报告
Muck Rack 的「AI读什么？发现报道记者AI引用股份跌至22%
在 Google 上的首选来源中添加 Unite.AI
莫克·拉克(Muck Rack)的2026年10月，新闻在AI引用中的比例从5月的约27%下降到22%，第四版借鉴了超过4000万次AI回复和2.07亿次引用，跨越ChatGPT,Claude,Perplexity和Gemini.
发布 October 28，研究指出支持AI生成的答案的材料组合发生了改变。 新闻报道仍然是一个重要的来源，但更大份额的报导是提供给其他种类的第三方材料和品牌自己出版的内容。 对于出版商和通信团队来说，调查结果使能见度成为一个提供者、迅速和内容格式的问题。
新闻在收益来源仍为中央时失去份额
从27%下降到22%，约为5个百分点。 而不是新闻连线或通讯量的绝对减少。
区分新闻与媒体对了解结果至关重要。 包括学术研究、政府和非政府组织来源、百科全书网站、社会平台、第三方公司内容，因此，报告的数字大约为84%，代表着比专业新闻室更广泛的源头生态系统。
拥有的媒体从13.7%上升到15.2%，而政府、百科全书和学术材料也获得了份额。 付费和广告内容仍为0.3%，自5月起保持不变。 这种小的引文分享了这种网页作为来源的频率；它没有衡量广告的总体效力。
ChatGPT 驱动 Shift 的大部分
Muck Rack认为ChatGPT是新闻业份额下降的主要驱动力。 它的纵向图显示在2026年宣布了变化，9月的终点为22%. Claude和Gemini在大部分期间表现出相对稳定的图案。
这些差异对试图理解为何某一出版物或品牌出现在一个助手的答案中而不是另一个答案的人来说都很重要。 知名度的改变可以反映提供者不断变化的源选择，而不是出版商产出或相关性的类似变化。
调查结果涉及所收到的答复中引用的资料来源。 它们没有确定模型培训数据中包含哪些材料，引用份额不应作为读者人数或信任的直接衡量标准。
只有212个域出现在所有四家供应商中
在9月收集的每家供应商1 020个提示中，在所有四个系统中只出现了212个域。 这就是研究样本中的重叠之处，而不是这些服务能够利用的每一个来源。
结果削弱了将AI能见度作为单一排名来优化的理由。 一个提供商经常提到的安置办法在另一个提供方的覆盖范围可能有限。 同样，一个有用的公司网页和独立报道可发挥不同的作用，取决于所回答的问题。
报告登机页突出了另一个变化来源：再次问同一问题可以产生不同的引用。 其实际含义是，单一反应是判断可见度持续变化的薄弱基础。
什么方法可以——不能——显示
完整的报告和方法结合了专门的研究收集，主要使用无品牌的消费者提示，与基因脉冲的持续，主要是以品牌为重点的监测。 这些数据集服务于不同目的，不应作为统一的样本处理。
该报告还提供了研究期间的指出和退役模型。 因此，随着时间推移的比较反映了模型和产品的变化以及时间的流逝。 观测结果描述了经过测试的系统、问题和收集期。
Muck Rack先前对其研究方法的解释明确了迅速选择的重要性。 它描述了来自行业、品牌和现实世界问题的建筑查询，然后按媒体类别和内容格式对引用进行分类。 研究者们选择帮助确定他们观察到的来源。
出版社和通信小组的变动
即刻的教训是在提供者和问题层面跟踪源选择。 一组稳定的与观众相关的提示，随时间推移而重复，可以揭示出不止一次的出现。 与这些结果一起记录模型变化有助于区分编辑或分发效果与AI服务本身的变化。
对出版商来说，新闻业总体份额较低并不会消除其在解释新发展或评价产品和组织方面的作用。 对于品牌来说，自有内容的成长创造了另一条通向答案的路线，但第三方来源的持续突出意味着在一个公司网站上只发布来源生态系统的一部分。
2026年10月AI读什么？. 报告提供了全部调查结果和方法。 其核心信息是，大赦国际的引用主要基于第三方材料，而信息来源的组合则因提供者、问题和时间而大不相同。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Reports
Muck Rack’s ‘What Is AI Reading?’ Report Finds Journalism’s AI Citation Share Falls to 22%
Add Unite.AI to your preferred sources on Google
Muck Rack’s October 2026 What Is AI Reading? report finds that journalism’s share of AI citations has fallen from about 27% in May to 22%, even as earned and third-party sources continue to account for roughly 84% of citations. The fourth edition draws on more than 40 million AI responses and 207 million citations across ChatGPT, Claude, Perplexity, and Gemini.
Released October 28, the research points to a change in the mix of material supporting AI-generated answers. News coverage remains an important source, but a larger share is going to other kinds of third-party material and to content published by brands themselves. For publishers and communications teams, the findings make visibility a question of provider, prompt, and content format.
Journalism Loses Share While Earned Sources Remain Central
The drop from 27% to 22% amounts to roughly five percentage points. It describes journalism’s share of citations, rather than an equivalent decline in the absolute number of news links or traffic delivered to publishers.
The distinction between journalism and earned media is essential to understanding the result. Muck Rack uses a broad earned-media category that includes academic research, government and NGO sources, encyclopedic sites, social platforms, and third-party corporate content, alongside journalism. The report’s roughly 84% figure therefore represents a much wider source ecosystem than professional newsrooms alone.
Owned media rose from 13.7% to 15.2%, while government, encyclopedic, and academic material also gained share. Paid and advertorial content remained at 0.3%, unchanged from May. That small citation share measures how often such pages appear as sources; it does not measure advertising’s overall effectiveness.
ChatGPT Drives Much of the Shift
Muck Rack identifies ChatGPT as a major driver of the decline in journalism’s share. Its longitudinal chart shows pronounced changes during 2026 and an endpoint of 22% in September. Claude and Gemini show comparatively steadier patterns over much of the period.
Those differences matter for anyone trying to understand why a publication or brand appears in one assistant’s answers but not another’s. A change in visibility can reflect a provider’s changing source selection, rather than a comparable change in a publisher’s output or relevance.
The findings concern sources cited in generated responses. They do not establish what material was included in a model’s training data, and citation share should not be read as a direct measure of readership or trust.
Only 212 Domains Appeared Across All Four Providers
In the September collection of 1,020 prompts per provider, only 212 domains appeared across all four systems. This is the overlap within that research sample, rather than a count of every source those services can access.
The result weakens the case for treating AI visibility as a single ranking to optimize. A placement that is frequently cited by one provider may have limited reach in another. Likewise, a useful company page and independent coverage can play different roles depending on the question being answered.
The report’s landing page highlights another source of variation: asking the same question again can produce a different set of citations. Its practical implication is that a single response is a weak basis for judging a sustained change in visibility.
What the Methodology Can—and Cannot—Show
The full report and methodology combine dedicated research collections, primarily using unbranded consumer prompts, with Generative Pulse’s continuous, mostly brand-focused monitoring. These datasets serve different purposes and should not be treated as one uniform sample.
The report also notes that providers introduced and retired models during the study. Comparisons over time consequently reflect model and product changes as well as elapsed time. The observations describe the tested systems, questions, and collection periods.
Muck Rack’s earlier explanation of its research approach makes the importance of prompt selection explicit. It describes building queries from industries, brands, and real-world problems, then classifying citations by both media category and content format. The questions researchers choose help determine the sources they observe.
What Changes for Publishers and Communications Teams
The immediate lesson is to track source selection at the provider and question level. A stable set of audience-relevant prompts, repeated over time, can reveal more than a one-off appearance. Recording model changes alongside those results helps distinguish an editorial or distribution effect from a change in the AI service itself.
For publishers, journalism’s lower overall share does not eliminate its role in explaining new developments or evaluating products and organizations. For brands, the growth of owned content creates another route into answers, but the continued prominence of third-party sources means publishing on a company website addresses only part of the source ecosystem.
The October 2026 What Is AI Reading? report provides the full findings and methodology. Its central message is that AI citations remain grounded largely in third-party material, while the mix of sources varies substantially across providers, questions, and time.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.