anVendor vs AI: can an AI agent find who uses your competitor?

AI research agents are good at reading the public web. That makes them a natural first try for a common sales question: does this company use my competitor, how widely, and what is it worth?

We tested that question directly. We took some of the largest findings in anVendor's vendor directory: big companies using popular SaaS such as Atlassian, SurveyMonkey, Canva, Zoom and Google Workspace. We then asked an AI agent with web search to research the same company-service pairs. The agent was not told what anVendor had found.

How we ran the test

• The AI agent: Claude Opus 5.5 on Max, used as a general-purpose research agent with web search and page fetching, run on September 29, 2026. It had the same brief a salesperson would give it: for each pair, find evidence of current use, the share of employees using the tool, and a defensible annual spend estimate. Weak evidence was not to be treated as confirmation.
• anVendor: the saved findings behind each service's public customer page, with the check date, estimated adoption and estimated annual spend range shown in the app.
• Scope: 11 pairs, chosen from the biggest companies and spend ranges on the directory's most popular service pages. The nine below appear on public service pages, and each links to its page.

This is a small, qualitative test, not an accuracy benchmark. The companies were chosen from anVendor's own findings, so it measures whether AI can recover what anVendor finds. It does not measure what anVendor misses.

The biggest findings, side by side

Second test: “list the largest companies using this vendor”

We also asked a separate AI agent to build the list itself. For Atlassian, SurveyMonkey, Canva and 1Password, it had to name the five largest companies using each, with evidence, adoption and spend. The agent was not allowed to copy third-party technographics lists.

It took 21 searches and 3 min 37 s. Naming well-known logos was easy, because every vendor publishes customer stories. Ranking them by size was not possible. No source stated how many employees used a tool, and most evidence covered one team, such as a single business unit, a market research team or “1,400 teams”. It did not show company-wide use.

Third test: AI finds the names, anVendor checks them

We then put the AI agent's 20 companies into anVendor's Search leads, one vendor at a time, as a company list, and revealed each one.

anVendor detected 15 of the 20 companies the AI named. It added an adoption estimate for 14 and a spend range for 13, where the AI agent had none. Some AI claims came from a 2021 press release or a single team. When anVendor finds no evidence, add the company to your list only after further checks. This does not prove non-use: the company may use the vendor somewhere anVendor cannot see.

Why the AI agent struggled

The agent was careful. It refused to upgrade weak evidence, which is exactly what you want. But being careful exposed the limits of public-web research:

1. Vendors appear as suppliers, not buyers. Searching “Amazon Atlassian” returns AWS hosting Atlassian. “OpenAI 1Password” returns a Codex integration. “Meta SurveyMonkey” returns a Facebook integration. A less careful agent would count all three as customers.
2. Large companies rarely announce internal tools. Most SaaS use never appears in a press release, case study or job ad. For many of the biggest accounts, no public evidence is the normal result.
3. Evidence is old, undated or team-level. A 2014 study, a 2013 blog claim or one team's post says little about company-wide use today.
4. Adoption and spend are almost never public. Even the one confirmed deal, Keller Williams and Canva, disclosed no user count or contract value.
5. Third-party lists cannot be checked. Some sites list big companies as customers with no date, method or scope.

What anVendor does better

Company-level findings where public sources say nothing. anVendor identifies SaaS use behind a login, not only what a company publishes or its website loads. That covers pairs like Meta and SurveyMonkey, where the agent found only integration pages.

Fresh, current evidence on demand. Every finding shows when it was last checked. When you need a fresh answer, anVendor runs a live check: usually under a minute for one company and one service, and five to ten minutes for all of a company's subscriptions, longer for a large company. An AI agent can only repeat what was published, often years ago.

Adoption and spend in the same place as the finding. Estimated adoption helps separate a company-wide rollout (IBM and Atlassian, 76%) from a small team (Amazon and Asana, 2%). An estimated annual spend range, based on the vendor's published plans, helps rank accounts by opportunity size. AI would have to invent both.

It starts from the competitor, not the company. In Search leads, enter a competitor and filter by location, industry and size. Discovering companies is free; you pay only for a confirmed company-service result. An AI agent has to guess which companies to research first.

anVendor's unique strength: ask about any company, for any service

AI can only report what someone has published, and publishing favors big brands and big deals. anVendor turns the question around: name any company and the service you care about, and anVendor checks it.

• Any company, any size. A 30-person agency, a mid-sized manufacturer, a university or a global enterprise can be checked the same way. The company doesn't need a case study, a press release or a job ad.
• A clear answer to your question. You get a detected result with estimated adoption and annual spend, or a dated check that found no evidence. That tells you which accounts to prioritize and which to verify before outreach.
• A large catalog of popular SaaS and business services. Collaboration, design, CRM, HR, security, e-learning, developer tools, AI and more. Browse the vendor directory to see which services are covered.
• Both directions. Start with a competitor to find its customers, or start with a company to see the services it uses.

Where AI is still the better tool

• Published customer stories. The list test surfaced named customers from vendor case studies, such as Mercedes-Benz and Cisco for Atlassian, and Ryanair for SurveyMonkey. These are useful references even without adoption or spend figures.
• Context and narrative. The agent found the Keller Williams press release, the rollout timing and the fact that agents are independent contractors. That is useful for a first sales conversation.
• Signals anVendor does not report. PepsiCo's 2026 announcement naming a competing AI platform is a buying signal. anVendor does not track press releases.
• Anything outside anVendor's coverage. For a service anVendor does not track, web research is still the place to start.

The better way: use anVendor inside your favorite AI

You don't have to choose. Connect the anVendor MCP server to the assistant you already use, such as Claude, ChatGPT or any MCP client. Sign in and approve the connection; there's no key to copy. Then ask in plain language: “find companies using SurveyMonkey in Germany with 1,000+ employees”, or “what SaaS does acme.com use, and what is it worth?”

The assistant gets anVendor's dated findings, estimated adoption and spend ranges, and adds what AI is good at: context, news and a draft for the first call. Discovering companies is free. Confirmed results use the same credits as the app, within the budget you set for the connection.

What these results do and do not mean

A finding shows that a company uses a service. It does not show payment terms, satisfaction or an intent to switch. Adoption and spend are estimates, and coverage varies by company and service. When anVendor finds nothing, that means no evidence was found, not proof of non-use. The same caution applies to the AI results above.

Create a free anVendor account and connect it to your AI assistant. You get 20 free credits every 30 days, and no credit card is required.