LO
Worker adoptionCognitive automation2026-06-26
Anthropic's usage telemetry: the median Claude Code session that produces a blog post contains one human prompt, against 13 rounds of back-and-forth in chat and Cowork
Copywriteroccupation page →
Event date / reported
2026-06-26
Evidence stage
Worker adoptionMeasured, large-scale use of a tool for real work, where the decision to use it was the worker's rather than an employer's. It is more than a capability record — the work is real, not a demo — and less than a deployment record, because no employer put it into production, required it, or built a process around it. Weighted `cautious`: `automating` means the machine can do the task AND there are adoption signs, and this is an adoption sign — but usage can be experimental, and much of the measurement comes from a party with a stake, so one record is never enough and two independent ones are. Note who is counting. Vendor telemetry sees this directly and sells the tool, so such a record names that stake in its scope; a statistics agency asking firms whether their workers use AI in tasks sees the same channel with no stake at all, and that is the better source where it exists.
Tasks this bears on
Producing volume copy
Product descriptions, ad variants, SEO articles, social captions, email sequences — many pieces, small stakes each.
Automating✓ Evidence-backed
Where this applies
Measured across Claude chat, Cowork, Claude Code and Anthropic's first-party API during the report's sample period (spring 2026, covering the 15 April US tax deadline). It counts conversations, not people and not employers: 81% of conversations producing a blog or article and 80% of those producing marketing content were classified work-related, but nobody asked whether an employer commissioned, paid for or accepted that output. Higher delegated autonomy on the agentic surface appears across 26 of 31 output types — 0.37 points on a 1-5 scale across all conversations — and it survives holding the model constant: among Sonnet-served conversations the gap is still 0.26 points, which is why the authors read it as a property of the product rather than the model. Anthropic operates the tool being measured and sells it, and it is the only party that can see this data at all - both facts belong beside the numbers. Nothing here is occupation-specific: the mapping from conversation to occupation is a classifier's, not an employer's.
What this means
The thing that changed for volume copy is not that a machine can write a passable article — that was true in 2023 — but how many times a person now touches one. In chat and Cowork the median blog post takes thirteen exchanges: a brief, a draft, corrections, a restructure. In Claude Code the median is one prompt and no further human turn. Thirteen touches is a job with a person inside it; one is a request. Higher delegated autonomy on the agentic surface shows up in 26 of 31 output types, so it is not a quirk of writing.
What it does not yet show
It does not establish that anyone stopped being paid to write. These are conversations, not commissions: nobody asked whether the output was published, accepted, invoiced, or rewritten afterwards by a person whose turn the telemetry cannot see. A single-prompt session is also compatible with an experienced writer who knew exactly what to ask for and edited the result in their own document. And the count comes from the company selling the tool, which can see this and no competitor's equivalent.
What you can check
Count the exchanges on your own last three pieces — brief, draft, revisions, sign-off. If the number is close to thirteen, the part of your work that is being paid for is the exchanges, and it is worth knowing which of them the person commissioning you would notice missing.
Does it change the assessment?
No. The impact index is never moved by a single event. Nor did this record change a layer: all 1 linked judgement above already rested on earlier evidence. This one adds to them.
Source
Anthropic Economic Index report: Cadences · verified 2026-09-12 · Claude (VOLO agent) — page fetched over HTTPS and read in full as text; the one-prompt/13-turn, 26-of-31, 0.37 and 0.26 figures each located in the source text before being quoted · interpreted 2026-10-16 · Claude (COO), source re-verification 2026-10-16
Primary source — published by the party that did this, or the authority of record. No co-signature needed.
This record is cited in