation as a structural threat to public reasoning, not a series of isolated hoaxes Inteligencia artificial y desinformación.

What would help

The literacy that prevents these failures is not technical fluency but a habit of asking three questions: what is this output, who profits from my believing it, and what happens to what I just typed in? Tools exist to support the first question — AI is showing real promise at flagging fabricated claims AI shows promise in the fight against fake news. But be honest about the limit: no curriculum closes a gap that a vendor profits from keeping open. Detection literacy helps individuals; it does not, by itself, fix the incentives that produce the flood.

Evidence Synthesis

Synthesizing 4,168 sources from this week, the evidence on AI literacy points to a finding that should reframe how citizens think about the skill: the literacy that matters for civic life is not the ability to use AI but the ability to recognize when AI is being used on you. This goes beyond technical competence. Where prior coverage framed AI literacy as a balance between workforce readiness and ethical engagement, the week’s evidence shifts the ground — the most consequential literacy gap is now defensive, not vocational, and it shows up wherever an algorithm sorts, scores, or speaks to a citizen who never consented to the arrangement.

What the evidence shows

The convergent finding across sources is that exposure does not produce judgment. People fall for AI-generated misinformation not from ignorance of technology but from the ordinary cognitive shortcuts that AI now exploits at scale — fluency, confidence, and emotional resonance read as credibility What Makes Students (and the Rest of Us) Fall for AI Misinformation?. A systematic scoping review of generative AI and misinformation confirms that the threat is structural rather than incidental: synthetic content degrades the shared evidentiary baseline citizens use to reason together GenAI and misinformation in education: a systematic scoping review - Springer. What works, where anything works, is calibrated skepticism — and AI detection tools are precisely the wrong instrument for building it. Colleges have paid millions for detectors that misfire Colleges pay millions for AI detectors that are flawed - CalMatters, and the tools remain unreliable even as institutions deploy them anyway AI detection tools are unreliable. Teachers are using them anyway : NPR. Outsourcing judgment to a machine that claims to detect machines is the opposite of literacy.

Contested terrain

“Literacy” stays contested because its proponents cannot agree on what it defends against. One camp treats it as a market skill — prompt engineering as the new twenty-first-century competence Frontiers | Prompt engineering as a new 21st century skill, echoed in vendor curricula that frame fluency as the goal Introduction to AI Literacy - Training | Microsoft Learn. Another treats it as civic defense against deepfakes and the subtler harm of the “liar’s dividend,” where the mere existence of synthetic media lets bad actors dismiss real evidence as fake Watch out for false claims of deepfakes, and actual deepfakes this election year - Brookings. The Slovak case shows the dividend operating in a real election. These are not the same skill, and conflating them lets vendors sell fluency while the civic deficit widens.

Across domains

Tool-specific literacy means understanding what a system does to you when you are not the customer. When the EU adopts US-developed AI to rank job candidates L’UE se tourne vers une IA développée aux États-Unis pour classer ses candidats, or Göteborg automates pupil assignment into demonstrable injustice Un cas pratique d’injustice algorithmique - l’attribution automatisée des élèves, the literacy that matters is knowing such systems exist and can be challenged. The equity dimension is sharpest in accessibility: AI agents now read websites through the accessibility tree, and that pipeline is breaking The Accessibility Tree Is How AI Agents Read Your Site & It’s Breaking — meaning the citizens most dependent on machine mediation bear the least-visible failures AI and Accessibility: Incredible Potential, Inconvenient Questions.

Gaps and uncertainty

What we do not know is whether literacy interventions durably change behavior, or merely test performance. We lack longitudinal evidence that any curriculum reduces susceptibility to synthetic persuasion in the wild, and we have almost no data on populations outside formal instruction — exactly the citizens the participation question is about.

For citizens

Individually: treat fluency and confidence as warnings, not credentials; verify provenance before sharing; assume any consumer chatbot logs you Parents’ Ultimate Guide to AI Chatbots and Mental Health Support. Collectively: the systems sorting citizens will not be fixed by smarter individuals — they require disclosure mandates and contestation rights that no amount of personal literacy can substitute for.