Obility Editorial · · 4 min read
Hiding the price leaves the number to someone else
Most sales-led software companies keep pricing behind a “contact sales” button, and for real reasons: deals vary, discounts are common, and competitors read pricing pages. The assumption underneath is that a buyer who wants a number has to talk to sales first. AI answers have quietly broken that assumption. Ask ChatGPT what a contact-sales product costs and it will usually give a figure anyway. What you control is whose figure it is.
Omnius, an SEO agency, tested this in October 2026 with 300 prompts to ChatGPT about 20 AI customer service vendors, half of which showed no plan price on their own pages. ChatGPT put a figure on all ten of the hidden-price vendors. Of the 49 answers about them that completed, 38 named a figure. For the ten vendors that publish, 49 of 50 did.
The figures come from listings, rivals and estimates
When the vendor’s page had no number, the number came from somewhere else. In the Omnius test, listings on software marketplaces, most of them posted by the vendors themselves, supplied 15 of the 38 first figures. Most of the rest came from rival vendors’ blog posts, pricing websites and contract-price data. ChatGPT labelled 92 of the 147 figures it gave for these vendors as estimates, and stated them anyway.
A larger study points the same way. Kyle Poyar’s Growth Unhinged newsletter and Profound, an AI search vendor, analysed about 7,600 answers to the prompt “Evaluate Company X on pricing” for the Cloud 100 companies across six engines: ChatGPT, Google AI Mode, AI Overviews, Gemini, Copilot and Perplexity. A company’s own pricing page appeared in 46 percent of answers but came first in only 12 percent. The outside sites cited most were Vendr, Reddit and G2. For one vertical AI company with no public pricing page, 67 of 76 answers used negative pricing language, and the quoted range ran from about $100 to $2,400 per user a month for the same product.
Hidden pricing costs less shortlist presence than the warnings claim
The common warning is that AI will not recommend a company that hides its price. Omnius tested that directly by adding a budget sentence to 100 shortlist requests. The contact-sales group’s average recommendation rate fell 2.8 points and the publishing group’s rose 0.8, a gap its own authors say could be chance. Their splits by budget hint that the loss sits at mid-range budgets, where ChatGPT appears to treat a hidden price as an enterprise price, but that pattern still needs confirming.
So the honest case for acting is not that hidden pricing makes you invisible. It is that buyers now form a price expectation before the first sales call, and a competitor’s blog post or a Reddit thread may be setting it. That expectation arrives at the call as an anchor, or as a reason the call never gets booked.
You can publish the shape of the price without the price list
Full transparency is not the only alternative to silence. An answer engine needs a fact it can quote, and most sales-led companies can state several without giving up their discounting: the unit they charge by, an entry configuration with a starting price or range, any minimum or platform fee, what moves a quote up, and which parts are custom and why. Growth Unhinged’s strongest example is Plaid, whose billing documentation says plainly that a price list is not available there and then explains each billing model. Answers about Plaid cited those docs 70 percent of the time.
Write those facts as plain text in the page the server sends, not only inside a calculator, a tab that opens on click, or structured data no visitor sees. A sentence that states the unit and the starting point is easier for an engine to quote than a table it has to interpret.
Find the prices you are already publishing
Before writing anything new, look for the numbers that already exist under your name. The contact-sales vendors in the Omnius test had prices coming back from their own old blog posts, from their own cloud marketplace listings, including a placeholder figure for a private offer that ChatGPT repeated with a caveat, and in one case from a development copy of the vendor’s website left on a third-party server. Each of those is fixable by the company itself, which a Reddit thread is not.
Then measure on a fixed set of pricing questions a buyer would ask: what it costs, the entry price, how it compares with a named rival, and whether it fits a stated budget. Run the same set across engines on a schedule and record, for each answer, the figure, the first source cited, and whether the figure matches what sales would quote. In Obility, those prompts sit in Answer Engine Insights as their own group, and a wrong figure becomes a FactCheck & Sentiment item that keeps the answer, engine, date and cited source together, so the correction has an owner and can be rechecked.
Two limits are worth keeping in mind. An engine will still quote a third party when that source is more specific than you are, so a vague page loses to a precise estimate. And a fix on your own pages does not reach answers the same day, so recheck after every pricing change rather than assuming the old number has gone.