Ask Claude How Much AI Is Actually Used in Your Industry

Most of what small business owners hear about AI adoption is a guess dressed up as a trend piece. "Everyone is using AI now" is a vibe, not a number. On July 22, 2026, Anthropic gave that question an actual answer: a new connector that lets you ask Claude directly about the Anthropic Economic Index, the company's own research on how AI gets used across occupations, tasks, and industries.

To be direct about what this is and isn't: it's a data tool, not a productivity feature. It won't write your invoices or answer your phones. It lets you replace a guess with a real number in about the time it takes to type a question.

What is the Economic Index connector, exactly?

It's a toggle inside Claude that connects any conversation to Anthropic's Economic Index research. Once it's on, you can ask which occupations use AI the most, what people in your state or line of work tend to ask Claude for, or how usage in your field has shifted over the past year, and Claude answers using the underlying dataset instead of its general training.

Setup is close to nothing: in Claude.ai, open the connectors menu, find "Anthropic Economic Index" in the directory, and enable it. Anthropic says it works with any Claude model, takes about a minute, and there's no separate app or account to create (Anthropic). The data was already public as downloadable reports; the connector's real contribution is making it conversational instead of a document you have to dig through.

What does the index actually track?

The Index classifies real Claude conversations against roughly 20,000 specific work tasks pulled from O*NET, the U.S. Department of Labor's occupational database, then rolls those tasks up into occupations and broader categories like "business and financial" or "education and library" (Anthropic). A privacy-preserving classification system called Clio does the matching, so researchers see aggregate patterns, never individual conversations.

That matters for how much you should trust the numbers. This isn't a survey of how people feel about AI. It's a record of what tasks people actually brought to Claude, and how often, which is sturdier than most "AI adoption" headlines you'll see this year.

Which occupations show the most AI usage right now?

Overwhelmingly, computer and technical work. Anthropic's most recent usage report found computer and mathematical occupations make up roughly 30% of surveyed Claude users, against about 4% of U.S. employment overall, a 7.5x overrepresentation. Management roles are next, at 23% of surveyed users versus 7% of employment (Anthropic).

Task-level data tells a similar story. Computer programmers show roughly 75% task coverage by Claude, and data entry work sits around 67% (Anthropic). Pooled across multiple Economic Index reports, about 49% of jobs in Anthropic's sample, roughly half, had at least a quarter of their tasks touched by Claude at some point (Anthropic).

What about the industries Arios actually works in?

This is where the honesty matters more than the excitement. The same labor-market research found about 30% of occupations in the U.S. economy currently show close to zero measured Claude usage, including cooking, motorcycle repair, lifeguarding, and bartending (Anthropic). Farming fares no better in Anthropic's own category breakdown: "farming, fishing, and forestry" is the single lowest-usage occupational group the Economic Index tracks, at just 0.1% of queries (Anthropic).

That overlaps with real chunks of Arios's own client base: agribusiness, logistics, manufacturing, oil and gas field operations. If your business runs on physical, hands-on work, the honest answer right now is that direct AI usage inside the tasks themselves is still thin. What's usually not thin is the office layer sitting on top of that work (scheduling, dispatch, paperwork, customer intake), which shows real, growing usage in the Index even when the trade itself doesn't.

Does high usage mean the tool is actually working well?

Not automatically, and Anthropic's own data says so. The labor-market report distinguishes raw task coverage from an effectiveness-adjusted measure that weights coverage by how often Claude's output actually gets used successfully. Under that adjustment, some high-coverage categories like data entry look different once quality and reliability are factored in, not just frequency (Anthropic).

I'd treat this as the single most useful caution in the dataset. A task showing up a lot in Claude conversations means people are trying it, not that it's reliable enough to hand over unsupervised. Pilot first and check the output yourself: the data tells you where to look, not whether the result is right.

Does AI usually do the whole task, or just help with it?

Mostly the second one, for now. Anthropic separates conversations into "automation," where Claude does most of a task with minimal back-and-forth, and "augmentation," an iterative back-and-forth toward a finished output.

Augmentation still leads at 52% of conversations against 45% for automation, though automation's share has been climbing (Anthropic). If you're hoping to fully hand off a task like drafting client emails, you're closer to today's minority use case than the majority. Staying in the loop is still the norm, and a reasonable way to start either way.

So what does this mean for your business?

It means you now have a free way to sanity-check an AI vendor's pitch before you spend money on it. If a vendor claims "everyone in your industry is automating this," open Claude and ask whether that's actually true for occupations like yours. That's real money protected: a failed AI pilot costs more than the software, it costs staff hours and trust.

It also points toward where the fast wins actually are: office and administrative work (scheduling, intake, reporting, drafting) shows consistently higher measured usage than physical trade work. That lines up with what I see with Arios clients directly, the quickest, most defensible first AI project is almost always in the paperwork layer, not the floor operations. Time saved there is real time, even at a business whose core work is still mostly hands-on.

How do you try this yourself?

Open Claude.ai, go to the connectors menu, find "Anthropic Economic Index" in the directory, and enable it. Ask something specific: your occupation, a task you're weighing whether to automate, and ask for source numbers, not just a summary. "How much is Claude used for dispatch scheduling in logistics" gets a usable answer; "Is AI good now" doesn't.

For the pricing side of this story rather than the adoption side, see what Sonnet 5 becoming Claude's free default model means for cost. Once you know where usage is already proven, the natural next steps are sizing the investment with our piece on cost per successful automated task, and our breakdown of where AI delivers the fastest wins in operations. If the honest answer for your field is "adoption looks thin," our piece on the SMB AI adoption skills gap covers why, and what closes it faster than waiting.

The data won't make the decision for you. It just means the decision doesn't have to start from a guess.