AI-Driven Storytelling and Narrative Engineering

In 2023, a science fiction story generated almost entirely by AI won second prize in a Chinese provincial competition—only one or two of the six judges recognised its non-human origins. The same year, AI-illustrated children’s books flooded self-publishing platforms, and Hollywood writers walked out in part over fears that studios would use large language models to draft scripts on the cheap. These are small episodes in a much larger shift. Generative AI has moved from novelty to genuine creative force—and with that transition comes a set of increasingly urgent questions. As AI begins to play a larger role in cultural production, what guardrails are in place to ensure that its outputs reflect human values, intentions, and diversity? And more provocatively: what happens when the ability to generate stories at scale is directed not toward expression, but toward influence?

This essay explores that dual reality. On one hand, AI offers genuine promise: broader access to creative tools, new forms of adaptive narrative, and the democratisation of storytelling itself. On the other, there lies a speculative but plausible risk—that such systems could evolve into instruments of subtle narrative manipulation. While AI-driven storytelling remains in its infancy, and large-scale narrative orchestration is more theoretical than realised, it is precisely in these early stages that the frameworks for future ethical and technical development must be laid.

Generative AI is already reshaping how stories are conceived, produced, and shared. Whether through language models generating story drafts or text-to-image tools illustrating imagined worlds, AI is augmenting human creativity in ways that would have been unthinkable a decade ago.

One of the most promising developments is accessibility. Tools that once required formal training, expensive software, or studio backing are now available to anyone with a web browser. For emerging creators without institutional support, this levelling of the playing field opens doors: a teenager in Nairobi can visualise a comic series, a teacher in São Paulo can generate interactive fables for her students, a retiree in rural Canada can write a science fiction novel aided by machine prompts. AI has not eliminated the work of storytelling, but it has transformed the scaffolding around it.

Additionally, AI hints at a future where stories are no longer static artefacts but dynamic, adaptive systems. Narrative can respond to reader emotion, context, or history. It can become non-linear, endlessly recombinant, or even co-authored in real time. While these capabilities are currently limited by technical constraints and interface design, the direction of development suggests that storytelling may increasingly shift from fixed product to fluid process—from authored sequence to interactive experience.

“AI has not eliminated the work of storytelling, but it has transformed the scaffolding around it.”

Yet with new tools come new limitations. Most generative models are trained on large, publicly available corpora, much of which reflects dominant linguistic, cultural, and aesthetic norms. If left unexamined, these models may reinforce the patterns and biases already present in that data, producing outputs that favour established tropes over novel or marginalised perspectives. When the same models underpin creative tools worldwide, there is a risk that storytelling itself becomes flattened—varied in surface detail, yet convergent in deeper structure.

Economic pressures compound this risk. As automation becomes more capable, incentives may prioritise speed and volume over craft. Early signs are already visible: stock photography platforms have been flooded with AI-generated images, games localisation increasingly relies on machine translation with light human oversight, and marketing agencies experiment with AI-drafted copy to reduce costs. While widespread deskilling has not yet occurred, the trajectory is clear enough to warrant attention.

There is also the question of authenticity. A machine can simulate emotional tone, thematic depth, and stylistic nuance. But simulation is not the same as lived experience. There remains something distinctively human in how we narrate pain, joy, uncertainty, and contradiction—something rooted in embodied experience that AI may approximate but not originate. Audiences may eventually become indifferent to this distinction, but creators should not.

And crucially, an audience habituated to AI-generated content may become less equipped to notice when that content is designed not merely to entertain, but to persuade. A media environment saturated with structurally similar narratives—stories that share underlying tropes, emotional beats, and moral framings even when they differ in surface content—may be more susceptible to coordinated influence. If audiences lose the habit of encountering genuinely divergent perspectives, they may also lose the capacity to recognise when a particular worldview is being systematically reinforced.

This brings us to the more speculative concern: what I will call, following a term increasingly used in disinformation research, narrative engineering. By this I mean the deliberate, coordinated generation of thematically aligned stories across multiple channels—not to persuade through argument, but to shift perception through repetition and emotional resonance. The idea is not of a single AI-generated story that deceives, but of many small stories, each independent on the surface, that collectively cultivate a particular worldview.

Imagine an AI system tasked with promoting a soft ideological agenda—not by issuing propaganda, but by generating thousands of emotionally resonant, stylistically diverse stories that subtly echo a shared moral frame. No single narrative contains the message outright. Rather, the repetition of motifs, character arcs, or thematic resolutions begins to normalise certain values or foreclose others. Over time, such distributed storytelling could influence how people perceive truth, identity, authority, or collective action—without the directness of argument or debate.

“The idea is not of a single AI-generated story that deceives, but of many small stories, each independent on the surface, that collectively cultivate a particular worldview.”

At present, there is little public evidence that such AI orchestrated narrative engineering is being pursued at scale. The current technical and institutional limitations of generative AI make comprehensive scenarios unlikely in the near term. But the components of narrative engineering already exist and have been documented in adjacent domains. Coordinated inauthentic behaviour on social media—networks of fake accounts amplifying particular messages—has been extensively studied by researchers, including work published by the Stanford Internet Observatory and the Oxford Internet Institute’s Computational Propaganda Project. Political campaigns routinely A/B test messaging at scale, optimising for emotional response. Recommendation algorithms already shape what stories reach which audiences, creating feedback loops that can amplify certain narratives while suppressing others.

What remains speculative is not the individual techniques, but their synthesis: the possibility that generative AI could automate and integrate these capabilities into a coherent system of narrative influence. The logic is coherent, the incentives (political, commercial, or ideological) are clear, and the infrastructure—algorithmic content distribution, data-driven feedback loops, scalable media generation—is already taking shape. This makes narrative engineering not a present threat, but a future vector of concern that demands proactive attention.

Precisely because the more troubling uses of AI in storytelling remain speculative, now is the time to shape the systems and norms that will govern them. The challenge is not only technical but ethical: how do we ensure that the tools enabling broader access to creativity are not co-opted into instruments of subtle manipulation?

Transparency must become foundational. As AI-generated media becomes more widespread, audiences deserve to know when and how machine systems have contributed. Attribution should not be a matter of compliance or disclaimer, but an act of narrative honesty. Efforts are already underway: the Coalition for Content Provenance and Authenticity (C2PA)—whose founding members include Adobe, Microsoft, the BBC, Arm, Intel, and Truepic—has developed technical standards for embedding verifiable metadata in digital content. Major AI labs including OpenAI and Google DeepMind have experimented with watermarking techniques that can identify AI-generated text and images. These initiatives remain imperfect and incompletely adopted, but they represent a foundation on which more robust transparency norms can be built.

Beyond transparency, we need a cultural shift toward narrative literacy. Traditional media literacy focuses on evaluating source credibility or identifying bias in reporting. Narrative literacy asks deeper questions: How do stories construct identity? What do they frame as possible or impossible, good or evil, desirable or shameful? When such patterns repeat across seemingly unrelated narratives, they often become invisible—not because they are subtle, but because they feel like background truths. In practice, this might involve classroom exercises that ask students to identify recurring character types across AI-generated stories, or critical reading frameworks that prompt readers to ask: Whose perspective is absent here? What conflicts are resolved, and how? What emotions is this story designed to evoke, and to what end?

“When such patterns repeat across seemingly unrelated narratives, they often become invisible—not because they are subtle, but because they feel like background truths.”

Finally, the design of generative systems must incorporate ethical constraints as defaults, not afterthoughts. A system that helps a user write a memoir should not optimise for virality. An AI trained to generate fiction should not be engineered to maximise engagement at the expense of complexity or discomfort. These guidelines face genuine tensions: what happens when ethical constraints conflict with a platform’s business model? When responsible design reduces user engagement, who bears the cost? These are not reasons to abandon the goal, but rather to recognise that ethical AI design is not merely a technical problem. It requires institutional incentives, regulatory frameworks, and cultural expectations that reward responsible development even when it is not the most profitable path.

It is worth noting that some researchers argue these concerns are overblown. Audiences, they suggest, are more resilient to manipulation than critics assume. People have always navigated environments saturated with competing narratives—from religious texts to advertising to political rhetoric—and have developed informal heuristics for evaluating credibility. AI-generated content, on this view, is simply the latest iteration of a familiar challenge. This counterargument deserves serious consideration. But even if audiences prove more resilient than pessimists fear, the asymmetry of effort matters. Generating thousands of tailored narratives costs almost nothing; critically evaluating each one costs time and attention that most people do not have. Resilience is not a reason for complacency.

AI-based storytelling stands at a formative juncture. The question is not whether AI can generate compelling stories—it already can. The deeper question is what values guide the systems that generate them, and whether we are willing to interrogate the motives behind the narratives we increasingly consume. Will we use these tools to widen the scope of human expression, or to flood it with coherent, affective, beautifully rendered influence?

We remain, for now, the authors of the systems that may soon author much of our culture. The choices we make in these early years—about transparency, literacy, and design—will shape the narrative landscape for decades to come. The opportunity is still ours to determine what stories we let them tell, and why.

Sources

Bradshaw, Samantha, Hannah Bailey, and Philip N. Howard. “Industrialized Disinformation: 2020 Global Inventory of Organized Social Media Manipulation.” Working Paper 2021.1. Oxford, UK: Programme on Democracy & Technology, Oxford Internet Institute, 2021. https://demtech.oii.ox.ac.uk/research/posts/industrialized-disinformation/

Chik, Holly. “A Chinese Professor Used AI to Write a Science Fiction Novel. Then It Was a Winner in a National Competition.” South China Morning Post, 20 December 2023. https://www.scmp.com/news/china/science/article/3245725/chinese-professor-used-ai-write-science-fiction-novel-then-it-won-national-award

“ChatGPT-Written Books Are Flooding Amazon as People Turn to AI for Quick Publishing.” South China Morning Post (via Reuters), 22 February 2023. https://www.scmp.com/tech/big-tech/article/3211051/chatgpt-written-books-are-flooding-amazon-people-turn-ai-quick-publishing

Coalition for Content Provenance and Authenticity. https://c2pa.org/

“Hollywood Writers Went on Strike to Protect Their Livelihoods from Generative AI.” Brookings Institution, 23 May 2024. https://www.brookings.edu/articles/hollywood-writers-went-on-strike-to-protect-their-livelihoods-from-generative-ai-their-remarkable-victory-matters-for-all-workers/

“Stanford Internet Observatory.” Wikipedia. https://en.wikipedia.org/wiki/Stanford_Internet_Observatory