A customer opens ChatGPT and asks how to move up from their current plan. A prospect uses Gemini to compare your pricing and editions with other vendors.
The answers sound confident and specific: the tier structure, what’s included, roughly what it costs.
They’re also wrong.
Pricing changed a quarter ago, and some of the sources feeding those answers were never updated to reflect it.
The dark funnel doesn’t stop at the sale, as we’ve written before. This is what one of those failures actually looks like, start to finish.
The customer doesn’t know that. They take the number into their renewal call, and now the conversation starts with a correction. The account team spends the first ten minutes explaining why the price the customer found doesn’t exist anymore, before they’ve said a word about value or renewal terms.
The customer walks away wondering why the company can’t get its own pricing straight.
Nobody lied to them. Nobody gave them a bad quote. An AI system simply answered a question with information that used to be true.
Nobody sees that failure
The tricky part is that this failure leaves almost no trace. Nobody on the account team knew ChatGPT had quoted the wrong price. Nobody in product marketing knew outdated information was still influencing the answer. The only visible symptom was a slightly harder renewal call, the kind that gets chalked up to a difficult customer or a tough quarter rather than traced back to its actual source.
And that’s when you’re lucky enough to see the symptom.
Another customer might spend thirty seconds confused, correct course, and move on. Another might decide the upgrade isn’t worth the hassle. Neither creates a support ticket or sends an angry email.
The underlying problem stays exactly where it was.
Multiply that across customers asking similar questions on ChatGPT, Claude, Gemini, and other AI systems, and those small failures can eventually show up in churn and net revenue retention (NRR).
The same thing happens before the sale
A prospect asks an LLM to compare three vendors on a specific capability.
The model draws on outdated information about one vendor and a competitor’s more recent content. The resulting comparison favors the competitor.
The prospect never mentions it to a sales rep. Why would they?
They simply move one vendor down the shortlist as they progress from initial discovery to product evaluation.
From inside that company, nothing appears to have gone wrong. There was no lost opportunity in the CRM because there may never have been an opportunity in the first place. No objection to handle. No conversation to review.
In fact, there may not even be a lead in your CRM. The prospect never raised their hand, filled out a form, or interacted with your sales team.
The company may never know it was evaluated, or why it lost.
Everything you control can be correct
Sometimes the source is the problem: an old page, outdated documentation, or third-party content that was never updated.
But everything you currently control can also be correct.
Your pricing page can be accurate. Your current documentation can be accurate. Your sales deck can reflect the latest positioning. Your product marketing team can have done everything right.
And an AI system can still give a buyer or customer the wrong answer.
The gap lives between the information you’ve published and the answer someone actually receives.
Companies spend enormous amounts of time testing what they control: websites, campaigns, sales messaging, documentation, support experiences.
Very few continuously test what AI systems are telling buyers and customers about them.
That’s becoming a significant blind spot as more product discovery, evaluation, troubleshooting, and expansion research happens through AI.
You have to test the answers
The fix isn’t complicated to describe, although it takes discipline to run.
Ask AI systems the questions your prospects and customers are actually asking. Test different scenarios and personas. Check the answers for factual accuracy, messaging, positioning, and competitive differentiation.
Then do it again.
This can’t be a one-time audit. Pricing changes. Products change. Competitors publish new content. Industry bloggers publish new content, or never update old content. Documentation gets rewritten. Sources change. AI answers change with them.
An answer that was accurate last month isn’t guaranteed to be accurate today.
Until continuous testing becomes standard practice, dark funnel failures will keep happening without your knowledge: no complaint, no support ticket, and often no opportunity in the CRM.
Just a buyer or customer making a decision based on an answer you never saw.
This is one of the problems isalo addresses by continuously testing what AI tells buyers and customers about your products and how they’re evaluated across public LLMs and owned digital channels, both before and after the sale.
Learn more about Buyer Experience Optimization and Customer Experience Optimization.