Show Me the Receipts: Why Every AI Answer Should Prove Where It Came From
Tracy Thayne7/14/2026·6 min read
Last month I watched a director of marketing do something that has become the quiet ritual of the AI era. Her analyst had used an AI tool to produce a competitive brief, a clean six-page summary of a rival's positioning, pricing moves, and recent wins. It looked great. Then she spent the next forty minutes doing what she called "the homework check": googling claims, opening the competitor's site, pinging a sales rep to confirm a pricing detail the brief stated as fact.
The brief took eight minutes to generate. Verifying it took five times that. And she was right to check, because one of the pricing claims was six months stale.
That ritual is now everywhere, and the numbers behind it are worse than most leaders realize. A global study of over 48,000 people by KPMG and the University of Melbourne found that 66% of people rely on AI output without evaluating its accuracy, and 56% admit to making mistakes in their work because of AI. Only 46% are willing to trust AI systems at all. And the trust problem follows the output out the door: Gartner found that 69% of B2B buyers now turn to sales reps specifically to validate AI-generated insights before they act on them.
Read those numbers together and you get the real picture of mid-2026. Half of us are trusting AI blindly and paying for it in errors. The other half are paying a verification tax that quietly eats the productivity gains we bought AI to deliver. Neither is a way to run a business.
The Verification Tax Is the Hidden Cost of Ungrounded AI
Every AI answer that arrives without evidence forces a choice: trust it or check it.
Trust it, and you join the 66% running on unevaluated output, where the 56% error rate lives. Check it, and you have converted an eight-minute generation into an hour of homework. Multiply that across every brief, every persona, every campaign summary, every account plan your team generates, and the arithmetic turns ugly. The tool saved minutes on production and charged them back, with interest, on verification.
This is the same trap I described in Why Most Companies Are Getting AI ROI Wrong: measuring the speed of the task while ignoring what happens to the work downstream. A fast answer you cannot act on without checking is not a fast answer. It is a rough draft with confidence.
Subscribe
Get notified by email when we publish a new post. No spam, unsubscribe anytime.
Interestingly, the people who use AI the most have already figured this out. TrustRadius found that the most frequent AI users are also the most rigorous checkers, with 62% of frequent users saying they always or very often fact-check AI output. Heavy users are not more trusting. They are more disciplined, because they have been burned. That discipline is rational, and it is also an indictment of the tools.
A Claim Without Provenance Is an Opinion With Good Grammar
The root problem is structural. Most AI tools generate answers the way a confident stranger does: fluently, plausibly, and with no obligation to show where any of it came from.
That is not a bug in the model. It is the natural behavior of a system with no grounding in your business. When an AI has no access to your actual customer history, your competitor intelligence, your past campaigns, and your product facts, it has nothing real to cite, so it cites nothing and lets the fluency do the persuading. I made this argument about output quality in Context Is the Whole Game, and it applies with even more force to trust. Ungrounded generation does not just produce generic work. It produces unaccountable work.
And unaccountable work cannot be delegated. This is the part that should worry every leader betting on AI leverage. You can only hand real decisions to a system whose reasoning you can audit. If the answer arrives naked, a human has to re-dress it in evidence before anyone senior will act on it. The bottleneck never went away. It just moved.
The Receipt: What Accountable AI Actually Looks Like
There is a simple fix, and it is not a better model. It is a receipt.
Every answer an AI gives you about your business should arrive with its provenance attached: what sources it drew on, how fresh each one is, how confident it is in each claim, and, just as important, what it did not know. Not footnotes buried in an export. Receipts attached to the answer itself, where each cited source is one click from inspection.
A real receipt changes the economics of trust. The forty-minute homework check collapses into a ten-second glance, because the checking has been done by the system and exposed rather than hidden. The stale pricing claim in that competitive brief would have announced itself: this fact was last updated six months ago, refresh before using. Honesty about freshness and gaps is not a weakness in an intelligence system. It is the single most trust-building thing one can do.
This is only possible when AI runs on a compiled, living body of business knowledge rather than a context window full of pasted documents, a distinction I unpacked in RAG Is Only Half the Story. A system that actually maintains your intelligence, with every fact typed, timestamped, and traceable to its source, can show its work because the work exists. A chatbot with uploads cannot, because there is nothing underneath the answer to point at. We built Expona around exactly this principle: every answer and every deliverable carries a receipt showing which personas, competitors, documents, and past campaigns it was composed from. The receipt is not a feature bolted on. It is what a grounded answer looks like by construction.
What to Demand From Every AI Tool This Quarter
If you are evaluating AI for your team, add one test to the process. Ask the tool a question about your business, then ask it three follow-ups. Where did that come from. How current is it. What are you not sure about.
A tool that can answer all three is operating on real context and can be trusted with real work. A tool that answers none of them is a fluent stranger, and everything it produces will carry the verification tax forever. Price that tax into the evaluation, because your team will pay it every single day.
The Takeaway
The AI trust crisis is not a perception problem to be managed with change-management decks. It is a product deficiency. People distrust AI answers because most AI answers give them nothing to trust: no sources, no freshness, no confidence, no admission of gaps.
The fix is provenance. An answer with a receipt can be checked in seconds, delegated with confidence, and defended in front of a CFO. An answer without one is an opinion with good grammar, and your team will keep spending their afternoons doing homework checks on it.
Do not ask your team to trust AI more. Ask your AI to earn it. Show me the receipts.
Tracy Thayne* is the founder of Expona, an AI-powered operational intelligence platform for B2B marketing. Read the Expona founder story or subscribe to the blog (below) for weekly insights on context, AI, and the operating model of the next decade.*
