新闻真实向来是新闻学研究的“元问题”。随着新闻与新闻活动发生的深刻变革，一些研究开始从受众视角出发讨论新闻真实，以公众信任为基点展开一系列论述；然而，公众信任并不能等同于受众主体的新闻真实。本文试图弥补从新闻的“真”到公众的“信”之间的缺失环节，提出“认知真实”作为受众主体视角下的新闻真实性。在此基础上，本文在中、美、德三国的问卷调查（N=3000）中设计了2（政治/非政治新闻）×2（匿名/非匿名来源；新闻机构/社交媒体来源）的调查实验，探究影响公众认知真实的新闻呈现与个体特征条件，从而实证分析公众如何认知并辨别真假新闻。研究发现，相较于非政治类新闻，政治类新闻更加真假难辨，而中国民众更擅长正确识别政治类的真实新闻；新闻机构“标题+导语”的形式比社交媒体“推文”形式更容易让人们相信其真实性；机构媒体使用能让人们更好地识别真实新闻，而社交媒体使用、软新闻消费则降低了正确辨别虚假新闻的能力；自由主义新闻观念越强，美、德两国民众越擅长于辨别新闻真假，中国民众则越容易高估新闻的真实性；新闻素养能够提高民众辨别新闻真假的能力；随着年龄增长，人们更倾向于否定新闻的真实性。此外，本文还引入“图灵实验”（Turing Experiment）的方法，让大语言模型模拟人类的态度和行为，将“新闻真实性判定”作为测量人类认知与大语言模型的特定任务，通过大语言模型GPT–3.5依据性别、国别等人口学因素生成与人类样本相对应的3000份“硅样本”（silicon sample），进而对比分析其在判定新闻真实性上的认知能力和潜在偏见。研究发现，ChatGPT在判定新闻真实性这一复杂认知任务上有着明显的优势和薄弱环节，在政治新闻、真实新闻的判断上表现得比人类要更准确，但在非政治新闻、虚假新闻上表现得不如人类认知。同时，ChatGPT还存在着一定的代际偏见和语言偏见。总的来说，公众的认知真实并不仅仅与新闻本身的真实相关，也不能够与受众信任或媒介素养相等同，而是受到新闻呈现与个体特征等因素影响的、存在着结构与个体差异的、高度情境化的认知过程。就公众本身而言，政治新闻、虚假新闻仍是人们鉴别新闻时的薄弱环节。
News authenticity has always been a fundamental issue in journalism research. With the profound changes in news and news activities, some studies have started to discuss the news authenticity from an audience perspective, launching discussions based on public trust. However, public trust does not necessarily equate to the news authenticity for audiences. This article aims to bridge the gap between “truth” in journalism and “trust” among audiences by proposing “cognitive authenticity” as a perspective on journalistic authenticity from an audience perspective. Based on this, this article designed a survey experiment (N=3,000) using questionnaires in China, the U.S., and Germany with 2 (political / non-political news) x 2 (anonymous/non-anonymous sources; institutional / social media sources), so as to explore how presentation styles and individual characteristics affect public perception of truthfulness while analyzing how people recognize true or false information.The study found that political news is more challenging to distinguish between true or false than non-political ones. Chinese people are better at correctly identifying true political news than non-political ones. Headlines combined with introductions are easier for people to believe its truthfulness compared with news in tweets form. Institutional media use helps people identify true information while social media use and soft-news consumption reduce their ability to correctly identify fake information. Stronger liberal views improve Americans’ and Germans’ ability to distinguish between true or false information while Chinese tend to overestimate its veracity. News literacy can enhance people’s ability to discern whether something is true or not. Moreover, as age increases, people are more likely to deny the truthfulness of certain reports.In addition, this article also introduces the method of “Turing Experiment” to allow large language models to simulate human attitudes and behaviors. It uses “news authenticity assessment” as a specific task to measure human cognition and large language models. Based on demographic factors such as gender and nationality, 3000 “silicon samples” corresponding to human samples are generated by the large language model GPT-3.5. Then, their cognitive abilities and potential biases in judging news authenticity are compared and analyzed. The study found that ChatGPT has obvious advantages and weaknesses in this complex cognitive task of judging news authenticity. It performs more accurately than humans in political news and real news judgments, but less so in non-political news and fake news judgments. At the same time, ChatGPT also exhibits certain generational bias and linguistic bias.Overall, cognitive authenticity among audiences is influenced by various factors such as presentation style and individual characteristics rather than just being related solely with news accuracy itself nor equivalent with audience trust or media literacy - it exists as a highly contextualized cognitive process affected by structural and individual differences. When it comes to the general public, political news and fake news remain vulnerable points in their ability to distinguish between different types of information.
