An international study on meme creation reveals that AI-generated captions on famous meme templates tend to score higher on humor, creativity, and shareability than captions produced solo by humans. Yet the most exceptional memes—those that stand out for both wit and resonance—still come from humans or from human–AI collaborations. The findings paint a nuanced picture: AI can boost productivity and broad appeal, but human creativity remains essential for content that connects on a deeper level. Across three familiar contexts—work, food, and sports—the research scrutinizes how AI and humans perform in humor-driven content creation, challenging simple narratives about machine versus person in the realm of memes.
Study Scope and Context
The study undertook a comprehensive examination of meme-caption creation by comparing three distinct creative setups: humans working alone, humans collaborating with a large language model (LLM), specifically GPT-4o, and memes generated entirely by the AI model with no human input. The experimental design deliberately used well-known, pre-existing meme templates rather than generating new images, ensuring that the variable under test was the caption creator rather than the visual format itself. This choice isolates the core question: can AI-captioning approaches leverage the cultural and linguistic patterns embedded in vast internet text to produce captions that are funnier, more creative, and more shareable than those created by people?
In line with this aim, the researchers organized their evaluation around three perceptually resonant categories that span everyday life and popular culture: work, food, and sports. By selecting these domains, the study sought to capture how humor functions within commonly understood social contexts, where jokes can hinge on shared experiences, professional life, everyday routines, and ubiquitous cultural references. The decision to use three broad but relatable contexts helps illuminate contextual sensitivity in humor generation: what resonates in a workplace joke might differ markedly from what makes a sports-related pun land, or what triggers humor in a culinary setting. This contextual dimension is essential for understanding how AI systems generalize humor across domains and how human sensibilities compare when the same material is treated by different creators.
A key methodological note is that while the image templates were pre-existing meme formats, the caption generation process was strictly what varied across the experimental arms. The AI-generated captions came from the GPT-4o model, and in the collaborative condition, humans produced captions with AI-assisted prompts or guidance. Conversely, the human-alone condition relied entirely on unaided human creativity. The study’s researchers emphasize that the images themselves did not come from AI; the emphasis was on caption-level creativity and humor. This distinction matters because it clarifies where AI’s strengths lie: narrative and linguistic patterning that can be applied to familiar images, rather than visual originality of the images themselves.
To evaluate the outputs, crowdsourced raters assessed the memes on three main dimensions: humor, creativity, and shareability. The researchers operationalized shareability as the probability that a meme would be widely circulated—an outcome influenced by how funny a caption is, how relatable it feels, and how timely or culturally relevant its content appears. This multi-metric approach acknowledges that a successful meme is not merely funny; it must also feel novel enough to attract attention and be easily shared across social networks.
In reporting their findings, the researchers note a provocative comparison to the classic meme of the “meme Turing Test.” A well-known online remark framed the results in dramatic terms: AI’s performance in meme captioning could be interpreted as a signal that machines have crossed a threshold of human-like content creation. Yet the paper itself includes a cautious caveat: although AI can boost throughput and attract broad audiences, human creativity remains indispensable for content that creates deep connection. This balance—between AI efficiency and human depth—emerges as a central insight from the study.
The international research team behind this work includes scholars from leading institutions, working across Europe and beyond, pooling expertise in AI, cognitive science, and human–computer interaction. Their collaboration underscores a growing trend in AI-enabled creativity research: that interdisciplinary approaches are crucial to disentangle the ways machines and humans contribute to culturally meaningful artifacts like memes. The overarching implication is clear: AI’s capabilities in producing broad appeal should be viewed as a complement to human talent, not a replacement for it.
In interpreting these results, it’s important to remember the study’s framing: AI can learn to imitate and remix humor patterns found across the internet, enabling it to generate captions that land with large audiences. Humans, by contrast, can draw on personal experience, nuanced social understanding, and a sense of timing that may be harder for a machine to replicate consistently. The combination of AI’s breadth with human depth—when orchestrated effectively—appears to yield the most compelling outcomes in terms of creativity and shareability.
Methodology and Experimental Setup
The study’s experimental architecture was designed to parse out the contributions of human intuition, AI amplification, and their synergy. Three distinct creation conditions formed the backbone of the investigation: solo human authorship, human authorship augmented by AI assistance, and AI-only caption generation. In all cases, the same set of widely recognized meme templates served as the canvas, ensuring that performance differences reflect captioning strategies rather than visual variation. This consistency is crucial for isolating the efficiency and quality of captioning across modalities.
The meme templates were chosen from a pool of widely familiar formats to maximize recognizability and potential relatability. The researchers explicitly state that the images themselves were not produced by AI; instead, AI and humans produced captions that could be paired with these templates. This methodological emphasis on caption-level creativity aligns with a broader inquiry into how language models can contribute to humor without necessarily altering the iconic status of the underlying memes.
The three relational categories—work, food, and sports—were selected for their resonance with everyday life and cultural touchpoints. Each category provides a unique lens through which humor can be tested. Work-related memes often hinge on shared professional experiences, workplace culture, and common career challenges. Food memes draw on culinary habits, dining culture, and seasonal or trend-driven topics. Sports memes leverage athletic contexts, fandoms, and the social dynamics surrounding games and competition. By evaluating these domains, the study aims to identify patterns in humor generation that are sensitive to context and domain-specific knowledge.
In the human-plus-AI condition, participants collaborated with GPT-4o to generate captions. The collaboration was designed to explore how AI can scaffold human creativity—offering prompts, suggested phrasings, or alternative angles that a human creator could then refine. The AI-only condition tested GPT-4o’s capability to autonomously compose captions for the same templates, thereby measuring the machine’s capacity to generate novel, humorous, and shareable content without human intervention.
Crowdsourced evaluators played a central role in the assessment process. Raters evaluated memes on a triad of metrics: humor, creativity, and shareability. These measures were defined to capture not only immediate comedic impact but also the potential for meme virality and cultural relevance. The researchers indicate that shareability is influenced by humor, relatability, and the meme’s alignment with current cultural topics. By using a crowd-based evaluation, the study sought to approximate real-world reception among diverse audiences with varied tastes and backgrounds.
A qualitative dimension accompanies the quantitative scores. The study includes illustrative examples of meme captions generated under each condition, enabling readers to see concrete instances of how AI and human authorship diverge in tone, word choice, and humorous strategy. While some AI-generated captions may rely on broad, pattern-based humor, human-generated captions can reveal sharper cultural nuance, insider jokes, or highly specific situational humor. This juxtaposition is central to understanding why AI may yield high average scores even as top individual memes still favor human creativity.
The researchers acknowledge several constraints that shape the interpretation of results. First, the caption-creation sessions were relatively short, potentially limiting the opportunity for participants to explore extended creative strategies and deeper AI collaboration. Second, the crowdsourced evaluation introduces subjectivity and possible biases toward mainstream humor, which could inadvertently advantage AI-generated content tailored to broad appeal. These limitations point toward avenues for future research, including longer collaboration cycles, richer prompting paradigms, and diversified evaluator panels, to gain a more granular understanding of humor and creativity across demographic and cultural lines.
In terms of data presentation, the study reports comparative averages across the three conditions and analyzes best-case performances to identify how the top memes differ from the overall trend. The researchers also examined productivity metrics, such as the number of meme ideas generated, and qualitative indicators of process ease or difficulty. The “top memes” analysis highlights that while AI excels at producing a large volume of broadly appealing content, the most compelling individual memes—and the most creative or highly shareable outputs—often emerge from human or human–AI collaborative workflows.
The paper concludes with a forward-looking perspective: there is value in designing AI tools that emphasize rapid ideation while preserving human curation and final decision-making. The balance between AI’s capacity to generate ideas quickly and humans’ ability to select, refine, and infuse nuance appears to be a critical factor in producing memes that are not only widely consumed but also deeply resonant.
Performance Across Categories and What It Means
Across the three categories—work, food, and sports—the study uncovered notable patterns in how AI and human creators perform when tasked with humorous captioning. Memes in the work category tended to receive higher humor and shareability scores relative to those in food or sports categories. This observation suggests that workplace-related humor often touches on universal experiences, shared frustrations, and recognizable scenarios that transcend specific cultures or subcultures, enabling AI-generated captions to connect with a broad audience. In contrast, memes centered on food or sports sometimes hinge on niche tastes, specialized knowledge, or highly specific fandoms, which can limit the mass appeal of captions generated by AI or humans alike.
The AI models demonstrated strong performance in producing broad appeal across contexts, which aligns with the model’s training on massive, diverse internet datasets. Patterns of humor—such as surprise, incongruity, wordplay, and relatable misfortunes—tend to recur across various online communities. The AI’s ability to identify and manipulate these patterns at scale can yield captions that “feel” familiar and engaging to many viewers, a quality that translates into higher average scores for humor, creativity, and shareability when aggregated across large rater samples.
Nevertheless, the study highlights a clear asymmetry between average performance and peak quality. While AI-generated captions generally outperformed human-created captions on average, the most humorous memes—those that generate the strongest laughter or sense of delight—were predominantly produced by humans or by human–AI collaborations. This finding implies that while AI can produce consistently appealing content, the spark of originality or the precise comedic timing that elevates a meme to a standout status still often hinges on human insight or the nuanced interaction between human judgment and AI suggestion.
Context matters. The research team emphasizes that the same AI or human approach can yield different results depending on the category. Work-related memes benefited more from AI-driven approaches in terms of humor and shareability, reflecting perhaps a broader resonance with professional or organizational experiences that many people recognize. In the food and sports domains, AI’s broad pattern recognition sometimes collided with more specialized humor cues, cultural references, or insider knowledge that humans are more likely to possess or tailor to a particular audience. This contextual sensitivity underscores the value of tailoring creative processes to the intended audience and purpose of meme content.
The study also makes a crucial distinction regarding the images themselves. The AI did not generate the meme visuals; instead, the researchers used established meme templates. The captions—whether produced by AI, by humans, or by humans with AI assistance—were the variable under study. This distinction is important because it clarifies the locus of AI influence: the captioning and linguistic creativity, rather than image generation, which remains a separate creative task with its own dynamics.
From a broader perspective, the results underscore a fundamental truth about humor and cultural content in the digital age: large-scale data-driven models excel at detecting and reproducing patterns that work across broad audiences, which is why AI often lands with high average scores for broad appeal. Yet the most memorable or provocative jokes—those capable of provoking strong laughter, empathy, or a sense of shared identity—often arise from personal experiences, insider knowledge, or mischievous wordplay that resonates on a more intimate level. Thus, AI’s strength lies in amplification and pattern recognition, while human creativity remains essential for depth, novelty, and emotional resonance.
In practical terms, these findings suggest a pragmatic approach to meme creation. AI can be a powerful tool for ideation, rapid drafting, and broad-salience testing, enabling creators to generate a large volume of candidate captions quickly. Humans can then curate, refine, and elevate the most promising options, infusing them with experiential nuance, cultural specificity, and inventive phrasing that a machine may miss. The combination—human curation guided by AI-assisted ideation—appears to offer a path toward maximizing both volume and quality, especially for creators seeking to balance efficiency with memorable impact.
The study also reveals a striking dynamic in the relationship between productivity and quality. When AI assistance is used, participants reported generating significantly more meme ideas and experiencing a smoother, less effortful captioning process. Yet the presence of AI did not universally translate into higher average ratings for the collaborative memes. In other words, the sheer quantity of output did not automatically yield superior results; rather, it is the selective curation and refinement of AI-suggested ideas, guided by human judgment, that drives top-tier outcomes. This insight resonates with broader conversations in AI-assisted creativity: tools that increase output must be paired with thoughtful human oversight to preserve originality, relevance, and emotional depth.
Ownership and motivation also emerge as important psychological dimensions of AI-assisted creativity. Users working with AI reported feeling only modestly more detached from their outputs compared with solo creators, indicating that AI involvement can dilate the creative process without fully diffusing authorship. Given that a sense of ownership fuels motivation and satisfaction in creative tasks, the study’s authors suggest that researchers and tool developers should consider how to balance AI contributions with opportunities for human authorship and final decision-making. Calibrating this balance could help maintain creative engagement while leveraging the productivity benefits of AI.
Taken together, these findings illuminate a nuanced landscape where AI contributes to both the breadth and speed of meme production, while human expertise remains indispensable for depth, originality, and context-sensitive humor. The practical takeaway for content creators and platforms is clear: deploy AI for ideation and rapid iteration, but preserve human curation and final content selection to ensure memes retain distinctive voice, cultural relevance, and emotional impact. This hybrid approach aligns well with the observed pattern in which AI-generated content tends to be broadly appealing, while the most remarkable memes arise from human creativity or effective human–AI collaboration.
Productivity, Ownership, and Collaboration Dynamics
A central takeaway from the study concerns the productivity dynamics of AI-assisted meme creation. When participants used AI assistance, they reported a marked increase in the number of meme ideas generated and described the process as easier and less laborious. The AI partner functioned as a prolific ideation engine, supplying a wide array of linguistic options, phrasings, and framing angles that could be quickly evaluated by human creators. This productivity uplift aligns with expectations about AI’s capacity to process vast textual patterns and propose alternative phrasings that human creators might not consider within traditional brainstorming windows.
Yet there is a trade-off to consider. The same study highlights that the increased output from human–AI collaborative workflows did not translate into higher average quality on the evaluative metrics. In essence, more ideas does not automatically equate to better memes. The researchers summarize this dynamic with a concise observation: the increased productivity of human–AI teams yields more content, not necessarily superior content. This distinction is critical for practitioners who might otherwise equate quantity with quality. It suggests that the value of AI assistance lies not merely in volume but in its strategic use to surface high-potential ideas that a human curator can select and refine.
Ownership emerges as another nuanced dimension of AI-assisted creativity. Respondents using AI tools reported a slight decrease in perceived ownership over their final memes compared with solo creators. This sense of ownership—closely tied to creative motivation and personal satisfaction—can influence an artist’s willingness to invest time and effort into the work. When creators feel less ownership, there is a risk that they may disengage or rely more heavily on AI suggestions, potentially eroding the unique voice that marks standout memes. The study’s authors propose that AI tool designers and organizations should consider mechanisms to preserve or even enhance perceived ownership in AI-assisted workflows. Possible strategies include transparent attribution for AI contributions, more explicit human curation roles, and user interfaces that foreground the human author’s decision-making as the final arbiter of the meme’s content.
The study’s findings also pose important considerations for the broader adoption of AI in creative tasks beyond memes. If AI accelerates ideation and enables rapid production of concept-rich outputs, content teams can scale up production, test more variations, and iterate more quickly. However, this acceleration must be balanced with safeguards that ensure the output remains emotionally resonant, culturally sensitive, and aligned with the creator’s intended voice. The researchers imply that the most compelling outcomes arise when AI acts as a collaborator rather than an autonomous ideator, offering suggestions that humans can validate, adapt, and personalize.
From a workflow design perspective, several practical implications emerge. First, prompt engineering and structured ideation prompts can help steer AI toward outputs that align with the creator’s sensibilities and target audience. Second, design features that encourage quick human curation—such as side-by-side comparisons of AI-generated options, one-click editing suggestions, and easy toggling between AI drafts—can streamline the process and reduce friction. Third, explicit ownership cues within the final meme, including author credits and transparent delineation of AI assistance, may mitigate concerns about authorship and enhance creator motivation. Fourth, a hybrid model—where humans set the creative brief, guide AI exploration, and perform final edits—seems best positioned to maximize both productivity and quality.
The collaboration dynamic also invites reflection on how AI should be integrated into creative teams. Rather than substituting human decision-making, AI can be positioned as an augmentative partner that expands the creative palette. In meme creation, where the line between cleverness and cliché can be thin, AI’s capacity to generate high-volume, diverse caption options can be especially valuable for producers who need to respond quickly to trending topics, cultural moments, or platform-specific formats. By coupling AI’s breadth with human sensitivity to nuance and context, teams can cultivate a steady pipeline of content that is both scalable and distinctive.
Limitations, Biases, and Critical Perspectives
No study is without limitations, and this research acknowledges several that shape the interpretation of its conclusions. The caption-creation sessions were relatively brief, which may constrain the depth of exploration into long-form humor strategies or more complex joke structures that require extended development. Longer or repeated sessions could reveal greater potential for improvement through AI–human collaboration or prompt refinement that might translate into higher-quality memes over time.
Crowdsourced evaluation, while pragmatic and scalable, introduces subjectivity and the possibility of bias toward mainstream humor. Evaluators’ cultural backgrounds, personal tastes, and exposure to internet memes can influence ratings, possibly skewing results toward content that conforms to widely shared norms. This bias could advantage AI-generated captions that are optimized for broad appeal while underrepresenting more niche, subcultural humor that may be highly impactful within specific communities. The study notes that future work could counterbalance this by incorporating expert panels or targeted demographic groups to better capture subtleties in humor, culture, and creativity across diverse audiences.
Another limitation concerns the ecological validity of the experimental setting. While meme templates were chosen to be familiar, the controlled environment may not perfectly mirror real-world meme creation processes, where creators have access to a larger repertoire of templates, image sourcing, editing tools, and audience signals. The AI models’ performance in unconstrained, real-world conditions could differ from laboratory-style evaluations, where reaction times, prompt quality, and collaboration dynamics can shape outcomes. The researchers suggest that future studies explore longer-term collaborations, richer interactivity with AI systems, and more varied templates to approximate authentic production pipelines more closely.
A related concern concerns model semantics and cultural specificity. The AI’s success in achieving high average humor and shareability scores likely rests on broad pattern recognition rather than a granular understanding of social norms in every subculture. In this sense, AI’s strengths may lie in mimicking widely recognized humor styles rather than consistently producing culturally sensitive or contextually precise jokes. As humor is deeply rooted in contingent social cues, future research will need to probe how to tailor AI humor more precisely to local contexts while preserving the broad appeal that currently characterizes these captions.
The study explicitly invites future exploration of AI ideation at scale, including scenarios where the AI rapidly generates multiple ideas for humans to curate. Researchers propose that this “idea sprint” approach may unlock new dynamics in human–AI collaboration, where rapid generation is complemented by human selection, refinement, and ethical considerations in humor and representation. Such future work could also examine cross-cultural humor, multilingual contexts, and content that challenges harmful stereotypes, offering a more comprehensive understanding of AI’s role in creative expression.
In addition, the research underscores the distinction between AI-generated images and AI-generated captions. The study’s design did not involve AI-produced visuals, focusing instead on linguistic generation. While this clarifies the scope of the AI’s contribution, it leaves open questions about what happens when image generation, visual aesthetics, and captioning are simultaneously AI-enabled. Exploring integrated AI pipelines that jointly optimize both visuals and captions could yield additional insights into how AI can influence meme effectiveness across the entire medium.
Despite these limitations, the study’s results contribute valuable evidence to ongoing debates about AI’s role in creative tasks. They provide a measured counterpoint to sensational narratives that either celebrate AI as a universal creator or condemn it as a threat to human artistry. Instead, the findings point toward a collaborative future in which AI acts as a powerful partner for ideation, testing, and rapid content generation, while humans provide the qualitative judgment, cultural sensitivity, and personal touch that elevate memes from clever to profoundly resonant.
Implications for Meme Culture and AI in Creative Tasks
The study’s outcomes carry broad implications for meme culture, social media practices, and the adoption of AI in creative workflows. The demonstrated ability of AI-generated captions to achieve high average humor and shareability suggests that AI can play a substantial role in helping content creators scale their output and test a wider array of humorous approaches. For social platforms that prize rapid, engaging content, AI-assisted captioning could become a standard tool for creators seeking to maximize reach while maintaining stylistic consistency.
However, the tendency of AI to optimize for broad appeal also raises concerns about homogenization. If AI systems increasingly generate captions that fit mainstream tastes, there is a risk that meme ecosystems could become more uniform, with distinct voices from individual creators or niche communities subsiding as AI-driven patterns dominate. This potential drift toward uniformity could dampen the diversity and subversive edge that often make memes culturally impactful. Content creators and platform designers may need to implement strategies that encourage originality and highlight unique perspectives, ensuring AI tools augment rather than erode distinctive voices.
Ethical and copyright considerations also surface in AI-assisted meme creation. When AI models contribute to content that is widely shared and monetized, questions arise about attribution, ownership, and remuneration for AI-generated ideas, prompts, and creative contributions. While the study notes that ownership perceptions can waver with AI involvement, practitioners will need clear policies and practical workflows that recognize human authorship and, where appropriate, provide transparency about AI assistance. Transparent crediting and user empowerment in deciding how much influence AI can have on the final meme could help sustain creator motivation and trust.
For marketers, brands, and educators, the implications extend beyond entertainment. Memes have become a rapid channel for information dissemination, brand storytelling, and social commentary. If AI-generated captions can reliably boost engagement while preserving emotional resonance and cultural sensitivity, organizations may increasingly deploy AI-assisted meme creation as part of broader communication strategies. Yet this also raises the bar for accountability: as the boundary between human and machine authorship blurs, it becomes essential to maintain ethical standards and ensure that humor remains respectful and inclusive.
The cultural impact of AI-assisted humor is another fertile ground for discussion. Memes often reflect collective mood, current events, and shared jokes that circulate within communities. AI’s capacity to identify patterns that resonate across large audiences can help surface topics with broad appeal, but creators must remain vigilant about misinterpretations, stereotypes, and misrepresentations that can inadvertently provoke harm. The study’s emphasis on context-specific humor underscores the importance of combining AI’s statistical prowess with human discernment to navigate nuanced social landscapes responsibly.
Finally, the study provides actionable guidance for practitioners who want to harness AI in meme creation without sacrificing creativity or human agency. The recommended approach centers on a balanced workflow: use AI for rapid ideation and initial drafting, then apply human judgment to curate, refine, and finalize. This model protects the integrity of voice, ensures cultural and situational relevance, and preserves a sense of authorship. It also aligns with the broader trend toward AI-assisted creativity across industries, where iterative collaboration and careful curation can yield outputs that are both scalable and meaningful.
These implications extend to education, media literacy, and digital culture studies as well. As AI becomes more embedded in everyday creative tasks, there is a growing need to examine how people interact with AI assistance, how they interpret AI-influenced memes, and how audiences respond to content with hybrid authorship. The study’s results offer a data-driven foundation for these discussions, illustrating both the potential and the caveats of AI-enabled humor in real-world social ecosystems.
Future Research Directions and Practical Takeaways
Building on the study’s findings, several avenues for future research emerge. One key direction is to explore longer and more iterative collaboration cycles between humans and AI. By extending caption-creation sessions and enabling more nuanced interactions with AI prompts, researchers can assess whether prolonged collaboration enhances the quality of AI-assisted memes beyond what is observed in shorter experiments. This could reveal whether deeper engagement with AI tools helps close the gap between average AI performance and the top-tier memes produced by humans or human–AI teams.
Another area for exploration is prompting strategies and prompt-design methodologies. By refining how humans interact with AI—through structured prompts, targeted constraints, or domain-specific templates—creators can potentially steer AI toward outputs that align more closely with nuanced audience preferences. Investigating how prompt design influences humor style, cultural sensitivity, and contextual relevance could yield practical guidelines for optimizing AI-assisted meme creation in diverse settings.
Diversity and cultural specificity should also be foregrounded in future work. Studies could examine humor across different languages, regions, and cultural subgroups to determine how AI models perform in multilingual or multicultural contexts. This line of inquiry is crucial to ensure AI-generated humor remains respectful, accurate, and resonant across a wide range of communities.
Methodological expansion is another objective. Researchers might incorporate expert panels—comprising comedians, language professionals, and cultural critics—alongside crowdsourced evaluators to triangulate judgments about humor and creativity. Mixed-methods approaches, combining quantitative ratings with qualitative analyses of captions and memes, could yield richer insights into what makes humor land in particular contexts and how AI can be tuned to support or enhance those dynamics.
From a practical standpoint, creators can take away several actionable strategies. First, adopt a hybrid AI workflow: leverage AI to generate a broad array of caption options and seed ideas, then let a human creator curate, tailor, and finalize the content to preserve voice and nuance. Second, design AI-assisted tools that emphasize ownership and collaboration, offering clear attributions and interactive interfaces that foreground human decision points. Third, experiment with workflow configurations that balance speed and depth—for example, using AI for ideation in the early stages and reserving more time for thoughtful refinement as memes approach publication. Fourth, remain vigilant about cultural sensitivity and ethical considerations, ensuring AI-generated humor does not perpetuate stereotypes or cause harm, and that content aligns with platform expectations and community guidelines.
In conclusion, the study’s nuanced findings offer a practical blueprint for integrating AI into meme creation. AI-generated captions can boost average humor, creativity, and shareability, increasing content volume and broad-reaching appeal. Yet the most compelling memes—the ones that stand out for wit, originality, and emotional impact—still belong to human creators or to productive human–AI collaborations that combine the strengths of both sides. The path forward is not one of replacement but of thoughtful collaboration: AI as a powerful accelerator of ideation, and humans as the judges, curators, and soulful storytellers who give memes their distinctive voice and lasting resonance.
Conclusion
AI-generated captions for well-known meme templates show a clear advantage in average humor, creativity, and shareability compared with captions produced by humans alone. However, the most memorable and outstanding memes still emerge from human creators or from human–AI collaborations, indicating that human nuance, personal experience, and cultural awareness remain indispensable for content that resonates deeply. The findings point to a practical, hybrid approach to meme creation: harness AI to rapidly generate a broad set of caption ideas, then apply human judgment, refinement, and curation to select and finalize the most compelling memes. This synergy—combining AI’s pattern recognition and ideation capacity with human creativity and contextual insight—offers the most effective path for producing memes that are not only widely engaging but also culturally nuanced and emotionally resonant. As AI continues to evolve, creators, platforms, and researchers should embrace this collaborative model, continually refining prompts, workflows, and attribution practices to foster innovation while preserving originality and accountability in digital humor.
