A new independent benchmark from Send puts ZoomInfo's GTM.AI connector firmly atop Anthropic's connector directory for nine consecutive weeks between July and September 2025, tracking over 100,000 snapshots. But the ranking measures a public popularity signal—not data accuracy or real-world outcomes—leaving IT leaders and revenue teams with a tougher question: does a directory leaderboard mean their AI assistant is getting better answers?

Send's popularity score is a discovery metric, not a verdict on data quality

The study, titled "The State of GTM Connectors," polled Anthropic's connector directory every minute for nine weeks, logging more than 100,000 total snapshots and 3,610 for the 11 go-to-market providers included. ZoomInfo finished with a normalized index of 100.0, while the second-place provider trailed by roughly two points—a tight race that suggests the top slots were regularly swapping positions beneath the surface. Send itself calls the directory's built-in popularity score "the closest public signal to how widely each connector is actually used," but that's far from a usage audit. The score reflects installation interest and discovery, not active sessions, query volume, or the accuracy of data returned.

This distinction matters because Anthropic's connector framework lets Claude access connected applications under a user's existing permissions. According to Anthropic's own documentation, enabling a connector can allow Claude to access—or potentially modify—data within those permissions. The metric that Send tracked is, at best, a rough proxy for mindshare among Claude users who've poked around the directory. It says nothing about whether a sales rep's buying committee map is correct, whether a technographic profile is up-to-date, or how many times a connector was actually queried in a production workflow.

Why a Windows-centric enterprise should look past the leaderboard

ZoomInfo's connector runs on GTM.AI, a headless context layer that the company positions as a single governed source for sales intelligence across AI assistants. It supports not just Claude but also Microsoft Copilot, ChatGPT, Salesforce Agentforce, and HubSpot Breeze. For an organization running Microsoft 365 where sales teams use Copilot in Word or Excel alongside a Claude-powered research workflow, that multi-assistant coverage can reduce data silos—but only if governance holds.

When a user asks Copilot to research a target account or Claude to map a buying committee, GTM.AI returns verified company, contact, technographic, and buying-signal data without requiring a static spreadsheet export. ZoomInfo claims its data on over 100 million companies and 500 million contacts is refreshed continuously rather than exported and left to decay. That's a genuine advantage for AI agents that need fresh signals, given that roughly 70% of B2B contact data decays each year. But the convenience also expands the surface area for misconfiguration. Administrators must verify that connector authorization maps cleanly to least-privilege roles, that CRM permissions are consistently enforced, and that a broad research request doesn't turn into uncontrolled list building or data extraction.

For Windows shops already deep into Microsoft's Purview and identity tools, the question isn't whether ZoomInfo is popular among Anthropic users; it's whether the connector's scope, audit logs, and data handling align with existing compliance policies. If a single MCP-based layer can truly deliver the same permissioning and lineage across Copilot, Claude, and ChatGPT, then a directory ranking becomes a footnote. If not, then the risk of data sprawl across AI surfaces outweighs any popularity metric.

What the study didn't measure—and why it matters

Send's methodology intentionally avoided weighting for revenue, data accuracy, or user satisfaction. It's a pure popularity index drawn from a public-facing score that Anthropic says is "based on recent engagement." That means a connector could rank first because of a well-timed marketing push, a prominent listing position, or a single influential influencer's recommendation—not because it returned better answers than its rivals.

Moreover, the study's tight grouping—top providers within two index points—makes the "No.1" designation a matter of statistical noise rather than a durable lead. A 2026 benchmark that lasts only nine weeks can't capture seasonal shifts, enterprise rollouts that take months to surface in directory installs, or the long tail of user testing that precedes a production commitment. For IT buyers, the real due-diligence checklist remains unchanged: run your own accuracy benchmarks, test the connector's worst-case scenarios (aged-out contacts, ambiguous company names, foreign-language signals), and trace how the tool handles your most sensitive account data before trusting an agent with it.

Practical steps for IT teams and sales ops

If your organization is evaluating GTM connectors for Claude, Copilot, or both, a directory ranking should be one of the last things you look at. Start with a concrete plan:

  • Audit the connector's permission model. Ask the provider for a full list of OAuth scopes or MCP permissions. Map them against your identity provider's least-privilege principles and see if the tool can be restricted to read-only operations for non-admin personas.
  • Demand data freshness guarantees. Stale data is an even bigger liability in an AI agent context because the answer sounds confident even when it's wrong. Press vendors to provide service-level objectives for refresh cycles, especially for high-velocity fields like job changes or funding events.
  • Test with your own accounts. Don't rely on demos. Run a dozen of your own accounts through the connector—both well-known enterprises and tricky mid-market firms—and verify against your CRM's record of truth. Look for phantom contacts, outdated technographic flags, and missing buying signals.
  • Check the audit trail. When an agent pulls data, what gets logged? Can you trace a Copilot recommendation back to a specific ZoomInfo record? If compliance requires showing an auditor that a sales pitch didn't use prohibited datasets, you need that traceability.
  • Align with existing Copilot governance. For Windows-focused organizations, Microsoft's Copilot Studio and Purview already give you hooks for data loss prevention and sensitivity labeling. Confirm whether the GTM connector respects those controls or operates in a separate lane. A governed AI stack demands that all connectors speak the same compliance language.
  • Set usage guardrails. Decide whether broad research queries should be allowed or whether connectors should only surface data after a specific intent trigger. Unrestricted access can lead to data scraping or accidental exposure of sensitive buying signals.

The ranking from Send is a useful signal of market momentum, but it's the starting line, not the finish. ZoomInfo's first-place finish across nine weeks suggests that many users are kicking the tires—but until those tires prove themselves on your own data, the leaderboard is just a scoreboard.

How GTM AI connectors became an enterprise chessboard

The rise of agentic go-to-market tools has put a premium on real-time, permissioned data. Two years ago, most AI assistants relied on manually exported CSV files uploaded into prompts. That model broke down as soon as a sales rep forgot to refresh the file, missed a buying committee change, or breached a data-use agreement by sharing an ungoverned export. Connectors emerged as a cleaner alternative, but they also introduced a new dependency: the quality of the underlying data layer.

ZoomInfo's bet with GTM.AI is that one context layer serving multiple AI surfaces will become the default architecture for revenue teams. The company has been rapidly building integrations—Claude, ChatGPT, Copilot, Agentforce, Breeze—in an effort to make its verified data the plumbing that agents tap into, regardless of which assistant sits on top. This week's directory ranking is a marketing win that validates, at least in public perception, that the bet is resonating with early adopters.

Yet the landscape is far from settled. Other large data providers are lining up their own connectors, and major platforms like Microsoft are deepening native data connections within Copilot. The real fight isn't over Anthropic's directory; it's over who becomes the primary source of truth for AI-powered revenue motions. For Windows-centric enterprises, that fight will be decided not by popularity scores but by which provider can demonstrate airtight security, cross-platform permissioning, and measurable improvement in sales outcomes.

What to watch next

Expect connector directories from Anthropic, Microsoft, OpenAI, and others to become a new distribution channel for enterprise data services. That will accelerate investments in MCP-based context layers, but it will also increase pressure for transparency. Industry groups are likely to push for standardized connector audits that measure more than installation popularity—covering accuracy, freshness, security scopes, and real-world usage patterns. For now, treat any directory ranking as a glance at foot traffic, not a guarantee of quality. Your own evaluation will remain the only one that counts for your business.