One AI Engine Quoted a SaaS Seat at $100. Another Quoted $2,400. Neither Price Came From the Pricing Page.
A 7,600-response study found AI engines quoting the same SaaS seat 24x apart. Here's why that breaks renewals, not just signups.
Every SaaS pricing page used to have exactly one audience: a human, reading it before they signed up. That's no longer true. A growing share of pricing questions — from prospects, and now from existing customers deciding whether to renew — get asked to an AI engine instead of typed into a URL bar. Stripe's own Monetize 2026 conference put a number on what happens when that engine can't actually read your page, and the number is worse than most pricing teams assume.
What the study actually measured
Pricing analyst Kyle Poyar ran 7,600 AI-generated responses across six major AI engines, asking about pricing for companies on the Cloud 100 list — a reasonable proxy for well-funded, high-growth software companies that, in theory, have the resources to get their pricing page right. For one unnamed company in the sample, 88% of the AI-generated answers used negative language when describing its price. The quoted figure for a single seat ranged from $100 to $2,400 a month, depending on which engine answered and which source it pulled from.
Source: Kyle Poyar, Growth Unhinged, presented at Stripe's Monetize 2026 conference
That gap isn't a model being creative with currency conversion or plan tiers. It's the predictable result of an AI engine that couldn't find a real price and reached for whatever it could find instead — old coverage of a funding round, a stale review, a forum post from someone comparing tools eighteen months ago. None of those sources know about your last price change. All of them sound confident enough that a user has no reason to doubt the number.
Why this happens: the pricing page was never built for a reader that can't click
Most pricing pages were designed, tested, and iterated against a human visitor — someone who scrolls, clicks a toggle to switch between monthly and annual, maybe opens a calculator to size seats. An AI crawler doesn't do any of that. It fetches whatever HTML comes back from the first request and works with that. If your actual price only appears after a script runs, after a login, or not at all because the page is gated behind a "request a demo" form, the crawler gets nothing useful — and nothing useful is exactly what produces a confidently wrong answer downstream, because the model still has to say something.
| Common practice | What the AI engine does instead | What the reader sees |
|---|---|---|
| Every tier gated behind "Contact sales," no number shown | Falls back to old funding coverage or a competitor comparison blog | A guessed figure, sometimes off by 10x or more |
| Price only renders after a JS calculator finishes loading | Crawler sees an empty page and cites a cached snapshot instead | A price from months or years ago, presented as current |
| Pricing or plan-comparison page blocked in robots.txt | Pulls from Reddit threads, G2 reviews, or third-party roundups | Someone else's anecdote, treated as your official price |
| Price changed, but the old page is still ranking in search | Cites whichever indexed version it finds first, old or new | A number you haven't actually charged in months |
Every row in that table describes a pricing page that would look completely normal to the person who built it. Gating enterprise pricing behind a sales call is a defensible go-to-market decision. Building your plan comparison as an interactive calculator is good UX. Neither one was designed with "what does a crawler see on first request" as a requirement, because until recently, nothing was asking that question at scale.
This stops being an acquisition problem the moment it touches a renewal
Most of the commentary around this finding treats it as a top-of-funnel issue — a prospective buyer gets a wrong number and doesn't convert. That's real, but it's the smaller half of the risk. The larger half shows up at renewal, where the question isn't "should I sign up" but "should I keep paying," and we already know a meaningful share of that decision is starting to move away from the customer themselves. In our piece on agentic cancellation churn, we covered survey data showing 74% of consumers say they'd delegate a subscription renewal decision to an AI agent outright. Put that alongside Poyar's pricing-accuracy numbers and the mechanism writes itself: an agent (or a customer who just asked an AI engine the same question a human used to ask a salesperson) checks "is this still worth it" or "is there something cheaper," and the answer it gets back is built on a number that's wrong in either direction.
Wrong-high is the obvious damage — a customer sees an inflated price for their own plan, or sees your real competitor's price understated, and cancels over a comparison that was never accurate. Wrong-low is just as costly in a quieter way: if the AI understates what a competitor actually charges, a customer might switch expecting savings that don't exist, discover the real invoice later, and churn from the new vendor too — except by then you've already lost the renewal that would have kept them. Either direction, the defining feature of this churn is that nobody on your team gets a chance to correct it before the decision is made. A sales rep on a renewal call can say "that number's not right, here's what you're actually paying." An AI engine answering a question in someone's chat window has no equivalent correction step, and your cancellation flow is often the first moment you'd even find out the decision was based on bad information at all.
This sits next to a pattern we've described in AI substitution churn — customers increasingly routing a task through a general AI assistant instead of the tool built for it — but it's a distinct mechanism. Substitution churn is about the AI doing the job your product used to do. This is about the AI doing the job your sales and pricing pages used to do, badly, because it was never given the real data to do it with.
What to actually check this week
Start with the simplest test: fetch your own pricing page the way a crawler would, with JavaScript disabled, and see whether a real number is in the HTML that comes back. If it isn't, that's the single highest-leverage fix available, well ahead of anything more elaborate — a page that renders its price as plain server-side text, even behind a toggle for billing frequency, is legible to an AI engine in a way a page that waits for a calculator to hydrate is not.
Second, check robots.txt and any noindex tags on plan-comparison or enterprise-tier pages specifically, since those are the ones most commonly locked down for sales-process reasons and the ones most likely to get a guessed number in their place when they're invisible. Third, treat a price change the same way you'd treat a renewal price change communicated to a human — don't just update the live page and assume search and AI indexes catch up on their own. A stale indexed copy sitting next to your current one is exactly the condition that produced Poyar's 24x spread.
Finally, add a specific option to your cancellation survey for this: something closer to "I was told this costs more than it does" or "I found a cheaper price elsewhere" rather than letting it blend into a generic "too expensive" reason. Right now almost no cancellation flow distinguishes a customer who did the math correctly from one who's leaving over a number an AI engine invented. CancelFlow can only ask that question if you've built the option into the flow first — and once you have, a customer citing a wrong price is one of the few cancellations where the honest fix isn't a discount, it's just telling them what they're actually paying. Our churn calculator is a quick way to see what even a small share of price-perception-driven cancellations is costing you before you decide how much engineering time this is worth.
Frequently asked questions
Why would an AI chatbot give the wrong price for a SaaS product?+
Because it never read your actual pricing page. If that page sits behind a login wall, renders its numbers only after a JavaScript calculator loads, or is blocked in robots.txt on the enterprise tier, an AI crawler sees either nothing or an empty shell where the price should be. When a user then asks an AI engine what your product costs, the model doesn't say "I don't know" — it answers from whatever secondhand source it can find instead: a two-year-old funding-round article, a Reddit thread, a G2 review written before your last price change. None of those sources update when you do.
What did Stripe's Monetize 2026 conference find about AI and SaaS pricing?+
Pricing analyst Kyle Poyar, writing as Growth Unhinged, ran 7,600 AI-generated responses across six AI engines asking about pricing for companies on the Cloud 100 list, and presented the results at Stripe's Monetize 2026 conference (covered on Stripe's blog on October 13, 2026). For one unnamed company, 88% of the AI-generated responses used negative language about its pricing, and the quoted price for the same product ranged from $100 to $2,400 per user per month across different engines and sources — a 24x spread for one product in one month.
Can bad AI pricing information actually cause an existing customer to cancel, not just stop a new signup?+
Yes, and it's a different failure than losing a lead who never converted. We've covered separately that 74% of consumers say they'd delegate a subscription renewal decision to an AI agent — the mechanism in agentic cancellation churn. Put those two findings together: if a customer or their agent asks an AI engine "is there a cheaper alternative" or "what does my current plan actually cost" right before a renewal, and the engine answers from a stale cache or a scraped secondhand number, the customer is making a stay-or-go decision on a price that isn't real. Unlike a sales call, there's no human in that exchange to correct the record before the decision gets made.
How do I make my SaaS pricing page readable to AI engines?+
Four checks, in order of how often they're the actual problem: confirm your pricing page returns real price text in the initial server-rendered HTML, not only after a client-side calculator finishes loading; remove any robots.txt block on pricing or plan-comparison pages, including the ones for your highest tier; don't gate every tier behind "Contact sales" with no number attached, since an engine with nothing to read will substitute something from elsewhere; and after any price change, check what a fresh crawl of the page actually returns within a day or two, because a stale cached version competing with your live page is exactly how a 24x spread happens.
Stop losing subscribers today
One script tag. One function call. A live cancellation flow in under 10 minutes.
Start free trial →