ImplementationAdoption storyFor: Legal and compliance · Marketing

Anthropic legal: marketing review went from days to 30 minutes an asset

Anthropic says Claude checks marketing copy before its lawyers do, taking review from 2-3 days to 30 minutes an asset. It gives no volume, cost or error rate.

At a glance

Company
Anthropic · Technology and software · Size not stated
Team and use case
Legal and compliance · Legal review of marketing content
Tools
Claude Enterprise, Claude Tag, Claude skills, Slack
Stage
In production, one team, since 2025 (self-review tool); Claude Tag channel by August 2026
Results
  • Marketing review turnaround, self-review tool (December 2025 post): Two to three days → 24 hours
  • Marketing review turnaround, Claude Tag channel (August 2026 article): "A day (or longer)" → 30 minutes per asset
  • Time to draft a privacy impact assessment: Roughly two hours → Thirty minutes
  • Contract redlining per agreement: "Hours" → "Minutes"
Cost
Not disclosed: no licence cost (Anthropic makes the product), no price for Claude Tag, no hours for building or maintaining the instructions.
People
Lawyers without engineering backgrounds built the workflows. No headcount change is mentioned. In December 2025 a lawyer still read every blog post; the August 2026 article says counsel review the statements Claude flags and does not say whether each asset is still read in full.
Evidence
Vendor-reported

Published by the company selling the product, usually with the customer. Wins are selected; cost is usually missing.

What this leaves out: Anthropic sells Claude and is describing its own legal team. The figures are elapsed turnaround, not lawyer hours, and Anthropic says they reflect individual employees' experiences. No volume, period, error rate or cost is given.

Anthropic's legal team says a Claude tool that lets marketers check their own copy before filing a legal ticket cut marketing review turnaround "from two to three days down to 24 hours", in a [company blog post](https://claude.com/blog/how-anthropic-uses-claude-legal) dated 8 December 2025. A [second Anthropic article](https://claude.com/resources/articles/how-anthropic-employees-use-claude-tag), dated 28 August 2026, says a later set-up in Slack has compressed turnaround "from a day (or longer) to 30 minutes per asset". Anthropic sells Claude; both accounts are its own.

How the review works

In the first version, described by associate general counsel Mark Pike, a marketer pastes a draft into a tool pinned in Slack. Claude reads it against a "skill", a file holding the legal team's past guidance and review framework, and flags problems such as logo and publicity rights, overstated claims and unverified statistics as low, medium or high risk, with suggested fixes. Only then does the marketer file a ticket, which goes to the right lawyer with the flags attached.

In the second, product counsel Molly Villagra set up a Slack channel where marketers post a document link and Claude Tag, Anthropic's product that puts Claude in Slack threads, examines every asset first. Claude works through its flags with the requester, checks factual statements against company Slack, an internal knowledge index and the public web, and tags a product counsel for what still needs sign-off. Each Friday it reviews the week's counsel feedback and proposes changes to its instructions, which Villagra approves.

Anthropic's [legal industry guide](https://claude.com/resources/guides/claude-for-the-legal-industry/how-anthropics-legal-team-uses-claude) gives two more results from the same team: drafting a privacy impact assessment, the document that records how a project uses personal data, fell "from roughly two hours to thirty minutes", and comparing contract versions against the commercial playbook went "from hours to minutes per agreement".

The catch: elapsed time, one team, no cost

  • Turnaround is not lawyer time. Two to three days is how long a request waited and bounced back and forth. Neither post says how many minutes a lawyer spent on each asset before or after, so the saving in lawyer hours cannot be worked out.
  • No method. The August article ends with this note: "Turnaround times in this post reflect individual employees' experiences with specific tasks; results vary with the task, the tools connected, and how Claude Tag is set up." No sample size or period is given, and no volume beyond "dozens" of assets before a launch.
  • No error rate. Neither post says how often Claude missed an issue a lawyer later caught, or flagged one that was not there.
  • No cost. The August article says Claude Tag is in public beta on Team and Enterprise plans and gives no price; neither post gives the hours spent writing and maintaining the instructions.
  • The seller is its own customer. What the same set-up costs a buyer is not shown.

The people side

Lawyers built both versions; Pike says he does not code and the article says Villagra has no engineering background. Neither post mentions a change in legal headcount. On who checks the output, the December post is plain: "I still read the blog post. I'm still reviewing the work," Pike says, and "AI systems can still hallucinate". The August article says counsel "can quickly review the flagged statements" and does not say whether a lawyer still reads each asset in full. Wren's read: that is the control to ask about, because a reviewer who reads only what the tool flagged will not catch what it missed.

What to do if you run legal operations

  • This week, pull your last 20 marketing review requests and record two numbers for each: days from request to approval, and minutes a lawyer worked on it. If the gap is large, the delay is queueing and rework, which a pre-check can remove.
  • Have one lawyer write down the ten issues the team flags most in marketing copy, and give it to marketing as a checklist to run, with or without an AI tool, before filing a ticket.
  • If you trial an AI pre-check, keep a lawyer reading every asset in full for the first month and log what the tool missed. Decide what lawyers may skip from that log.
Was this useful for your work?

Take this to work

Forward this

Anthropic says its legal team cut marketing review turnaround from two to three days to 24 hours with a Claude check marketers run before filing a ticket, then to 30 minutes an asset (Anthropic posts, 8 December 2025 and 28 August 2026); it gives no volume, cost or error rate. The part we could copy without buying anything is the pre-check: a written list of what legal flags most, run by the requester.

Questions to ask

  1. Of our own review turnaround, how much is waiting and rework and how much is a lawyer reading?
  2. How often does the pre-check miss an issue a lawyer later catches, and who is logging that?
  3. Does a lawyer still read every asset in full, or only what the tool flagged, and who decided?
  4. What does it cost per seat and per month at our volume, and who maintains the instructions when our guidance changes?
  5. Which drafts and internal systems does the tool read, and are unreleased product details in them?

The arithmetic

Lawyer hours saved a month = assets reviewed a month × (lawyer minutes per asset before − lawyer minutes per asset after) ÷ 60. Queue time saved is a separate sum: assets × (days waiting before − days waiting after).

Anthropic gives only the second sum, and only per asset: two to three days down to 24 hours, then to 30 minutes. It gives no assets a month and no lawyer minutes per asset, so the first sum cannot be done from its figures. For privacy assessments it does give work time: roughly 2 hours − 30 minutes = 1.5 hours saved each, with no count of assessments a year.

Questions this article answers

How does Anthropic's legal team use Claude to review marketing content?

Marketers run their draft past Claude before legal sees it; Claude checks it against the legal team's written guidance, flags issues by risk level and suggests fixes. What still needs sign-off goes to a lawyer with the flags attached, according to Anthropic posts of 8 December 2025 and 28 August 2026.

How much faster is legal review of marketing copy at Anthropic?

Anthropic says turnaround fell from two to three days to 24 hours with its first tool, and from a day or longer to 30 minutes an asset with a later Slack set-up. These are elapsed times from Anthropic's own account, with no volume or method given.

Do lawyers still review the content after the AI check?

In the December 2025 post, associate general counsel Mark Pike says he still reads every blog post. The August 2026 article says counsel review the statements Claude flags and does not say whether each asset is still read in full.

What did it cost Anthropic to automate legal review?

Not disclosed. Anthropic makes Claude, gives no price for Claude Tag, which is in public beta on Team and Enterprise plans, and does not say how many hours its lawyers spent building and maintaining the workflows.

Send this to someone who runs a teamLinkedInXBlueskyHNEmail