The deals you never had

There’s a category of lost revenue that doesn’t appear in any dashboard. No “closed-lost” in the CRM. No bounce in the analytics. No abandoned form fill. The buyer asked an AI about your category, the AI got something wrong about you, and the buyer moved on. You never knew they existed.

We started thinking about this after a pattern emerged in our benchmarks. Companies would see their scores, fix the errors, and then notice a change in inbound quality they couldn’t attribute to anything else. More buyers showing up with accurate expectations. Fewer first calls starting with “I heard you don’t support X” when X has been a core feature for a year.

The question was whether we could put a rough number on what the misinformation was costing before they fixed it.

How much does AI misinformation cost? The back-of-envelope math

For a typical mid-market B2B company, our framework lands around $150,000 per month in structurally inaccessible pipeline. The calculation is crude but directional.

Start with the evaluation volume. 51% of B2B software buyers start research with an AI chatbot. For a company with 1,000 monthly category-relevant searches, that’s roughly 500 AI-mediated evaluations per month the company has zero visibility into.

Apply the error rate. In our study of 33,000 AI evaluations across 47 B2B companies, 72% had at least one verifiably wrong factual claim about themselves stated with confidence. Working conservatively at the individual-evaluation grain — assuming roughly one in four evaluations carries a verifiable error — that’s about 125 of those 500 monthly evaluations where the AI is telling buyers something wrong.

Estimate the decision impact. Not every error changes the outcome. A wrong founding date rarely loses a deal. Wrong pricing, a missing critical feature, or a competitor recommendation built from the competitor’s content almost always does. In that same study, pricing errors alone accounted for 30% of every factual error — the single largest category — and once you add feature misattribution and competitor-framed comparisons, roughly half of all errors fall in categories that directly move a purchase decision.

That leaves approximately 60 evaluations per month where a buyer received incorrect information that could have changed their decision. For a B2B company with an average deal value of $25,000, even a 10% conversion rate on those evaluations represents $150,000 in monthly pipeline that was structurally inaccessible.

That 10% assumption is deliberately conservative. AI-referred traffic converts at 4.4x the rate of other channels — the buyers who start an evaluation inside ChatGPT or Claude are high-intent and largely pre-qualified — so the deals lost to a wrong fact skew toward the ones that were closest to closing, not the tire-kickers.

The number grows quickly for companies with higher deal values, more competitors, or categories where AI evaluations are common.

Why doesn’t this show up in attribution?

Because attribution systems track things that happen, and AI misinformation causes things to not happen. Someone visits your site, fills out a form, attends a demo: the system records each event and credits a source. With AI misinformation, the buyer never visits, the form never gets filled, the demo never gets scheduled. The absence of an event is invisible to every attribution system that exists.

This is what makes it different from any other marketing problem. A bad ad produces measurably bad results. Bad SEO produces measurably less traffic. AI misinformation produces nothing, literally nothing, and nothing looks exactly like “the market is slow” or “inbound is down this quarter for reasons we can’t explain.”

The signal is in the delta. Companies that benchmark their AI presence and fix the errors consistently report changes in inbound quality they can’t attribute to any other channel. Not more leads. Better leads. Buyers who show up correctly positioned by the AI instead of incorrectly positioned by it.

Why does the problem compound?

Because AI misinformation feeds itself. When a buyer asks AI about your category and the AI doesn’t recommend you, that buyer may write a review, post in a forum, or share their experience without ever having evaluated your product. “I looked at the category and went with [competitor].” That post becomes a source AI models cite in future responses.

69% of B2B buyers changed which vendor they chose based on AI guidance. Those buyers become your competitor’s customers. Those customers write case studies, leave reviews, and generate content that further strengthens the competitor’s AI presence. Every buyer you lose to misinformation makes the misinformation harder to overcome.

This is why timing matters. The winner-take-all dynamics of AI recommendations mean early accuracy advantages compound. Fixing a factual error today doesn’t just recover the deals you’re losing now. It prevents the compounding that makes future recovery harder.

What changed our thinking

We used to treat AI misinformation as a content problem. Fix the content, fix the representation, move on. The data showed us it’s a pipeline problem first and a content problem second.

The content fix is the mechanism. But the thing you’re actually fixing is a leak in your pipeline that’s invisible to every other measurement tool. When a company corrects a stale pricing claim across all four AI models, they’re not optimizing their content. They’re reopening a pipeline channel that was structurally closed.

43% of consumers report making purchasing decisions based on false AI-generated information. They didn’t know it was false. The AI didn’t either. The cost is the gap between the pipeline you have and the pipeline you should have, and it scales with how many buyers in your category start with AI. That number is 51% today. It was under 20% a year ago, and AI search traffic itself grew 527% year over year. The channel isn’t stabilizing — it’s still steepening.

What should you do about it?

Benchmark, prioritize, fix, re-measure. The sequence matters.

First, benchmark. Find out what AI actually says about you across all four models. You can’t quantify the cost until you know the errors. The benchmark should hand you misrepresentations with receipts: each wrong claim shown next to the source that contradicts it, so there’s no debate about what’s wrong or what evidence the fix needs to outrank.

Second, prioritize by revenue impact. Not all errors cost equally. Wrong pricing on a model 300 million people use costs more than a wrong founding date on a model 50 million people use. The scoring decomposition tells you which errors drive the most recommendation losses.

Third, fix and re-benchmark. Publish corrections, wait for models to update, then run the same prompts again. The run-over-run deltas are your ROI proof: which mentions shifted, which citations changed, whether the recommendation flipped. If it did, the cost of the misinformation was whatever you were leaving on the table before the fix.

The companies that do this systematically treat AI Presence as a pipeline metric, not a marketing project. The ones that don’t are leaking deals into a channel they can’t see, at a rate that accelerates every quarter.