A new study examining meme creation finds that AI-generated meme captions on well-known meme images tend to score higher on average for humor, creativity, and shareability than captions produced by humans. Yet the most exceptional individual memes still come from human creators, underscoring a nuanced balance between AI-assisted productivity and human originality. The research, slated for presentation at the 2025 International Conference on Intelligent User Interfaces, sheds light on how AI and humans perform differently in humor-creation tasks and raises important questions about the role of AI in creative processes. While some experts celebrated the results as a milestone for machine-generated humor, others cautioned that broad appeal does not necessarily equate to deepest impact, and that human nuance remains crucial for creating content with personal resonance. The study, conducted by an international team from leading European institutions, provides an in-depth look at how different experimental setups influence meme quality and the dynamics of collaboration between people and AI.
Study design, scope, and context
The research reported in this study positioned its inquiry at the intersection of humor, creativity, and online sharing dynamics. The experimental framework involved three distinct scenarios designed to compare how memes are produced under different conditions: humans working alone, humans working with the assistance of a large language model (LLM), and memes generated entirely by an artificial intelligence system without human input. The LLM used in the experiments was a state-of-the-art model, and the researchers specifically scoped the study to captions rather than image generation. In line with common meme practices, the images used were pre-existing, popular meme templates rather than AI-generated visual content. This choice allows the study to isolate caption-writing as the variable under investigation, while leveraging familiar templates to test how well AI and human creators can capitalize on contextual cues, cultural references, and shared experiences that meme audiences typically recognize and appreciate.
Three familiar thematic categories were selected to probe how context affects humor, creativity, and shareability. These categories spanned work-related scenarios, food-related situations, and sports contexts. By loading the study with these everyday domains, the researchers aimed to understand whether AI’s performance varies with subject matter that participants routinely encounter in daily life and online discourse. The methodology employed crowdsourced evaluators to rate meme captions, providing a broad cross-section of perceptions of humor, originality, and potential for wide circulation. This evaluative approach enables the study to approximate real-world reception across a diverse audience, rather than relying solely on expert judgments or narrow panels.
In addition to the primary comparison among the three production modes, the study introduces a nuanced perspective on collaboration between humans and AI. It creates a spectrum of human involvement, from solitary human authorship to joint human-AI production, and finally to AI-only generation. This design allows for a direct assessment of how AI assistance might influence not only the final product but also the process by which ideas are generated and refined. The participants in the study varied in background and expertise, reflecting a broad pool of crowd-sourced contributors rather than a homogeneous sample. The aim of this diverse participant base is to capture a range of taste profiles, humor sensibilities, and cultural references that shape the way meme captions are evaluated in the broader internet ecosystem.
The study’s framing of the meme creation task emphasizes the practical aspects of content production in the digital age. By focusing on captions for established meme formats, the researchers align their work with a real-world content creation workflow in which creators often select familiar templates to reach large audiences quickly. The evaluation metrics—humor, creativity, and shareability—were operationalized through predefined rating scales, with raters asked to judge how funny a caption is, how original or imaginative it feels, and how likely it is to be widely shared or propagated across social platforms. The authors well note that these metrics, while informative, are inherently influenced by subjective, cultural, and temporal factors, and that the study’s conclusions must be interpreted within that context.
The international collaboration behind the study included prominent European institutions, with a joint effort spanning multiple countries and research cultures. The cross-institutional nature of the team enhances the study’s generalizability by incorporating diverse perspectives on humor, creative critique, and human-computer interaction. The research team also engages with ongoing debates about the role of AI in creative fields, particularly around questions of authorship, originality, and the value of human perspective in content that seeks to connect emotionally with audiences. This broader context is essential for understanding why the study’s results matter beyond mere numerics: they contribute to a larger conversation about how AI can complement human skill while preserving the distinctive attributes of human creativity.
The study’s timing is notable in that it arrives during a period of rapid advancement in AI-assisted content generation, especially in the realm of online memes, social media culture, and digital entertainment workflows. The rapid evolution of large language models, coupled with growing accessibility for creators to leverage AI in everyday tasks, makes the study’s findings timely and relevant for practitioners, researchers, and policy makers who are interested in the practical implications of AI as a creative partner rather than a replacement for human talent.
Overall, the study constructs a comprehensive framework for evaluating AI and human-generated meme captions across multiple dimensions and in varying collaboration setups. By combining rigorous experimental control with ecologically valid stimulus materials (well-known meme templates) and a broad evaluative audience, the researchers provide a robust foundation for interpreting the results and situating them within the broader discourse on AI-assisted creativity. The paper’s conclusions emphasize both the strengths and limitations of AI in humor generation and highlight the enduring value of human insight for producing standout, memorable creative work. The authors also acknowledge that the dynamics uncovered in this study may translate differently across platforms, audiences, and cultural contexts, underscoring the importance of ongoing study as AI technologies and user behaviors continue to evolve.
Key findings: humor, creativity, and shareability across production modes
The core results of the study reveal a layered landscape of performance differences among AI-generated, human-generated, and human-AI collaborative memes across the three thematic categories (work, food, sports). On average, captions produced by AI alone demonstrated higher scores for humor, creativity, and shareability than captions crafted by humans or by humans who collaborated with the AI. This finding suggests that the AI model, trained on extensive internet data, has captured broad patterns of humor and topical relevance that tend to resonate with large audiences in a rapid, scalable way. The AI’s capacity to identify widely appealing humor and topicality likely underpins its superior average performance across these aggregate metrics.
However, one of the most striking nuances in the results concerns the distribution of quality across the meme set. While AI-generated captions excel on average, the best-performing individual memes—those most memorable and impactful in terms of humor and originality—tend to come from human creators. The study’s analysis indicates that humans, when crafting memes either alone or in collaboration with AI, can produce especially funny content that leaves a lasting impression, even if those instances are fewer in number relative to the AI-generated supply. In other words, AI tends to produce broadly appealing humor that looks “safe” and widely comprehensible, while humans can push the envelope to deliver moments of exceptional comedic impact.
Adding another layer, the combination of human creativity with AI assistance did not, on average, outperform all-human production in the measured metrics where humans act solo. The data show that memes created by humans alone and memes created through human-AI collaboration did not register higher average scores than AI-generated memes in aggregate. Yet the collaborative approach did yield notable advantages in creativity and shareability at the level of specific memes, indicating that AI assistance can augment creative possibilities and help surface more ideas—though that boost in quantity does not automatically translate into higher average quality. The study’s explicit framing captures this paradox: AI can increase productivity and provide a wider pool of content ideas, but the overall quality, as judged by average humor and originality across a broad set of memes, may still be anchored by human nuances and sensibilities.
A particularly telling insight concerns the dynamics of idea generation and ownership. When participants used AI assistance, they reported generating significantly more meme concepts and found the process easier and less effortful overall. This productivity uplift demonstrates the practical value of AI as a creative aid, substantiating claims that AI can help content creators brainstorm more rapidly and iterate designs more efficiently. However, the study notes that these efficiency gains did not automatically yield superior average results—more output did not guarantee better quality. Another dimension of the collaboration effect is perceived ownership: participants who relied on AI assistance felt slightly less ownership over their final memes than solo creators. Given that a sense of ownership has been linked to creative motivation and satisfaction, the authors suggest that creators considering AI tools should deliberately manage the balance between AI input and personal authorship to sustain motivation and engagement.
The paper also highlights a notable distinction between the average performance of AI-generated content and the qualitative excellence of human-generated variants. The AI model’s high average performance across humor, creativity, and shareability is attributed to broad patterns learned from vast datasets, enabling it to recognize widely appealing humor and culturally resonant themes. In contrast, human-generated memes often draw on more intimate experiences and nuanced personal references, which can yield moments of exceptional hilarity or relevance that do not always translate into broad appeal. This contrast underscores a central tension in AI-assisted humor: what is broadly funny versus what is strikingly funny in a personal or idiosyncratic way. The researchers emphasize that while AI can replicate and amplify common humor patterns, it may miss the distinctive spark that arises from deeply personal or context-rich experiences, which humans can deliver.
In the results, the researchers also present visual representations illustrating the relative performance of AI, human, and human-AI meme-generation pathways across the evaluation metrics. The diagrams highlight that while AI tends to outperform on average, the top memes—those that achieve the strongest impact in humor and originality—are more likely to emerge from human creators and from human-AI collaborations that harness human judgment for selection and refinement. These insights illustrate a dynamic where AI provides a broad palette of ideas and patterns, while humans curate and elevate particular captions into standout creations.
Finally, the study discusses a subset of results framed through the lens of context. When ratings were broken down by meme category, work-related memes often scored higher for humor and shareability than memes about food or sports. This pattern suggests that certain contexts may lend themselves more readily to meme humor, even when the underlying captions are AI-generated, a phenomenon the researchers interpret as evidence that contextual familiarity and work-related experiences can influence how humor translates into shareability. In short, context matters: a given caption may perform differently depending on the domain and audience expectations embedded in that context, regardless of whether it was produced by a human or an AI.
Production modes and interpretive implications
An essential theme running through the findings is the interplay between broad appeal and individualized wit. AI models, trained on expansive datasets from the internet, can detect and reproduce patterns of humor that tend to resonate with general audiences, leading to higher average ratings across humor, creativity, and shareability. Yet this strength can also mean that AI-generated captions might lack the personal touch, spanning a wide audience but offering fewer moments of uniquely resonant humor that come from lived experiences or niche cultural references.
Human creators, by contrast, are better positioned to inject nuance, sarcasm, insider jokes, or specific cultural signals that can elevate a single meme to meme-legend status. The study’s observation that the most exceptional memes arose from human authors or from human-AI collaborations aligns with longstanding insights in creative fields: while AI can expand the palette and accelerate iteration, human discernment, taste, and experiential depth often drive the peak moments of creativity and impact.
In practical terms, the study’s results suggest a model for effective meme production in a world where AI tools are readily accessible. For broad reach and rapid generation, AI-alone captioning can be a powerful baseline, offering many options that align with widely shared humor patterns. For high-impact memes that aim to be particularly memorable or feel deeply personal, human authorship remains essential, whether operating solo or in collaboration with AI to brainstorm, refine, and select the strongest candidates. The most compelling workflows may blend human judgment with AI-assisted ideation, allowing creators to explore a wide field of possibilities and then apply human insights to curate and polish the best ideas into standout memes.
The study also highlights an important dynamic for content-creators on platforms where engagement and virality hinge on rapid production. AI-assisted approaches can substantially increase the volume of ideas and reduce the cognitive and time burdens of generation, enabling teams to scale their meme production pipelines. However, platform managers and creators should be mindful of the potential trade-offs in perceived ownership, originality, and audience perception when AI tools are used extensively. Human creators who leverage AI should be intentional about maintaining a unique voice, brand alignment, and personal connection that can distinguish their memes in a crowded content landscape. The research contributes to a broader conversation about how creators can responsibly integrate AI into their creative workflows to maximize both efficiency and quality.
Contextual cues: why the AI-generated memes resonated on average
A central question emerging from the study concerns why AI-generated memes achieved higher average scores on humor, creativity, and shareability. The researchers attribute much of the AI model’s strength to its training with vast volumes of internet data, which enables it to identify and reproduce widely appealing humor motifs and topical patterns. This broad training helps AI detect contemporary references and cultural touchpoints that tend to resonate with a broad audience, enabling it to generate memes that feel timely and accessible.
In contrast, human-created memes tend to lean into more personal or experience-based humor. These memes often reflect specific moments, shared jokes, or industry insider references drawn from the creators’ lived experiences. While such memes can exhibit higher novelty or emotional resonance for certain viewers, they may not achieve the same broad appeal as AI-generated captions that tap into universal or widely recognized patterns. This dichotomy between broad appeal and individualized humor offers a plausible explanation for the study’s observed outcomes: AI produces broadly appealing content that is easy to share at scale, while humans can deliver standout moments that feel deeply authentic to particular audiences.
The researchers also discuss the potential ecological validity of the results—the way the study translates to real-world meme creation and distribution on social platforms. Given that online audiences are diverse and constantly shifting their humor tastes, AI-generated captions’ breadth could be advantageous for platforms seeking fast, scalable content that maintains consistent engagement. Meanwhile, for creators aiming to cultivate a distinctive voice, the ability to craft or curate memes that reflect nuanced perspectives remains valuable and necessary to sustain long-term audience loyalty and brand identity. The interplay between AI breadth and human depth suggests a future in which creators harness AI as a tool for exploration and iteration while reserving human judgment for quality selection and signature style.
The study also invites reflection on how culture, time, and platform dynamics influence humor reception. The evaluative context—crowdsourced raters with varied backgrounds—captures a snapshot of broad audience tastes but also introduces biases that can shape outcomes. The researchers acknowledge that evaluating humor is inherently subjective and influenced by social norms, trends, and cultural specificity. As AI’s role in content generation grows, ongoing research will be needed to understand how evolving humor norms, audience demographics, and platform algorithms interact with AI-assisted content to shape what becomes popular or influential over time.
Two broader implications emerge from this context: first, AI’s capacity to generate broadly relevant humor may be well-suited for tasks requiring high-volume, rapid content production across diverse topics; second, the unique, personal spark characteristic of human-authored memes may be best preserved through careful curation and creative direction that integrates AI outputs with human sensibilities. The study thus supports a model in which AI serves as a discovery engine and productivity booster, while humans retain critical roles in curation, refinement, and the expression of deeply human humor.
Limitations, caveats, and avenues for future research
No study is without limitations, and this research identifies several important caveats that shape how its findings should be interpreted and applied. First, the meme caption creation sessions were relatively short. The authors note that extended use of AI tools and more sophisticated prompting could potentially strengthen the quality of human-AI collaborations, enabling participants to coax more nuanced humor or sharper writing from the AI model. Longer sessions might also reveal deeper dynamics of co-creation, including how fatigue, iteration speed, and iterative feedback loops affect creative outcomes when humans and machines work together over extended periods.
Second, the evaluation relied on crowdsourced evaluators, introducing subjectivity and potential biases toward mainstream or conventional humor. This design choice, while reflective of broad audience reception, may overweight content that aligns with popular tastes and avoid more culturally specific, niche, or boundary-preaking humor that could be highly valued by particular communities. The researchers acknowledge that expert panels or targeted demographic groups might provide complementary insights into nuanced aspects of humor and creativity across different cultural frames. Future work could incorporate a layered evaluation approach, combining broad crowdsourcing with targeted expert judgments to capture a wider spectrum of humor and creative criteria.
Third, the study emphasizes captions rather than image generation, and it uses established meme templates rather than AI-generated visuals. While this approach isolates caption quality and aligns with common meme practices, it leaves open questions about how AI might perform when both text and image generation are integrated. Future research could explore end-to-end meme creation that includes image synthesis, layout decisions, and typography, assessing how joint text-visual generation interacts with human input in shaping humor and shareability.
Fourth, the researchers point to the potential biases introduced by the use of crowd-based evaluations and the possibility that certain humor styles may be favored by broad audiences at the expense of more niche or culturally specific humor. This caveat invites further exploration into how different evaluator populations—such as region-specific panels or age cohorts—might respond to AI-generated versus human-generated memes and how cultural context shapes humor appreciation in meme culture.
Fifth, the research points toward the possibility that AI can rapidly generate numerous ideas, enabling humans to act as curators who select and refine the best content. While this scenario holds promise for efficient content workflows, it also raises questions about originality and the boundaries of authorship when AI contributes extensively to the meme creation process. Future studies could investigate the dynamics of authorship rights, attribution, and creative credit in AI-assisted meme production, as well as the long-term effects on creator motivation and platform norms.
In terms of future directions, the authors propose several promising avenues. They suggest examining scenarios where AI can rapidly generate multiple ideas, after which humans act as curators to select, refine, and finalize the best content. This curator approach could illuminate how human judgment can elevate AI-generated outputs to higher levels of humor, creativity, and shareability. The study also invites exploration of whether longer engagement with AI tools or more advanced prompting strategies can further enhance the quality of collaborative memes. Additionally, broader experimentation with diverse audiences, expert panels, and culturally specific groups could help capture more nuanced understandings of humor and creativity across contexts.
Ultimately, the study concludes that, despite AI’s impressive average performance and the productivity gains associated with AI-assisted creation, humans remain the champions of truly exceptional meme captions. The results emphasize a prevailing truth in creative work: while AI can amplify output, nuance, personal experience, and distinctive voice are still uniquely human strengths. This insight has practical implications for writers, designers, marketers, and platform developers seeking to integrate AI into meme production pipelines in a way that preserves human artistry and personal resonance even as AI handles broader, scalable tasks.
Implications for creators, platforms, and policy considerations
The findings carry several practical implications for creators and digital platforms navigating the integration of AI into meme production. For creators, the study underscores the value of combining AI-assisted ideation with personal authorship to achieve both breadth and depth in meme output. AI can serve as a powerful engine to generate a wide range of captions rapidly, enabling creators to explore many directions and identify patterns that resonate with audiences. However, to craft memorable moments with lasting impact, creators should retain control over the curation and refinement process, ensuring that the final memes reflect their voice, perspective, and brand identity. The research implies that successful meme strategies in the AI era may hinge on balancing automation with authentic human expression, leveraging AI for idea generation and first-pass drafting while preserving human judgment for quality assessment and signature styling.
For platforms that host, promote, or curate meme content, the study suggests opportunities to optimize for both volume and quality. AI-assisted content workflows can increase engagement by delivering a steady stream of captions that fit popular formats and cultural references. But platforms should also consider human-aided curation mechanisms to highlight exceptional memes that carry unique voice or originality, ensuring that the platform remains a space for distinctive content creators and not solely a reservoir of broadly appealing AI outputs. This dual approach could help platforms foster a diverse ecosystem where AI and human creativity complement each other rather than compete.
The study also touches on broader policy considerations related to authorship and attribution in AI-assisted creative content. As AI tools become more embedded in content production workflows, questions arise about how to credit AI contributions and how to allocate recognition between human creators and AI systems. These discussions are likely to intensify as AI continues to evolve, and they will require thoughtful guidelines that balance transparency, intellectual property considerations, and the practical realities of collaborative creativity. The research emphasizes that while AI can facilitate productivity, human authorship and responsibility for final content remain central to the ethics and governance of meme creation.
From a broader cultural perspective, the study’s insights into the strengths and limitations of AI-generated humor raise important considerations for the long-term evolution of meme culture. The broad appeal of AI-generated captions may accelerate the pace at which memes spread and mutate across platforms, potentially accelerating trends and cultural references. At the same time, the enduring value of personal humor and nuanced social signaling suggests that communities will continue to celebrate and curate memes that reflect lived experiences, insider knowledge, and distinctive voices. In this sense, AI is a tool that can amplify reach and efficiency while human communities preserve the unique voice and collective memory that give memes their resilience and cultural significance.
The study’s careful treatment of context, collaboration dynamics, and evaluation challenges also has implications for education and training in creative AI literacy. As AI becomes an increasingly common collaborator in content creation, learning how to prompt effectively, curate AI outputs, and recognize the boundaries of machine-generated humor will become valuable competencies for students, designers, and marketers. The researchers’ emphasis on maintaining ownership and motivation in human-AI collaborations highlights the need for training programs that help creators integrate AI in ways that support, rather than erode, personal agency and professional satisfaction.
In sum, while AI-assisted meme generation offers clear advantages in speed, breadth, and baseline humor alignment, the study reinforces a timeless principle: the most impactful memes frequently emerge from human insight and reflection, and the most compelling creative work often arises at the intersection of human judgment and machine-generated inspiration. For practitioners, the takeaway is to design workflows that leverage AI for ideation and rapid drafting while preserving human curation, taste, and voice to produce the standout content that defines a creator’s brand and resonates with audiences over time.
Methodological reflections and critical takeaways for researchers
The study provides a robust methodological blueprint for investigating AI in creative tasks, particularly in humor and meme culture. The tripartite comparison—human-only, human-AI collaboration, and AI-only generation—offers a clear framework to disentangle the contributions of human creativity and machine computation. This design allows researchers to identify where AI adds value, where it complements human insight, and where human intervention remains indispensable. By testing across multiple familiar domains (work, food, sports), the study also demonstrates how context shapes humor reception, highlighting the importance of domain choice in evaluating AI’s creative capabilities.
One notable methodological strength is the use of well-established meme templates rather than synthetic visuals. This choice enhances ecological validity by ensuring that the stimuli align with real-world meme consumption and interpretation. It also avoids conflating image-generation quality with caption quality, allowing a more precise assessment of how captions contribute to humor, originality, and shareability. The reliance on crowdsourced evaluators captures broad audience perspectives, offering a realistic gauge of how memes would perform in everyday online settings. However, this approach also introduces variability related to rater demographics, mood, and trending topics that may influence judgments.
For future research, several directions emerge from the study’s design and findings. Researchers could explore longer-term experiments that track how memes evolve under iterative AI-influenced workflows, including the effects of repeated prompts, feedback loops, and iterative curation. Investigating expert panels alongside crowdsourced ratings could provide a more nuanced view of humor and creativity, capturing cultural specificity and technical critique that crowdsourcing alone might miss. Another promising avenue is studying end-to-end meme creation that combines image generation, captioning, layout decisions, and typography, examining how integrated AI systems and humans jointly navigate multi-modal content production.
Additionally, researchers may examine how different AI prompting strategies and user-interface designs influence collaboration outcomes. For instance, prompts that guide AI to propose multiple alternative captions, or interfaces that make it easier for humans to rate, rank, and remix AI-generated options, could reveal actionable insights into optimizing human-AI co-creation processes. It would also be valuable to investigate the social and psychological dimensions of AI-assisted creativity, including how creators’ sense of control, autonomy, and satisfaction are affected by varying levels of AI involvement.
Broader methodological lessons from this study emphasize the importance of defining clear, multi-dimensional evaluation metrics that capture both broad appeal and high-impact moments. By including humor, creativity, and shareability as distinct yet interrelated measures, the study provides a model for assessing creative outputs that must function in real-world social ecosystems. Researchers should continue to refine these metrics, consider other dimensions such as originality, relevance, memorability, and emotional engagement, and explore how platform-specific dynamics—like algorithmic feeds, moderation, and audience feedback loops—shape the reception of AI-generated versus human-created content.
In conclusion, this study contributes a valuable, methodologically rigorous perspective on AI-assisted humor and meme creation. It highlights the strengths and limitations of AI in generating broad appeal while underscoring the enduring value of human originality in producing truly standout content. The insights invite ongoing inquiry into how best to integrate AI into creative workflows in ways that maximize collaboration, preserve human agency, and foster a vibrant ecosystem of memes that resonates with diverse audiences across platforms and cultures.
Conclusion
The study presents a nuanced portrait of AI-assisted meme creation, revealing that AI-generated captions on popular meme templates can outperform human-generated captions on average in humor, creativity, and shareability. Yet the most memorable, high-impact memes tend to originate from human creators, and human-AI collaborations can surface the most creative and shareable outputs in some cases. AI tools also boost productivity by generating more caption ideas and lowering the effort required to produce content, but this increased output does not automatically translate into higher average quality, and users often feel a reduced sense of ownership over AI-assisted memes.
Context matters: the specific meme category influences how humor translates into engagement, with work-related memes sometimes garnering higher humor and shareability scores than those about food or sports. The study’s use of established meme templates and crowdsourced evaluators provides insights into broad audience reception while highlighting limitations related to session length, evaluative biases, and cultural specificity. The researchers emphasize a balanced approach that leverages AI for ideation and rapid drafting while preserving human judgment for refinement, curation, and the distinct human voice that drives lasting resonance.
Looking ahead, the study points to fruitful avenues for future work, including longer-term collaborations between humans and AI, more nuanced and diverse evaluative panels, and end-to-end explorations that integrate image generation with captioning and layout decisions. It also hints at broader implications for authorship, attribution, and platform strategy as AI becomes an increasingly integrated collaborator in creative workflows. Ultimately, the takeaways reaffirm a core principle: AI can expand the horizon of possibilities and accelerate creation, but human creativity—with its depth of experience, personal nuance, and cultural insight—remains indispensable for producing truly exceptional meme content that captivates and endures.
