Embedding AI into the creation of cultural tourism promotional videos can enhance the emotional atmosphere and artistic appeal of city promotional videos, allowing audiences to genuinely feel the historical and cultural heritage. It can also incorporate local characteristics and creativity to enrich visual effects, and tailor personalized creations based on the city’s traits and audience needs, comprehensively displaying local charm ([4]).

In terms of production workflow, fully intelligent production can greatly simplify manual steps, increase efficiency, adapt to various promotional scenarios, and strengthen the atmospheric and cultural immersion of the video. Currently, AI-assisted production of cultural tourism promotional videos generally follows a data-driven, intelligent synthesis approach. The whole process can be divided into three parts: data collection and analysis, intelligent script generation, and intelligent video synthesis. In practice, data such as scenic spot images, local historical records, local chronicles, and online tourist reviews are first collected through AI crawlers and public data channels. Then, computer vision technology is used to sort and classify image materials, while natural language processing extracts cultural keywords and regional characteristic elements from textual materials, building a dedicated cultural tourism material repository. With this resource base, the system automatically generates a complete script including copy, shot sequence, background music, and voiceover, taking into account public aesthetic preferences, platform characteristics, and promotional themes. The narrative style can be flexibly adjusted according to local cultural characteristics, for example, historical landscapes like the Qilou buildings in Lingnan are more suitable for a humanistic, nostalgic expression, avoiding hollow content. For image production, generative AI tools such as Midjourney and Runway are commonly used to automatically generate images according to the script, and speech synthesis technology is used to record suitable voiceovers. Later steps such as editing and color grading can also be completed with one click, greatly shortening the production cycle of promotional videos ([8]). Industry reports and case studies indicate that using the above AI process can compress a traditional production cycle of weeks or months to just hours or days; the efficiency improvement claims in this section are based on such empirical evaluations.

Furthermore, another prominent advantage of AI in creating cultural tourism promotional videos is that it completely breaks away from the stereotyped pattern of traditional promotional content. In terms of innovation advantages, AI can be used to generate precise, customized content based on the regional characteristics and cultural heritage of different cities and cultural tourism landmarks, as well as the preferences, age groups, and aesthetic tastes of various audience segments. Whether it is a classic promotional style suitable for the general public or a fast-paced, trendy short-video format favored by younger audiences, AI can flexibly produce them. It can also incorporate current hot topics and support social interaction, so that cultural tourism promotion is no longer rigidly uniform, but truly tailored to people and places, greatly enhancing the appropriateness and dissemination effectiveness of the promotion.

Virtual digital human guides represent a new combination of AI and cultural tourism guidance. Using technologies such as 3D modeling, speech synthesis, and natural language processing, they create virtual guide images with human-like features, making guide services more intelligent and immersive.

The innovation of virtual digital human guides can be seen mainly in three aspects.

To begin with, innovation in appearance, the virtual guide can be designed to embody the unique cultural temperament of a locality, incorporating local costumes, folk customs, and cultural symbols to create a strong sense of identity and realism. Many scenic spots recreate historical figures as virtual agents; the “Zuo Zongtang AI Agent” is a typical example.

Next, functional upgrades, beyond simple commentary, virtual guides can answer questions intelligently, customize narration content, and plan tour routes. Tourists can converse directly with the virtual digital human; the system responds to queries in real time and can adjust the tour sequence based on visitor preferences, truly providing on-demand service. Moreover, its interaction model is completely different from traditional guides, and it is very flexible in application scenarios. It can be used on-site at scenic spots, in online exhibitions, or on short-video platforms, connecting online and offline guidance services. Tourists can access scenic area information at any time through mini-programs, short videos, holographic projections, AR, and other means.

Finally, innovation in the visitor experience: this new interactive method also narrows the distance between scenic spots and tourists, enhancing the overall immersive travel experience.

To ensure that virtual digital human guides provide accurate, reliable services, the support of large AI models is essential. Taking Jilin’s regional culture as an example, a dedicated small cultural tourism model was built locally, integrating seasonal views of Changbai Mountain, the formation of rime, Korean folk customs, and other local cultural tourism resources. This ensures that the guide content is professional and matches the regional characteristics. When tourists ask questions about tourism in Jilin, the system gives priority to local data for answers, effectively avoiding distorted or false information. This approach, which integrates general large models with cultural tourism-specific small models, offers a valuable technical reference for the local promotion and application of virtual digital human guides. The effectiveness of this case in improving guidance accuracy is mainly based on technical white papers and some media field reports, representing preliminary practical evidence.

As cultural tourism content production increasingly moves toward intelligent upgrading, cultural and creative products themselves are a crucial link in cultural dissemination and tourism consumption. They also need to leverage AI to improve efficiency, enhance the expression of regional characteristics, and meet diverse market demands. The value of AI in cultural and creative design is mainly reflected in three areas.

One key aspect is the intelligent extraction of cultural elements. Using computer vision, big data analysis, and other technologies, local traditional patterns, folk motifs, ancient artifact styles, and various intangible cultural heritage techniques are organized. Through digital recording and knowledge graph mapping, usable design elements such as shapes, colors, and decorations are selected, laying the foundation for subsequent creative work.

Another important function is the intelligent generation of creative ideas. Based on the organized cultural materials and combined with current aesthetic trends and market consumption preferences, AI can directly produce product shapes, patterns, and color schemes. These can be flexibly adjusted according to usage scenarios, target audiences, and budget. A collaboration between Guangxi Tourism Investment Group and Evolution Intelligence ([1]) is a good example: using AI, they combined the classic patterns of Zhuang brocade with modern aesthetics to launch cultural and creative products with local identity, providing new ideas for the contemporary expression of traditional culture.

A third value lies in design optimization based on feedback. AI can integrate market reputation and public evaluations to adjust the complexity of patterns and overall color matching in a timely manner, aligning with ordinary people’s aesthetic habits. This gradually creates a virtuous cycle of design and adjustment.

On this basis, the integration of AI design with flexible supply chains enables rapid connection from design to production, greatly shortening the production cycle. Evolution Intelligence’s AI design service can increase the speed of cultural and creative design by 90%, and its flexible supply chain, covering more than 2000 partner factories nationwide, can shorten the production cycle by 70%. These data on efficiency improvement are directly cited from corporate publicity materials and related news reports, representing cutting-edge commercial practices but requiring independent academic evaluation for validation. This allows local cultural tourism to create distinctive cultural IPs at low cost and high efficiency, solving the problems of long production cycles and high costs typical of traditional cultural and creative product development.

AI is driving a new model of cultural tourism content production characterized by data as the core, human-AI collaboration, and closed-loop optimization. This model differs significantly from traditional production models. To clearly show the differences between the two models, a comparison table is provided below (Table 1).

Table 1. Summary of the innovation model.

Despite the strong potential demonstrated by the above cases, it is necessary to recognize that AI’s applicability in the cultural tourism field is not unlimited. First, AI is highly dependent on structured data and clear rules; it struggles with cultural interpretation that relies on unstructured knowledge, tacit experience, and intuitive judgment. Second, in complex scenarios requiring deep empathy and dynamic interaction, such as comforting guide services, emotional support during crises, AI's mechanical nature appears rigid and cannot replace the warmth and adaptability of human guides. Finally, for sites carrying heavy historical memory or high sanctity, completely AI-generated content may raise ethical concerns; such scenarios still require strong human-led interpretation. Clarifying these limitations helps to adopt a rational view of AI’s role and avoid technological determinism.

AI-generated cultural tourism content often captures only the superficial forms of culture, failing to reach deeper meanings. Factual and common-sense errors are common.

On the stone tablet of the Yaotou Kiln Ancient Town in Weinan, Shaanxi Province, the “Ode to Yaotou Kiln” had the signature “DeepSeek” directly. This AI-written text was publicly displayed in the scenic area for nearly a year, sparking public discussion about the seriousness of humanistic expression ([2]). In an early AI curation experiment at the Nasher Museum in the United States, the system fabricated non-existent artworks from the museum’s collection and produced a fake exhibition list ([9]). Problems such as training data bias and algorithmic black boxes mean that AI output is often inaccurate or completely wrong ([7]). Cultural connotations are simplistically deconstructed and alienated; cultural heritage that should be treated with rigor and respect is reduced to arbitrarily combinable graphic materials. The value of cultural inheritance that cultural tourism content should bear is thus undermined.

The deeper causes can be analyzed from three perspectives: data foundation, algorithmic mechanisms, and machine capabilities.

Starting with the data level, most of the training materials currently used in the industry are scattered, raw graphic and textual materials. There is a lack of targeted annotation by professional cultural researchers regarding contextual background, key historical facts, and folk taboos. As a result, there is no truly standardized dataset suitable for deep cultural applications.

Moving to the algorithmic mechanism level, AI learns by capturing co-occurrence patterns among different elements in massive data. It only performs shallow feature matching and struggles to grasp the historical context and humanistic depth embedded in folk traditions and regional symbols.

Finally, at the machine capability level, machines lack human-like critical thinking and fact-checking habits. When generating content, they do not distinguish truth from falsehood or actively verify historical facts. They mechanically replicate common surface associations present in the data.

The combination of these factors is the deep-seated reason why AI misinterprets culture and distorts its expression.

Cultural tourism projects from different regions and cultural backgrounds produce promotional and design styles that are highly similar, gradually erasing the unique character and cultural identity of each place. This homogenization is quite evident in content forms.

In terms of visual imagery and copywriting, AI-generated promotions for ancient towns universally adopt old streets, flowing water, red lanterns, slow-motion shots, and classical music. The copywriting focuses on phrases like “millennium ancient town,” “idyllic life,” and “poetic dwelling”. Whether it is a water town in the south of the Yangtze River, an ancient town in the north, or a characteristic town in the southwest, the visual feel and textual expression are almost indistinguishable.

In the field of cultural and creative product design, AI-involved outputs mostly imprint traditional symbols onto everyday objects. What sells in scenic spots are mobile phone cases, bookmarks, and umbrellas engraved with patterns; museums print artifact motifs on canvas bags and keychains. There is little creative transformation that aligns with contemporary aesthetics. The design expressions of intangible cultural heritage and cultural relics are becoming increasingly similar.

As for virtual IP images, most scenic spots’ digital guides adopt a rigid, ancient scholar image. Their language style, explanation logic, and interaction techniques are largely the same, failing to create a distinctive local IP memory.

In essence, AI lacks genuine cultural creativity; it merely replicates mature paradigms that have been validated in the market, dragging cultural tourism content creation into a low‑level cycle of reproduction.

The root of this homogenization lies in AI’s training logic. On one hand, during learning, the model naturally tends toward patterns that appear more frequently, fixing well-known traditional landscapes and cultural symbols into standard templates. It ignores unique cultural details of different regions and minority groups, inherently limiting the space for diverse innovation. On the other hand, AI’s “creative” method essentially involves splitting and splicing elements from existing high-quality works. It cannot actively explore cultural cores and then reconstruct them in modern expressions like human creators can. Over time, it abandons differentiation and clings to safe, mature frameworks, constantly repeating itself, and homogenization becomes increasingly severe.

When AI models are iteratively optimized, the cultural tourism content they produce generally prioritizes traffic and market appeal, weakening cultural heritage, historical accuracy, and spiritual substance. This deviates from the core mission of cultural tourism, which is to educate and cultivate through culture. This deviation is clearly reflected in various aspects of content production.

In content selection and traffic allocation, AI-made cultural tourism short videos often deliberately chase after popular check-in spots, photogenic angles, and niche places that look good in pictures, while downplaying core content such as historical explanations, folk custom introductions, and intangible cultural heritage skills. When tourists are surrounded by such content for a long time, their experience is gradually reduced to taking photos and checking in; cultural reflection and spiritual experience are noticeably compressed.

In the relationship between commercial and public-interest aspects, commercial demands persistently crowd out the public-interest nature of cultural tourism. AI-generated promotional content tends to highlight tickets, consumption, dining, and paid experiences, while paying little attention to public cultural services, heritage protection, and popular science. Cultural tourism promotion gradually turns into pure commercial marketing.

In historical narrative, the entertainment of historical content has become increasingly prominent. When recreating historical figures or ancient scenes, AI often alters images, fabricates plots, and playfully narrates history just to attract attention. This deviation is not simply a matter of inappropriate style; it reflects that AI’s technical logic overrides cultural logic, gradually reducing cultural tourism content to an appendage of commercial traffic.

To understand the roots of this value deviation, we can analyze from three dimensions: data ecology, optimization objectives, and value judgment. From the data ecology perspective, the structure of publicly available online resources is inherently unbalanced. Entertainment content is abundant, while serious cultural heritage materials and academic research are limited. AI passively replicates this data distribution, so its creative tendency naturally leans toward entertainment. From the optimization objective perspective, AI only optimizes around quantifiable metrics such as traffic, interaction, and conversion. However, the cultural creation values of historical accuracy, respect for ethnic cultures, and transmission of spiritual values cannot be measured by data and thus hardly enter the model’s optimization scope. Finally, from the value judgment capability perspective, machines have no formed historical cognition or humanistic ethical boundaries. They cannot distinguish between the appropriate uses of authentic historical records versus artistic interpretation. They lack the sense of reverence necessary for cultural creation and can only imitate existing expression patterns. These three dimensions reinforce each other, and over the long term, cultural imbalance and value orientation deviation are almost inevitable.

Technical limitations are an important factor causing cultural connotation alienation and content distortion. Technological iteration is needed to improve AI’s ability to understand cultural connotations.

Initially, at the algorithmic transformation level, continuously promote algorithm innovation to shift AI algorithms from a data-statistical type to a culture-understanding type. Use natural language processing, computer vision, knowledge graphs, and other technologies to build models with cultural understanding capabilities. Integrate regional cultural knowledge, historical background, folk customs, and other information so that AI can truly grasp the meanings behind culture, preventing cultural misinterpretation.

Next, at the data annotation and context construction level, integrate training datasets professionally annotated by cultural experts. Establish correspondences between cultural connotations and application contexts to improve the model’s accuracy in interpreting local culture.

Furthermore, at the generation precision level, focus on improving the precision and logic of content generation, reducing detail distortion. Explore directions such as multi-model integration and manual-annotation improvements to enhance the detailed depiction ability in cultural tourism promotional videos and cultural and creative designs.

Finally, at the application deployment level, rely on diverse scenarios such as museums, historical ancient towns, and intangible cultural heritage theme parks to advance technology pilots. Accumulate problems and solutions through real-world operation, continuously refine technical solutions, and steadily improve actual effectiveness and scenario adaptability.

Data is the carrier of AI-driven cultural tourism content production. Data quality and richness directly determine content quality and innovation. Current training data in the cultural tourism field is fragmented, repetitive, and lacks context. Improvements are urgently needed from multiple dimensions.

To begin with, establish a unified data standard system that sets rules for the collection, organization, storage, and sharing of cultural tourism data, specifying the scope, standards, and procedures for data collection.

Next, systematically organize and annotate existing historical documents, local chronicles, and cultural symbols to form high-quality datasets with cultural connotations and context.

Additionally, expand the scope of data collection, focusing on collecting text and image resources related to niche cultures, local characteristic cultures, and intangible cultural heritage. Process and augment them to avoid content homogeneity caused by identical data, enabling AI to create cultural and creative products with stronger unique characteristics and local cultural flavor.

Establish a sound human-AI collaboration mechanism that highlights the leading role of humans in grasping cultural connotations, guiding creativity, and reviewing content, so that humans and AI complement each other’s strengths.

First of all, clarify division of labor boundaries. AI should handle repetitive tasks such as data processing, first-draft generation, and basic editing. Humans are responsible for controlling cultural cores, adjusting creative directions, and final content review. This ensures that the generated cultural tourism content conforms to cultural connotations and value orientations. Humans should also modify AI-generated scripts for promotional videos and adapt to modern aesthetic and market needs, improving the design quality of AI-created cultural and creative products. Promote the integration of technology and culture, ensure that people use AI applications correctly, and prevent cultural risks.

Moreover, to make this collaboration work smoothly, process regulations alone are insufficient; talent reserves are needed. Universities can offer interdisciplinary courses on AI and cultural tourism to cultivate compound talents with both technical skills and cultural literacy.

In addition, at the review system level, establish a joint review panel composed of cultural experts and technical professionals. Assess AI-generated content from dimensions such as cultural authenticity, value orientation, and creativity. For content like scenic area narrations and intangible cultural heritage videos, strictly examine the accuracy of cultural connotations and resolutely prevent the dissemination of vulgar or distorted content.

A robust regulatory system can effectively prevent various risks that arise when AI is used to produce and disseminate cultural tourism content.

One crucial measure is to formulate industry standards that specify the standards, responsible entities, and review requirements for AI in cultural tourism content production. Set norms for the cultural authenticity and value orientation of AI-generated content, and prohibit the release of vulgar, distorted, or value-deviated cultural tourism content. Aim to establish industry standards for AI-generated cultural tourism content, stipulating that AI-generated content must be cultural in nature and cannot be published without human review.

Another key action is to strengthen the supervision of AI cultural tourism content production and dissemination by cultural tourism departments and cyberspace administrations. Guide cultural tourism enterprises and technology providers to establish a correct view of cultural development, adhere to cultural bottom lines, and standardize AI applications. Hold legally responsible persons accountable in cases of serious impact.

A further important step is to enhance industry self-discipline. Encourage cultural tourism enterprises and AI technology companies to establish a correct view of cultural development, adhere to cultural bottom lines, consciously avoid cultural risks, and proactively regulate AI application behaviors. Make cultural heritage the primary purpose of AI-driven cultural tourism content production, avoiding excessive use of AI technology just to chase traffic and commercialization. Cultural tourism enterprises can create internal self-regulation mechanisms to manage AI content production, thereby ensuring the cultural quality and value orientation of the content.