iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Clio is Anthropic’s research platform for finding aggregate patterns in AI-assistant conversations without making human reviewers reading raw conversations the main route of analysis. It summarizes and groups conversations so researchers can study how people use assistants and support safety analysis. Its published usage figures describe specific samples from 2024—not a current census of Claude users—and its privacy design is layered and empirically evaluated, not a formal differential-privacy guarantee.
What Clio does
Clio—short for “Claude insights and observations”—uses AI assistants to summarize conversations, organize summaries into clusters, and surface patterns across those clusters. Researchers can then explore topics and relationships through a hierarchy and visualizations rather than manually reading raw conversations as their primary analysis method. Anthropic describes the platform as a way to investigate questions such as which capabilities people use, how use varies across communities, and which expected benefits or risks appear in practice.
The authors’ paper describes Clio as analyzing patterns across millions of conversations. That does not mean every published result represents every Claude product, user, language, or period: each finding depends on the data source and sampling choices behind it. The paper’s methods and results provide the basis for interpreting its figures.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →What the published usage analysis found
For its general Claude.ai usage analysis, the paper sampled 1,000,000 conversations from Free and Pro users collected October 17–24, 2024. The published individual-use analysis excluded Team, Enterprise, and API customers. In that sample, web and mobile application development accounted for over 10% of conversations; writing, research, and educational use each accounted for 6–10%.
#1 Best Overall
These are shares of conversations in that defined sample, not shares of all users, all messages, or Claude use today. The paper also reports a multilingual analysis of 2,281,911 conversations sampled October 24–November 13, 2024, and describes language-linked topic differences, including elevated elder-care discussion in Japanese and Chinese conversations. The multilingual sample and the Free-and-Pro usage sample have different scope and dates, so their figures should not be treated as one interchangeable population.
The authors also note that observed mixes of non-coding use differ among Claude.ai, LMSYS-1M-Chat, and WildChat. Platform, sampling method, language, and period all affect what appears in the data; a finding from one dataset should not be generalized automatically to other assistants or user populations.
Rank #2
How Clio’s privacy protections work
The paper describes several safeguards applied at different stages of the analysis pipeline. Together, they are intended to reduce the chance that summaries or clusters expose private information:
Free tools Windows power users keep installed
One-click scans. No signup required.
- Conversation summaries: The model is instructed to omit private information when summarizing individual conversations.
- Cluster thresholds: Clusters that do not meet minimum requirements for unique accounts and conversation counts are discarded.
- Cluster summaries: The model is instructed to omit private information again when it summarizes a group of conversations.
- Audit and removal: Cluster summaries are audited, and clusters containing private information are removed.
Anthropic’s Privacy Center explanation, dated March 16, 2026, says research-use outputs are anonymized, aggregated broad trends and that this research use provides no employee access to raw user conversations or customer-specific data. This describes the aggregate research use, not every operational use of Clio.
Rank #3
Aggregate research and safety monitoring are different
Clio is also used for safety-focused analysis, including monitoring for coordinated abuse, surfacing unexpected issues around launches or world events, and examining safety classifiers. The paper gives an example in which analysis helped identify automated accounts involved in search-optimized spam, which were then banned.
Anthropic distinguishes these safety-focused runs from research that produces aggregate usage insights. According to its Privacy Center, safety-focused results can be linked to individual accounts, with access limited to a small number of authorized staff. The account-linked safety workflow should not be conflated with the research-oriented aggregate outputs described above.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the validation does—and does not—show
The paper reports 94% accuracy in reconstructing topic categories on a synthetic dataset with a known topic distribution. This is evidence about performance on that benchmark; it is not a guarantee that Clio captures the same accuracy for every real-world population, topic, or language.
The authors also report auditing 5,000 conversations as part of the paper’s privacy evaluation. They characterize their approach as layered and empirically evaluated, while explaining that formal guarantees such as differential privacy and k-anonymity are difficult to apply to rich textual summaries. The paper therefore does not establish a formal differential-privacy guarantee or prove that re-identification is impossible under every possible threat model.
How to interpret Clio’s results
- Check the population: Identify whether a result covers Claude.ai Free and Pro users, another platform, or a different data source.
- Check the dates and sampling: The paper’s figures come from defined 2024 collection windows, not a continuously updated public census.
- Check what is counted: A percentage of conversations is not necessarily the percentage of users or messages.
- Separate privacy mechanisms from guarantees: Clio uses multiple safeguards, but the paper presents empirical evidence rather than a simple formal guarantee for rich-text outputs.
- Distinguish research from safety use: Aggregate research outputs and account-linked safety-focused runs have different access models.
The paper also gives an estimated cost of $48.81 for an example Clio run covering 100,000 conversations. That is an estimate for the paper’s example run, not a current service price or a general cost schedule. Read the paper for the full sampling definitions, evaluations, and limitations.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
