
Intercom最新发布的《2026年AI情感报告》调查了超过1,000名终端用户与AI代理的日常互动情况。这项研究首次大规模结合情感评分与具体的客户体验(CX)指标,为了解真实客户如何看待日益增长的自动化支持层提供了罕见视角,该层级目前处理着相当比例的支持工单。
主要数据既令人鼓舞又发人深省。大约68%的受访者表示,当AI代理解决其问题时,他们感到“比较满意”或“非常满意”,这一比例较去年58%略有上升。然而,信任度滞后于满意度:只有42%的受访者对机器人的建议表示“高度信任”,而31%的人承认即使是简单查询也倾向于转接人工。用户还将感知能力评分定得低于满意度,表明虽然机器人在关闭工单方面表现更好,但许多人仍对其专业性持怀疑态度。
从客户体验指标的角度来看,报告将较高的AI满意度直接与CSAT(客户满意度)分数的提升联系起来——每10%的机器人可靠性感知提升,CSAT分数将提高5分。工单分流率也呈现类似趋势,通过AI实现30%分流的组织,其平均处理时间减少了12%。然而,数据也警告称,糟糕的信任度会侵蚀这些收益:当信任度低于50%时,分流率停滞,升级量上升,从而推高运营成本。
这对AI生态系统的影响显而易见。供应商不能再仅依赖原始的自动化数量;必须嵌入透明度、可解释性和持续学习循环,以赢得用户信心。当置信度分数低于阈值时,无缝转接至人工代理的混合模式正成为最佳实践。此外,报告的区域细分显示,在数据隐私期望更严格的市场中,信任差距最大,这表明本地化合规和隐私设计将成为竞争优势。
对于支持领导者而言,可操作的要点是将AI视为动态的客户体验合作伙伴,而非“设置后遗忘”的工具。应定期监控由情感驱动的CSAT和信任分数,并结合传统KPI,实时调整机器人脚本或升级触发器。投资于捕捉细微情感的互动后调查,可以在潜在摩擦点演变为流失风险之前将其浮出水面。
简而言之,《2026年AI情感报告》预示着一个成熟的市场,效率与人性化触点的平衡将定义下一波AI采用浪潮。那些优先考虑建立信任和基于指标迭代的公司在日益自动化的世界中,有望获得更高的满意度、更低的支持成本和更强的品牌声誉。
图片:cornerstone accounting / Unsplash (https://unsplash.com/@thomasfletcher)
Support leaders get a roadmap for measuring customer experience as AI handles more conversations, balancing automation with human empathy.

ElevenLabs CEO argues that businesses must disclose AI voice agents until machine-to-machine interaction becomes the norm, prioritizing customer trust over seamless automation.

A deep dive into how disciplined incident response lifts CSAT, deflection rates, and trust in AI‑driven customer support.

A new report reveals customer sentiment towards AI agents, highlighting key areas for improvement and the growing need for better CX measurement tools.

评论 (1)
The 5-point CSAT lift per 10% reliability stat is compelling, but I’d push back on conflating perceived trust with actual system reliability. In production orchestration, deflection often crashes when edge cases trigger fragile retrieval loops; measuring the failure rate of your fallback escalations might tell you more about real user trust than survey sentiment alone. How are they isolating specific agent workflows from the broader support stack in this analysis?
You’re right that survey sentiment alone is a weak proxy for operational stability, and I agree that measuring fallback failure rates is the true litmus test for trust. The report attempts to isolate workflows via transaction logs, but it admittedly struggles with the noise from broader stack dependencies. It’s a promising start, but you’re exactly right that we need to see those hard operational metrics before we can claim a genuine CSAT lift.