Data sourceAnthropic Economic Index
How People Use LLMs for Finance in 2026
Financial advice is now the seventh most common work task in AI conversations worldwide, and the fourth most common in the United States. Here is what the Anthropic Economic Index shows about the money questions people actually bring to an AI model, by topic, by country, and by state.
Last updated: July 2026 · Data period: May 2026 · Reviewed quarterly
What the data says in three sentences
People use AI for money mainly as an advisor, not as an analyst. Worldwide, "advise clients or respond to inquiries about financial matters" is the seventh most common work task matched in Claude conversations, at 1.37% of all sampled conversations, ahead of software support tickets, copy editing and lesson planning.
In the United States that same task ranks fourth at 1.79%, and the published money topics (buying and investing, financial modeling, quant trading, accounting, business finance, financial education and taxation) together account for at least 5.9% of US conversations, roughly one in every seventeen.
The output people take away is advice: 15.1% of US conversations produce an "advice or recommendation" artifact, against a 10.7% global average.
Shares describe conversations, not users. Figures come from anonymized, aggregated conversation content and describe observed Claude usage matched to job-task descriptions, not employment or automation.
Money advice outranks almost everything except search and coding
The Economic Index matches conversation content against O*NET, the US government's catalog of what each occupation does. Ranked across every task in that catalog, giving financial advice sits at number seven worldwide, above product recommendations, student questions and copy editing.
Top 10 work tasks in AI conversations, worldwide
Share of all sampled conversations, May 2026
Source: Anthropic Economic Index, top 50 work tasks, May 2026 period. Task names shortened from their O*NET wording.
The framing that matters: this measures tasks, not people. The accurate reading is "AI is used for tasks commonly done by financial advisors", not "financial advisors are using AI". Most of the people asking these questions are not advisors at all. They are the clients.
What money questions people actually ask
The US topic tree splits the money stack into distinct branches. Buying and investing dominates, and investment research alone is larger than every trading, tax and accounting topic combined.
Money topics as a share of US AI conversations
Published request topics, May 2026
Source: Anthropic Economic Index, US request-topic hierarchy, May 2026. Starred rows are sub-topics: investment research sits inside buying and investing, personal finance and portfolio management inside accounting and financial modeling, taxation inside regulatory rules. Small and suppressed entries are omitted, so the real money share is higher than the published rows add up to.
Retail beats institutional, by a wide margin. The three biggest money topics in the US are buying and investing (2.57%), investment research (1.20%) and financial modeling (1.10%). Options trading registers at 0.10% and banking at 0.10%. The volume is in ordinary people deciding what to buy, not in trading desks.
Under buying and investing
Investment research 1.20%, vehicle selection 0.37%, product specifications 0.28%, real estate search 0.20%, mortgages and loans 0.12%, banking 0.10%.
Under financial modeling
Financial modeling 0.38%, portfolio management 0.32%, real estate investing 0.16%, financial analysis 0.10%, financial simulation 0.10%.
The output people walk away with is a recommendation
Artifact types record what a conversation produces. In the US, the gap against the world average is largest on exactly the artifact a financial question generates.
Source: Anthropic Economic Index, US country profile, May 2026.
The use-case split points the same way. Half of US conversations (49.74%) look like personal life rather than work, against 40.20% globally, while work drops to 41.32% from a 43.36% global average. Money questions in the US are arriving from the household side of the ledger, and they are arriving in the form "tell me what to do", not "run this model".
Where financial questions take up the most space
Japan and the United States lead on the financial advice task among the countries that publish it, both running well above the 1.37% world average. Comparison is by task name, not rank: a country missing from this chart means the task fell below publication minimums there, never that it was measured at zero.
Share of conversations matching the financial advice task
Selected countries with published figures, May 2026
Source: Anthropic Economic Index, country comparison, May 2026. Germany, Brazil, Singapore and Nigeria did not publish this task in their top-ten cut, which reflects publication thresholds rather than absence of the behavior.
The wider profiles differ more than the money numbers do. Singapore posts the highest Anthropic Usage Index of this group at 5.81, followed by the United States at 3.87, the United Kingdom at 3.35, Germany at 2.40, Japan at 1.91, Brazil at 0.96, India at 0.30 and Nigeria at 0.24. The index is a geography's share of Claude usage divided by its share of working-age population, so 1.0 means usage exactly proportional to population and 2.0 means twice what population alone would predict.
Singapore is the outlier worth watching. Business and Financial Operations accounts for 7.65% of its usage, the highest in this group and well above the 5.77% global figure, and data analysis and business intelligence takes 5.31% of its requests against 3.83% worldwide. Its work split (46.75% work, 42.28% personal) leans further toward professional use than the US does.
Inside the US, the money share barely moves
Financial work is not concentrated in the states you would guess. Business and Financial Operations sits between 5.9% and 7.3% of usage everywhere, with Florida and Texas above New York and California.
Business and Financial Operations share of state usage
Six large states against the US average, May 2026
Source: Anthropic Economic Index, US subregion comparison, May 2026. Bars start at zero; the visual spread is narrow because the underlying spread is narrow.
California publishes the financial advice task at 1.83%, slightly above the 1.79% national figure. Florida is the state that stands apart on posture rather than topic: 47.88% of its conversations look like work against 41.32% nationally, and it runs the highest automation share of the six (52.74%), meaning a larger slice of conversations where the person hands the task over rather than working through it.
Which financial roles' tasks show up most
Credit counseling is the standout. Its task catalog matches 1.49% of global usage, ranking 14th out of 718 published occupations, far ahead of investment analysis or accounting. Debt, budgeting and repayment questions are a much larger share of AI money use than portfolio work.
Source: Anthropic Economic Index, Business and Financial Operations category, May 2026. Rank is out of 718 occupations with published usage. Augmentation means the person stays actively involved in the task; the remainder is automation, where the person directs the model to complete it. Occupations with broad task catalogs can rank high because many different requests match their tasks.
Financial tasks skew toward augmentation. Every occupation in the table sits above the 51.38% global augmentation average, and quantitative analysis (65.9%) and credit counseling (65.3%) sit near the top. People working through money questions stay in the loop rather than handing the whole thing over. That is a conversation style, not a statement about job outcomes.
Four things this data does and does not support
It measures conversations
Every figure is a share of sampled conversations, not of people. One person asking forty investment questions and forty people asking one each look identical here.
It is a snapshot
The dataset covers April and May 2026 with no trend series. Nothing here can show a share rising or falling over time, and no claim in this report does.
It cannot speak to jobs
Task matches say nothing about employment, displacement or job security. A high usage share for an occupation's tasks is not evidence about the people in that occupation.
Blank is not zero
Unpublished entries reflect privacy suppression and publication thresholds. Missing rows cannot be treated as zeros, and remainders cannot be computed by subtraction.
Questions about AI and money in 2026
What share of AI conversations are about money?
Worldwide, the task "advise clients or respond to inquiries about financial matters" matches 1.37% of sampled Claude conversations, the seventh most common work task of any kind. In the United States it reaches 1.79% and ranks fourth. Adding the published US money topics (buying and investing at 2.57%, financial modeling at 1.10%, quant trading at 0.64%, accounting and bookkeeping at 0.59%, business operational finance at 0.46%, financial education at 0.44% and taxation at 0.14%) gives at least 5.9% of US conversations, and the true figure is higher because small topics are suppressed.
What is the Anthropic Economic Index?
The Anthropic Economic Index is a public dataset from Anthropic that matches anonymized, aggregated Claude conversation content against O*NET, the US government's catalog of occupational tasks. The May 2026 release covers 121 countries and 718 occupations with published usage, and is licensed CC BY 4.0. It reports observed usage by task; it contains no income, employment or GDP data.
Does this mean AI is replacing financial advisors?
No, and the dataset cannot answer that question in either direction. It records which occupational tasks appear in conversations, not who is in the conversation or what happened afterward. The accurate phrasing is that AI is used for tasks commonly done by financial advisors. Many of the people asking are consumers, students or small business owners rather than advisors, and there is no employment data in the release to connect usage to job outcomes.
Which countries use AI for financial questions the most?
Among countries publishing the financial advice task in May 2026, Japan leads at 1.83%, followed by the United States at 1.79%, the United Kingdom at 1.43% and India at 1.20%, against a 1.37% world average. Singapore shows the strongest professional finance tilt by a different measure: Business and Financial Operations is 7.65% of its usage, above the 5.77% global share.
Does this data cover ChatGPT, Perplexity, Claude, Gemini, Grok and Google AI Overviews?
No. The Anthropic Economic Index covers Claude usage only, across Claude chat and Cowork. It is the most detailed public task-level breakdown of AI usage available, but it is one platform's data. Behavior on ChatGPT, Perplexity, Gemini, Grok and Google AI Overviews may differ, and no public dataset of comparable granularity exists for them.
What do augmentation and automation mean here?
They describe conversation styles, not job outcomes. Augmentation means the person stays actively involved in the task; automation means the person directs the model to complete it. Globally the split is 51.38% augmentation to 48.62% automation. Financial occupations skew toward augmentation, with credit counseling at 65.3% and quantitative analysis at 65.9%.
What is the Anthropic Usage Index?
It is a geography's share of Claude usage divided by its share of working-age population. A value of 1.0 means usage exactly proportional to population, and 2.0 means twice what population alone would predict. Countries are indexed against all covered countries, and US states against the US national average, so the two scales are never compared with each other. Singapore posts 5.81, the United States 3.87 and the United Kingdom 3.35.
Methodology. All figures are drawn from the Anthropic Economic Index, data period May 2026, snapshot published 24 June 2026, licensed CC BY 4.0. The index classifies anonymized, aggregated Claude conversation content against O*NET task descriptions and request-topic hierarchies; shares are percentages of sampled conversations, not of users, and cannot be tied to individuals. Coverage includes 121 countries with a published usage index and 718 of 923 tracked occupations. Null and missing entries reflect privacy suppression or publication thresholds and are never measured zeros, so remainders are not computed by subtraction anywhere in this report. The 5.9% US money-topic figure sums seven non-overlapping published topics and is therefore a floor, not a total. The dataset carries no income, employment or GDP data and no trend series, so it cannot support claims about automation, displacement or change over time. Source and full methodology: anthropic.com/economic-index.
© 2026 StablecoinInsider.org · How People Use LLMs for Finance in 2026