All posts
May 14, 2026 · 2 min read · claude

Anthropic's product chief says proactivity is next, which rewrites what 'being cited' even means

If Claude is setting up your customer's automations before they ask, the citation moment moves from query to setup

Anthropic's product chief says proactivity is next, which rewrites what 'being cited' even means

Photo by Randy Fath on Unsplash

Cat Wu, Anthropic's head of product for Claude Code and Cowork, sat down with TechCrunch and named what's coming in the next six months: "proactivity." Her example was Claude understanding what you work on and setting up automations for you, not waiting for the prompt. That's a small line in a long interview. It's also the part marketers should reread.

Synchronous, routine, proactive

Last year was synchronous: a person types, Claude answers, citations appear in the reply.

If that ships the way she's describing, the citation surface stops being the answer panel. It becomes the setup step. When Claude proposes "hey, I noticed you handle a lot of vendor onboarding emails, want me to draft a workflow?", whatever sources it pulled to understand vendor onboarding are the ones shaping the automation. The user may never see those sources. They'll see the suggestion.

Why this is harder than it looks

The model is inferring needs from context most marketers don't have access to (the user's calendar, their open tabs, their Claude Code repo).

I'm not sure how much of this ships in six months versus eighteen, and Wu's "our hope is that it continues" laugh about the model release pace suggests even Anthropic is hedging. But the direction is clear enough to plan against.

The shift in what gets indexed

Situation recognition rewards pages that describe workflows, job functions, and the specific pain points that trigger an automation suggestion.

If your product page reads "the leading platform for revenue teams," reactive Claude might still cite you for a direct query. Proactive Claude has no query to match against. It has a user spending three hours a week on commission disputes. The source it pulls is the one that described that exact situation in language the model can pattern-match.

Something to do this week

Pick your top three customer personas and write one page per persona that describes their week, not your product.

The other thing worth tracking is Glasswing, the cybersecurity model Anthropic isn't releasing publicly because, per the article, it "could be weaponized by bad actors." If Anthropic starts gating more capable models to enterprise partners only, the gap between what the public Claude knows and what the enterprise Claude can do widens. Which changes who you're optimizing for, and how much of the audience you can reach at all.

claude anthropic product-strategy

Track your own brand in AI search.

Five minutes from sign-up to your first visibility report. Free plan, no credit card.