More than half of corporate legal and tax departments want their outside firms using AI on client work, but fewer than a third know whether their firm actually does, a 2026 Thomson Reuters survey found. **The AI trust gap moved past adoption in 2026 — now it's about proof, and vendors who can't show their work are losing deals.**
Why Do Buyers Want Proof an AI Vendor's Claims Are Real?
Nearly all business buyers now use AI themselves while shopping for a vendor, but they don't take an AI pitch at face value. Forrester's Buyers' Journey research, drawn from roughly 18,000 global business buyers, found 94% use AI during the buying process, yet they compensate for AI-generated noise by seeking validation from peers, product experts, and independent analysts before trusting a claim, and more than 60% now insist on a trial before they'll commit to a contract (Forrester, 2026). A pitch deck with the word "AI" on it stopped being enough to close a deal in 2026. Buyers want to watch the thing work first.
The stakes for overclaiming went up too. In August 2026 the FTC finalized a $930,000 settlement with Cox Media Group and two smaller marketing firms over an "Active Listening" ad-targeting product the companies pitched as built on real-time voice data captured from smart devices with consumer opt-in — neither of which, the FTC found, was actually true (FTC, 2026). It was the 14th action under the agency's Operation AI Comply initiative, and the order carries a 20-year compliance term. The message to any business selling or buying an AI-branded service is the same: a claim without proof is a liability now, not just a marketing shortcut.
What Did the 2026 Thomson Reuters Survey Find About the AI Trust Gap?
Thomson Reuters Institute surveyed more than 1,500 professionals across 27 countries for its 2026 AI in Professional Services Report and found the same proof problem sitting inside the legal and tax industry specifically. Only 15% of organizations have adopted an agentic AI tool so far, though another 53% are actively planning or considering one, and just 18% of respondents said their organization tracks AI's return on investment at all (Thomson Reuters, 2026).
The sharpest finding is the client-firm gap: more than half of corporate legal and tax departments want their outside firms using AI on client matters, but fewer than a third of those same departments know whether their firm is actually doing it (Thomson Reuters, 2026). Firms aren't getting a clear signal either — 40% of firm respondents said they've received client instructions to both use AI and not use AI on the same kind of work, sometimes from different people at the same client (Thomson Reuters, 2026). Everyone wants an answer. Almost nobody has a way to check one.
That dynamic isn't limited to Am Law 100 firms or Big Four accounting shops. A Kalispell CPA practice fielding a client's year-end question about AI-assisted bookkeeping, or a Missoula law office getting asked the same thing about a filing, is working the identical problem at a tenth the scale — no compliance department, no internal audit trail, just a vendor's word standing between the client's question and an answer. The survey's gap doesn't shrink with firm size. It just gets harder to notice, because smaller firms have fewer people asking the question out loud.
Vendor Claims vs. Verifiable Proof: What's the Difference?
| What to check | A claim | Verifiable proof |
|---|---|---|
| What it's built on | "We use AI for that" | Shows you the system running on a real example, live |
| Your data if you leave | "You're fully protected" | Data export and deletion terms written into the contract |
| Who's accountable for a mistake | "Our AI is really accurate" | A named escalation path and error-handling policy you can read |
| How it actually works | "That's proprietary" | Third-party audit rights, or a summary you can verify independently |
When Is Just Trusting a Vendor's Word Fine?
Not every AI purchase needs an audit clause. If the tool is cheap, touches no customer data, and you can walk away in a month with nothing lost — a drafting assistant, an internal note-taker, a scheduling bot you're only testing — taking the vendor's word is a reasonable, low-risk bet. The verification bar should rise with what's actually at stake: the moment a vendor is handling customer phone calls, payment data, or anything that ends up in a contract or a filing, "trust us" stops being sufficient, because you're the one accountable if the claim turns out to be wrong, not the vendor.
How Do You Actually Verify an AI Vendor's Claims?
Four questions do most of the work. Ask to see the system running live on a real example from your business, not a demo script written for a sales call. Ask exactly what happens to your data and call history if you cancel — in writing, not verbally. Ask who is accountable, by name, when the AI gets something wrong. And if a vendor calls its methodology "proprietary" the moment you ask how it actually works, treat that as a question still unanswered, not a satisfying one — a legitimate vendor can show you results without handing over their source code.