AI in Communications: Insights from Anthropic's Economic Index

Last week, AI powerhouse Anthropic released the second edition of its Economic Index report, which analyzed 1 million anonymized Claude.ai conversations to offer objective measurements about AI's impact on work. In addition to the report, the company released the full dataset for further research and analysis. 

This type of data is timely, as internal comms pros experiment with AI and consider how greater investments in AI can influence their work, team and impact.  According to Contact Monkey's 2025 Global State of Internal Communications survey, 70% of respondents use some form of AI in their internal communications work, with ChatGPT being the most commonly used tool. AI ranked as the third highest topic internal comms pros are "most paying attention to” in 2025.

What stands out about Anthropic's analysis is that it shows actual usage patterns, i.e., potentially revealing not just if internal comms pros are using AI, but precisely how they're employing it compared to other professionals.

Intrigued, I dove into the data to see what, if anything, the report said about internal comms work. While I found limited data on internal comms tasks, there are some interesting points worth considering. 

Understanding the data

The latest Anthropic report analyzes data from the 11 days following the launch of Claude 3.7 Sonnet, covering 1 million anonymized Claude.ai Free and Pro conversations. 

Using a privacy-preserving analysis tool called Clio, Anthropic maps each conversation to tasks in the US Department of Labor's O*NET database, categorizing them into over 17,000 specific tasks. While Anthropic hasn't published detailed information about Clio's accuracy in categorizing conversations, this approach attempts to generate quantitative insights about AI usage across professions, industries, and tasks. The reliability of these categorizations, particularly for nuanced fields like internal communications, remains a limitation to consider when interpreting the findings.

To interpret the data, it's helpful to understand Anthropic's terminology:

• Automation: Users directing AI to complete a task (e.g., "write this newsletter") - measured by the percentage of conversations where users give direct instructions

• Augmentation: Users collaborating with AI in more interactive ways - measured as the combined percentage of learning, task iteration, and validation interactions

• Task Iteration: Drafting with AI, then refining outputs together (25% of conversations) - identified by multiple back-and-forth exchanges refining content

• Learning: Using AI to gather information or get explanations (28% of conversations) - identified by question-answer patterns focused on information gathering

• Directive: Giving AI specific instructions with minimal human involvement (30% of conversations) - measured by tasks completed with minimal user refinement

• Validation: Asking AI to verify or evaluate something (4% of conversations) - identified by requests to check or verify content

To focus on internal comms-related data, I located internal comms tasks such as:

Email communication tasks, such as

• "Draft professional emails about technical or business matters"

• "Draft professional business emails for various workplace communications"

Workplace communication improvement tasks:

• "Help me improve workplace communication and meetings"

Other internal communications tasks that were not specifically named Anthropic’s Index but were identified through keyword search, including tasks related to:

• Employee communications

• Staff updates

• Internal memos

• Company announcements

• Workplace communications

• Employee engagement

While internal communications tasks represented only 1.1514% of all analyzed records, this equates to approximately 11,513 conversations/sessions with Claude.ai, providing a sizable set of real-world AI usage to examine. 

Finding #1: Internal comms tasks show higher AI engagement

Internal communications tasks demonstrate significantly higher AI engagement than average, with:

• Tasks related to internal communications involve direct AI handling (automation) almost twice as often as other tasks - 35% of the time compared to just 20% for tasks in other categories.

• For collaborative AI work (augmentation), internal comms tasks involve working alongside AI nearly twice as much as average - about 64% of these interactions involve collaboration, compared to only 35% for tasks in other areas.

That represents a 78% higher automation rate and 84% higher augmentation rate than the average across all professions in the dataset.

All this underscores the notion that internal comms is a particularly AI-ready field, perhaps stemming from our content-heavy work. The data also indicates that AI tools are finding a natural fit in internal communications workflows, where content creation and refinement are constant requirements. Without historical data, however, it's unclear whether this represents a significant shift or a snapshot confirming strong usage by AI (in the case, Claude) early adopters and enthusiasts. 

Finding #2: Internal vs. external comms show different patterns

The data reveals distinct differences in how internal and external communications tasks utilize AI:

• Internal communications tasks show a greater tendency toward collaborative AI use rather than hands-off automation vs. external communications tasks.

• When looking at learning patterns, internal communications tasks involve using AI as an information resource almost twice as often as external communications tasks.

• Verification and checking activities with AI appear about 70% more frequently in internal communications tasks than in external communications work.

• Internal communications tasks involve less troubleshooting or error correction with AI compared to external communications tasks.

Within external communications, there are notable variations:

• PR/media relations tasks have higher automation (46.3%) and a lower augmentation-to-automation ratio (1.14)

• Social media tasks have lower automation (34.6%) and a higher augmentation-to-automation ratio (1.95)

The data suggests that internal communications work involves additional information gathering and verification when using AI compared to external comms. The higher learning and validation patterns suggest internal communicators are more likely to use AI as an informational resource and to verify content accuracy – perhaps reflecting the need to ensure messaging aligns with organizational knowledge and policies. 

Finding #3: Content type significantly influences AI usage

Analyzing the data by content categories reveals distinct patterns in how internal communicators use AI:

Policy Documents: When working with policy content, only about 18% of the tasks involve letting AI work independently. The remaining 82% of the time, people work alongside the AI. For every task where AI works alone, there are about 2 tasks where humans and AI collaborate.

• Policy Documents: When working with policy content, only about 18% of the tasks involve letting AI work independently. The remaining 82% of the time, people work alongside the AI. 

• General Internal Communications: For everyday internal communications content, AI handles tasks about 20% of the time independently. Most work (80%) involves some level of human involvement and collaboration with the AI.

• Newsletters: Newsletter-related tasks allow AI to work independently more often - about 32% of the time. In addition, AI is most often used as a drafting partner for newsletters - over half of newsletter tasks involve an iterative process where humans refine AI-generated content.

• Technical Content: When dealing with technical information, AI works independently on about 28% of tasks. Technical content shows the highest overall AI involvement (74% total), suggesting this type of content particularly benefits from AI assistance, whether working independently or collaboratively.

What this means: Different types of internal communications content appear to involve different approaches to AI usage. Policy communications show the strongest preference for human oversight – possibly because of their sensitive and compliance-heavy nature, though the data doesn't confirm causation. Policy content typically requires careful vetting and precise language that balances clarity with legal accuracy, which may explain the higher human involvement.

Newsletters, by contrast, have the highest automation potential, suggesting routine newsletter production contains more standardized elements that a person could delegrate to AI. Technical communications show high overall AI engagement (74.2% combined), likely reflecting how AI can help with technical content's specialized language and complexity.

* * *

The Anthropic data confirms people are using AI as a partner for internal comms tasks, and it sheds light on what that partnership looks like. While there’s a reasonable fear that AI may one day replace internal comms work by humans, at least for the moment, AI usage in internal comms looks like a quintessential workplace activity: teamwork. 

This analysis is based on Anthropic Economic Index data from February-March 2025, covering the 11 days following Claude 3.7 Sonnet's launch – a relatively short period that may reflect initial experimentation rather than established patterns. I used Claude to help analyze the original dataset with Jupyter notebooks, identify key patterns relevant to internal communications professionals, and interpret the findings. Longer-term data will be valuable to confirm whether these patterns persist over time.