By Elena Vasquez

Personal chats make up nearly 50% of Claude conversations on weekends, compared with about 35% on weekdays, according to Anthropic’s Economic Index.

Anthropic’s Economic Index finds that Claude use follows the ordinary calendar more closely than employers’ automation plans: weekends bring household questions, 6 p.m. brings recipe requests, and tax deadlines concentrate filings.

Federal Reserve Board Governor Michael S. Barr said little evidence so far shows AI displacing workers across the economy, putting daily behavior ahead of layoff claims.

Job and household questions share one chat window, so workplace AI requires a view of the whole week. The data lean toward well-paid technical jobs held by a small share of Americans; Anthropic said pay, computing use and optimism may reflect that occupational mix as much as AI’s effects, leaving gains for lighter users unresolved.

The report’s six patterns cover weekends, daily rhythms, tax deadlines, work and personal outputs, computing by wage, and optimism among heavy users.

When does Claude use rise?

On weekends, personal matters account for nearly 50% of Claude chat and Cowork conversations, versus about 35% on weekdays. Work requests remain the larger weekday share; the weekend narrows the gap without reversing it. Job-application activity also declines on weekends.

The software stays the same between Saturday and Tuesday; the calendar changes what users ask. The weekend rise also spans Claude Code and Anthropic’s API, products built for developers and businesses, not just the consumer app.

At 6 p.m., recipe requests run at 2.3 times Claude’s average rate, the sharpest hourly surge. Business correspondence rises only slightly between 10 and 11 a.m., when offices are busiest. Household obligations, not workplace schedules, drive the largest daily change.

Night and weekend work conversations skew toward better-paid occupations, while the two lowest wage quartiles account for a smaller share. Anthropic’s researchers suggest higher-paid workers may work beyond normal hours; they cannot determine users’ occupations with certainty, leaving the explanation plausible rather than proven.

Why did tax questions surge?

Tax discussions reached eight times their usual May daily share of Claude conversations on April 14, the day before federal returns for 2025 were due April 15, 2026, according to CNBC Select. The spike points to last-minute filers, not early preparers.

Near the deadline, users treated Claude less as a financial planner than a stand-in for an accountant or knowledgeable friend. CNBC Select reported that penalties and interest begin immediately when taxpayers owing money miss April 15.

Continued nonpayment can lead to wage garnishment, a property lien or passport loss. Even one day’s delay costs money. On April 16, tax topics fell sharply below the deadline peak despite penalties starting for unpaid filers.

The pattern shows users crossing a calendar deadline, not managing what they owe over time.

What does Claude produce?

Nearly every Claude chat and Cowork conversation yields something in hand, Anthropic reports. Across all use, the most common result is an explanation, at 17%. Work conversations skew toward outputs colleagues or clients can receive: a document or report appears in 20%, while explanations account for only 9%.

That complicates the idea that AI mainly removes tasks. Personal conversations reverse the mix: explanations make up 25%, far above their work-related share. The record separates workers seeking deliverables from private users seeking answers.

Who uses the most computing power?

Workers in the top third of occupations by pay consume 2.07 times as many Claude tokens—the units of text processed—as those in the bottom third. Marketing managers, paid more per hour than editors, generate about 2.5 times as many tokens per session.

The task mix explains 44% of the association between wages and computing use. App builders consume far more tokens than people having ordinary conversations, and a 2025 study by Handa and colleagues, cited by the Federal Reserve Board, found users in computer and mathematical occupations asked essentially only about programming.

A higher wage alone would not raise usage; coding demands more computation. Whether heavy use helps or displaces workers remains unsettled. Anthropic’s researchers wrote: “If the human remains involved in the highest-value tasks, the pattern looks more labor-augmenting than labor-displacing.”

Federal Reserve economists Leland D. Crane and Paul E. Soto found annual employment growth for coders—nearly all in occupations highly exposed to AI—had been about 3 percent lower than before ChatGPT. They cautioned against treating the slowdown as a simple AI effect, leaving the central question open.

Does heavy use signal optimism?

Users who delegate the most work to Claude expect more automation, yet report the greatest hope for their pay, job security and the meaning of their work. Anthropic presents heavy use as relief rather than livelihood anxiety.

Managers make up 23% of survey respondents but only 7% of U.S. employment, giving their views more weight than their workforce share. Their control over which tasks to delegate may also make automation feel less threatening.

The survey excludes workers who lost jobs before using Claude at work, including those with the most to lose. It measures heavy users’ views of their own futures, not the economy’s distribution of gains; if AI mostly boosts existing users, lighter users and non-users may be left behind.

The evidence shows an AI tool embedded in daily life, but not proof that gains from heavy use will reach lighter users or non-users.