AI-assisted teaching is the reform, not the threat
Shift assessment from answer-hunting to reasoning and disclosure
Train every teacher and set simple norms so AI boosts equity and learning

A single statistic should reframe the debate: in 2024, one in four U.S. teenagers used ChatGPT for schoolwork, a figure double that of 2023. The trend is steep, and it is still early. Most teens have not yet used it. But the growth is clear. AI is now a typical study habit for a significant minority, rather than a rare case. We can try to block it or oversee it. Or we can accept a simple truth: technology has already changed classrooms. The question is not how to fix the tool; it’s how to adjust teaching around it. This is at the heart of AI-assisted teaching. We need to set expectations that foster authentic learning, even when answers are just a prompt away. The goal is both straightforward and challenging: create classrooms where the easy path is not the learning path. The easy path should only help students do the hard work more effectively.

Reframing the problem: AI-assisted teaching replaces answer-hunting with thinking

The introduction of chatbots has revealed a problem with assessments rather than a cheating issue. We have asked students to gather facts and repackage them. Machines can do that much faster. AI-assisted teaching requires a shift in instruction. Tasks must encourage students to evaluate sources, compare different perspectives, test claims, and show their reasoning within time constraints. This is not an abstract ideal; it is the only way to keep learning central when AI provides fluent but shallow text. Surveys indicate that teens feel AI is suitable for research but not for writing essays or solving math problems. This instinct is healthy. We should respond with assignments that demand human judgment and original thinking. When a prompt can yield an outline, the assignment must require selecting, defending, and revising that outline using course concepts and cited evidence. Here, the tool becomes the starting point, not the endpoint.

The policy context has also moved in this direction. UNESCO’s global guidance encourages systems to focus on human agency, build teacher capacity, and create age-appropriate guidelines. This advice matters because teaching—not software—determines whether AI is beneficial or harmful. Countries and districts are beginning to respond. Training is increasing rapidly, but it is uneven and incomplete. In the U.S., the proportion of districts offering AI training jumped from about one quarter in fall 2023 to nearly one half in fall 2024, with plans indicating further growth. Teacher surveys show that usage is still concentrated in planning and content drafting, not in deep data use for learning. This is a story about capacity, not fate. Suppose we invest in teachers and redesign assessments. In that case, AI-assisted teaching can turn a potential risk into a tool for greater rigor.

What the data already shows: productivity gains and learning effects

The early evidence regarding teacher workload is clear. When teachers use well-guided prompts, lesson preparation time decreases without affecting quality. A large trial in England found that providing a simple guide along with ChatGPT reduced planning time by about a third. That level of gain is rare in education. It frees up hours for feedback, tutoring, and class discussions—elements of teaching that machines cannot replicate. U.S. surveys reflect this trend. Many teachers report using AI weekly, and those who do so regularly save several hours each week. These savings are not just conveniences; they create time for instruction that focuses on analysis and argument rather than recall. AI-assisted teaching starts by giving teachers back their time.

Learning effects are beginning to show, especially in writing. Randomized trials indicate that structured AI feedback can enhance revision quality for college students and language learners. Improvements happen when students use AI to identify problems and refine drafts, rather than skipping the drafting process altogether. This is the actionable takeaway for classrooms: make AI a critic, not a ghostwriter. When the task requires students to summarize complex readings, compare arguments, and defend a claim with quotations, AI can offer style advice or highlight weak topic sentences. Students must still engage in critical thinking. This approach aligns with AI-assisted teaching principles. The machine excels at providing rapid feedback and minor edits. At the same time, the student tackles tasks that require human reasoning, judgment, and decision-making.

The adoption gap: training, equity, and the new baseline for rigor

Adoption varies across systems and subjects. The OECD’s recent international teacher survey shows that, on average, about one in three teachers use AI, with leaders like Singapore and the UAE approaching three-quarters. In other areas, teacher usage remains below one in five. Within countries, use is primarily focused on lesson drafting and content adaptation, with limited application for formative analytics. This pattern is predictable and solvable. Training drives usage. Where teachers receive structured professional development, adoption and confidence grow. Equity must receive the same attention. If only well-resourced schools incorporate AI-assisted teaching, the existing gaps in feedback and enrichment will widen. The policy expectation should be clear: every teacher should understand how to use AI to create better lesson plans, develop richer examples, and design assessments that value reasoning over searching.

Student use is already prevalent. Global student surveys in 2024 revealed high rates of AI usage for academic support, and national snapshots confirm this trend. In the U.S., teen use of ChatGPT for schoolwork doubled in a year. Studies in the UK report that the majority use AI for assistance with essays or explanations. At the same time, students express concerns that AI might undermine study habits if misused. This tension should shape policy decisions. Banning and detection will not scale; teaching and designing must. Schools should clarify norms for acceptable uses of brainstorming, outlining, and language support; disclosure rules; and penalties for presenting AI-generated text as original work. Most importantly, we must raise the rigor of assignments. When tasks require live problem-solving, oral defenses, and iterative feedback on drafts, quick answers hold little value. AI-assisted teaching redirects student energy back to learning.

What to change now: curriculum, assessment, and teacher craft

Curriculum should include AI as a critical literacy. Students need to learn prompt design, source verification, bias detection, and the distinction between fluent language and genuine reasoning. This is not an optional unit; it belongs in core subjects. In writing courses, AI can model alternative thesis frameworks and illustrate counterarguments. In science, it can suggest experimental variations and help design data tables. In language classes, it can provide feedback focused on form while teachers emphasize meaning and interaction. The system's role is to assist; the student's role is to create. Institutions should also adopt standard disclosure policies so students can indicate where and how they used AI. When rules are straightforward and public, misuse decreases and AI-assisted teaching becomes standard practice rather than a loophole. Thoughtful national guidance already points in this direction; local leaders should implement it promptly.

Assessment must shift from recall to reasoning. This means more supervised, classroom-based writing, more oral defenses, and more multi-step tasks that require judgment at each stage. It also involves using AI to create better rubrics and examples. Teachers can ask AI to draft three sample answers—basic, proficient, and advanced—and then refine them to meet standards. Students learn what quality looks like and why it matters. Districts should update honor codes to include AI disclosure and provide clear examples of allowed and prohibited uses. Training remains crucial. As district surveys show, when leaders invest in ongoing professional learning, teachers’ use of AI becomes safer and more effective. We need the same investment that accompanied the transition from chalkboards to projectors to learning management systems—only faster. That is what AI-assisted teaching requires.

Anticipating the critiques—and answering them with design

Critique one argues that AI undermines originality. It can do so if assignments prioritize speed over thought. The solution is not to ban tools but to redesign assignments. When tasks demand unique evidence from class experiments, local data, or in-person observations, generic outputs will not meet expectations. Trials in writing support this point. AI feedback enhances revision quality when students write and then revise. It offers little support when students skip the draft. The solution lies in design, not in fear. We can establish process checkpoints, require annotated drafts, and include reflection as part of the grading. The outcome is more writing, not less, with AI serving as a coach. That is AI-assisted teaching in action.

Critique two warns that AI could worsen inequity. It might, if only some teachers and students knew how to use it effectively. However, policy can bridge the gap. Provide device access and high-quality AI tools in classrooms. Train all teachers, not just volunteers. Share prompt banks, model lessons, and example tasks created within districts. Current international and national guidelines emphasize user-focused methods and teacher development as key elements for safe AI adoption. Systems that prioritize training report higher usage and greater confidence. Equity is not complex; it is about resources, standards, and time. If we build the necessary supports, AI-assisted teaching can help close gaps by providing all students with quick feedback and more opportunities to engage in real thinking.

We should revisit that initial statistic and consider it carefully. A quarter of teens using ChatGPT for schoolwork is not, in itself, a crisis. It is a signal. This signal indicates that students are already studying in an environment where humans and machines coexist. The only real risk is pretending otherwise. If we improve our teaching—if we elevate the expectations for tasks, normalize disclosure, train teachers widely, and incorporate AI literacy into core subjects—we can keep learning at the forefront. We can ensure that speed serves depth. We can enable convenience to support judgment. This is the promise of AI-assisted teaching, and it is also a necessity. The tools will continue to improve. The only way to stay ahead is to teach what they cannot do: reason with evidence, explain under pressure, and adapt ideas to new situations. Let us create classrooms where these skills become the norm. The statistic will keep rising. Our standards should rise even faster.

The views expressed in this article are those of the author(s) and do not necessarily reflect the official position of the Swiss Institute of Artificial Intelligence (SIAI) or its affiliates.

References

Education Endowment Foundation (2024). Teachers using ChatGPT – alongside a guide to support them to use it effectively – can cut lesson planning time by over 30 per cent. (press release, Dec. 12, 2024).
OECD (2025). Teaching for today’s world: Results from TALIS 2024.
Pew Research Center (2025). About a quarter of U.S. teens have used ChatGPT for schoolwork—double the share in 2023. (Jan. 15, 2025).
RAND Corporation (2024). Using Artificial Intelligence Tools in K–12 Classrooms.
RAND Corporation (2025). More Districts Are Training Teachers on Artificial Intelligence.
UNESCO (2023; updated 2025). Guidance for generative AI in education and research.
Zhang, K. (2025). Enhancing Critical Writing Through AI Feedback (randomized controlled trial). Humanities & Social Sciences Communications.

Deepfake NIL licensing will surge as content costs collapse
Schools must use contracts and authenticity tech to protect communities
Provenance and consent rebuild trust when faces become licensable assets

Only three in ten UK adults say they feel confident in spotting whether AI made an image, voice clip, or video. This finding from the country's communications regulator highlights a critical gap in public life: the difference between what we see and what we can trust. Meanwhile, the influencer market is projected to reach about $32.6 billion in 2025, with content volume growing and production costs decreasing. Put these two facts together, and a simple truth appears. Deepfake NIL licensing—the paid use of a person’s name, image, and likeness through synthetic media—will grow faster than our defenses unless schools, universities, and education ministries take action. This issue goes beyond celebrity appearances. It affects students, teachers, and institutions that now compete for attention like brands and creators do. Classrooms have become studios, and authenticity is now part of the curriculum.

Deepfake NIL licensing is changing the cost curve

The reason for this change is technological. In 2025, OpenAI released Sora 2, a model that generates hyper-realistic video with synchronized audio and precise control. OpenAI’s launch materials highlight improved physical realism. Major tech outlets report that typical consumer outputs run for tens of seconds per clip. The app now includes social features designed for quick remixing and cameo-style participation. In short, anyone can create convincing short videos at scale using someone else’s licensed (or unlicensed) face and voice. When credible tools drop the cost of production to near zero, licensing becomes the business model. The market will follow the path of least resistance: pay creators or their agents a fee for permissioned “digital doubles” and then mass-produce content featuring them. This shift is already visible in pilot platforms that let users record a brief head movement and voice sample for cameo use, complete with revocable permissions and embedded watermarks. However, those marks don’t always hold up against determined adversaries.

Policy is progressing, but not quickly enough. The EU’s AI Act mandates that AI-generated or AI-modified media—deepfakes—must be clearly labeled. YouTube now requires creators to disclose realistic synthetic media. Meta and TikTok have introduced labeling and, importantly, support for Content Credentials (the C2PA provenance standard) to ensure that authentic data travels with files across platforms. However, these disclosures rely on adoption, which is inconsistent. Recent studies indicate that fewer than half of creators use adequate watermarking, and deepfake labeling remains uncommon across the ecosystem. This gap—in rules, enforcement, and infrastructure—provides an opportunity for deepfake NIL licensing to thrive in gray areas. At the same time, misuse spreads through the same channels.

The economics are compelling. Goldman Sachs estimates that the broader creator economy could reach $480 billion by 2027. Influencer marketing alone has jumped from $24 billion in 2024 to an estimated $32.55 billion in 2025. As volume increases, brands will treat faces as licensable assets rather than items to juggle. Unions have noticed this trend. The SAG-AFTRA agreements for 2024–2025 set minimum standards for consent, compensation, and control over digital replicas—including voice—creating a framework that education and youth sports can follow. However, the same time period also revealed how fragile these safeguards can be. Investigations and watchdogs have reported extensive non-consensual deepfake pornography, rising synthetic-fraud rates, and the ease with which realistic fakes can bypass naive checks. Deepfake NIL licensing will flourish in this environment.

Deepfake NIL licensing and campus policy

Education currently sits at a crossroads where NIL, influence, and duty of care intersect. In US college sports, a court-supervised settlement now allows schools to share up to $20.5 million each year directly with athletes for NIL starting in the 2025–26 season. At the same time, the collegiate NIL market itself is often valued at around $1.5 billion, while the broader influencer economy that students engage with is much larger. Many students are already micro-influencers, and many departments now function like media shops. In this climate, deepfake NIL licensing becomes a campus issue because licensing—and mis-licensing—can occur on school accounts, in team agreements, and within courses that require students to create and publish. The ethical guidelines on campus must keep up with the financial realities.

The risk profile is real. Channel 4 News, covered by The Guardian, reported that nearly 4,000 celebrities had been targeted by deepfake porn, which received hundreds of millions of views over just a few months. Teen creators and student-athletes face the same tools and risks, often without protection. Regulators and platforms stress labeling and disclosure, but confidence remains low: only 30% of UK adults feel they can judge whether media was AI-generated. At the same time, surveys in the US show that large majorities want labels because they cannot trust their ability to detect such content. Schools cannot pass this responsibility to platforms. Athletic departments, communications offices, and student services need consent processes, model releases that clearly cover digital doubles, and default provenance tagging on official outputs. This is necessary to protect students who license their likeness as well as those who never intended to do so.

Deepfake NIL licensing needs an authenticity infrastructure

Rules are essential, but infrastructure is crucial. The C2PA “Content Credentials” standard provides media files with a verifiable provenance trail. Major platforms and manufacturers have begun to adopt it, from TikTok’s automatic labeling of credentialed content to camera makers that include credentials in their devices. The Content Authenticity Initiative reports thousands of organizations are now involved. This is significant for education because provenance can be integrated into workflows: a journalism school, a design program, or a district communications team can require Content Credentials during creation and maintain them through editing and publishing. It is not a complete solution; it is a safety measure. Where labels are present, trust can grow. Where they are absent, disclosure rules become optional.

Yet adoption is inconsistent. A 2025 study found that only about 38% of popular image generators use adequate watermarking, and only about 18% employ meaningful deepfake labeling. Meanwhile, platform policies remain fluid. Meta announced labeling in early 2024 and later adjusted its placement and scope. YouTube’s disclosure requirement is now active, and TikTok has expanded auto-labels for media with credentials. Schools should treat these as baselines rather than guarantees. The transparency provisions in the EU AI Act offer a stronger framework and will influence global practices, but institutions cannot wait for uniform enforcement to arrive.

What educators should do now about deepfake NIL licensing

Start with contracts instead of code. Every school involved with student media or athletics should update release forms to include explicit consent, scope, duration, revocation, and compensation for digital replicas, following the principles of transparency, consent, compensation, and control present in recent SAG-AFTRA deals. The form must address voice clones, face swaps, and synthetic narration and specify whether training uses are permitted. For minors, obtain consent from a parent or guardian. For international programs, align forms with the EU AI Act’s disclosure requirements to prevent cross-border confusion. Then support the paperwork with technology: require Content Credentials on all official outputs, enable automatic credential preservation in editing tools, and keep a registry of authorized digital doubles so staff can verify if a “cameo” is licensed.

The curriculum also needs an upgrade. Media and information literacy should make checking provenance and identifying platform-specific labels standard practices. UNESCO warns that two-thirds of creators do not systematically fact-check before posting; PISA materials show how weak digital practices impact learning. Teach students how to interpret a Content Credentials panel, how to disclose synthetic edits, and how to file a takedown when a deepfake targets them. Develop lab exercises around real platform policies on YouTube, Meta, and TikTok. Teach detection skills, but do not base trust solely on visual judgment. The evidence is clear: the public’s ability to identify AI is limited, and in some scenarios, negligible. Trust should come from reliable processes rather than guesswork. Deepfake NIL licensing will hinge on how consent is gathered, how provenance is maintained, and how misuse is managed before it becomes a significant issue.

The existing gap remains a significant concern. Only 30% of adults feel confident identifying AI-generated media, even as the creator and influencer economies fuel the content boom. This serves as the backdrop for deepfake NIL licensing—a system that will monetize faces on a large scale, reward permissioned clones, and entice bad actors to bypass consent. Education cannot remain on the sidelines. It is the place where young creators learn the rules of the attention economy. It is where student-athletes negotiate NIL for the first time. It is where public trust in knowledge is either rebuilt or lost. The call to action is clear: align releases with the realities of digital replicas, require provenance by default, teach disclosure and verification as essential skills, and prepare for the day a viral fake targets your institution. If we do this, schools will not just keep pace with the platforms. They will set the standard for a public space where licensing a face is straightforward—but earning trust remains the goal.

The views expressed in this article are those of the author(s) and do not necessarily reflect the official position of the Swiss Institute of Artificial Intelligence (SIAI) or its affiliates.

References

Adobe. (2024). Authenticity in the age of AI: Growing momentum for Content Credentials across social media platforms and AI companies.
Coalition for Content Provenance and Authenticity (C2PA). (2025). C2PA technical specification v2.2.
Content Authenticity Initiative. (2025). 5,000 members: Building momentum for a more trustworthy digital world.
European Parliament. (2025, Feb. 19). EU AI Act: First regulation on artificial intelligence.
Freshfields. (2024, Jun. 25). EU AI Act unpacked #8: New rules on deepfakes.
Goldman Sachs. (2023, Apr. 19). The creator economy could approach half-a-trillion dollars by 2027.
Influencer Marketing Hub. (2025, Apr. 25). Influencer Marketing Benchmark Report 2025.
Meta. (2024, Feb. 6). Labeling AI-generated images on Facebook, Instagram and Threads.
Ofcom. (2024, Nov. 27). Four in ten UK adults encounter misinformation; only 30% confident judging AI-generated media.
OpenAI. (2025, Sept. 30). Sora 2 is here.
OpenAI. (2025, Feb. 15). Sora: Creating video from text.
Opendorse. (2025, Jul. 1). NIL at Four: Monetizing the new reality.
Reuters. (2025, Jul. 10). Industry video-game actors pass agreement with studios for AI security.
SAG-AFTRA. (2024–2025). Artificial Intelligence: Contract provisions and member guidance.
The Guardian. (2024, Mar. 21). Nearly 4,000 celebrities found to be victims of deepfake pornography.
TikTok Newsroom. (2024, May 9). Partnering with our industry to advance AI transparency and literacy. https://newsroom.tiktok.com.
YouTube Help. (2024). Disclosing use of altered or synthetic content.
Zhao, A., et al. (2025, Oct. 8). Adoption of watermarking for generative AI systems in the wild. arXiv.