TL;DR
- The Problem: The page with the most purchase intent on a B2B site is the one AI engines can least often read.
- The Cause: Most pricing and comparison content is built in the browser, and the major AI crawlers do not execute JavaScript.
- The Evidence: 37.5% of 80 B2B SaaS companies publish no price at all, while 62.3% of pages cited for comparison questions are comparison pages.
- The Fix: Publish the number and the comparison as static text in the first HTML response, then re-ask the engines every month.
- The Test: Fetch your pricing URL without a browser and look for digits in the output.
Ask ChatGPT what your product costs, then read the answer back.
Most of the time it will quote a range it found on a review site, or it will tell the buyer to contact sales. Your pricing page, the one your team rebuilt last quarter, never entered the sentence. The cause is usually mechanical. The page is assembled in the browser, and the crawler never receives it.
That makes this the cheapest growth fix on your site and the one nobody assigns.
The Highest-Intent Page on Your Site Is the Least Engineered
HockeyStack measured 31 million unique visitors across 80 B2B SaaS companies and published the results in December 2025. Pricing pages converted at 3.8%. For a marketing leader, that conversion rate is the clearest intent signal the site produces. The sample is anonymized customer data from 80 companies, which is worth knowing before you quote it to a board.
In the same dataset, 37.5% of those companies publish no price at all. They route every pricing visitor into a form. The page exists. The number does not.
I have sat in the room where that decision gets made.
Product wants to protect the deal. Finance does not want the number indexed and compared. Sales argues that a published price kills discovery. All three arguments have merit. None of them addresses whether a machine can read the page, because crawlability has never appeared on a pricing page’s job description. So the page gets tested for button color and never tested for what a crawler receives.
Then the buyer asks an engine for a price and gets silence.
AI Crawlers Do Not Run Your JavaScript
Vercel and MERJ measured crawler behavior across Vercel’s network and published the findings in December 2024. The result was blunt. None of the major AI crawlers rendered JavaScript. The list covers OpenAI’s GPTBot, OAI-SearchBot and ChatGPT-User, Anthropic’s ClaudeBot, PerplexityBot, Meta’s crawler and ByteDance’s Bytespider.
ChatGPT fetched JavaScript files on 11.5% of its requests. Claude fetched them on 23.84%. Neither one executed them. Google’s Gemini inherits Googlebot’s rendering stack and AppleBot renders through a browser, so those two are the exceptions that make the problem easy to miss.
Hold your own pricing page against that finding. If the price appears after a client-side fetch, a seat-count slider, or an annual toggle, the crawler receives a shell with a form on it.
If your price is assembled in the browser, the answer engine is reading an empty page. The buyer gets a number from whichever competitor published theirs.
OpenAI’s crawler documentation sets the stakes. Sites that opt out of OAI-SearchBot are not shown in ChatGPT search answers. Allowing the crawler gets your site fetched. It does not guarantee the crawler finds anything worth quoting.
The Page Format Decides Whether You Get Cited
Otterly analyzed 617,985 citations across 85,330 URLs in September 2026, spanning seven AI engines and five industries.
Listicles took 66.6% of cited pages and head-to-head comparisons took 7.9%. The mix shifts with the question. When a buyer asks a comparison question, 62.3% of cited pages are comparison pages. When the buyer asks for a ranking, 73.2% of citations go to listicles. The format that earns the citation is the format that matches the question.
One more result from that study matters for pricing. Brand-owned pages were the single largest publisher type at 38.9% of identified citations. The prevailing advice says engines favor third parties. In this corpus the brand itself was the biggest source, which means a direct route into the answer exists for companies willing to publish.
You need a page the crawler can read and a format that matches the question. Most B2B sites fail both, then buy a tool to measure the resulting invisibility. If you want the measurement side, I covered how to measure AI search visibility separately. This piece is about the page.
What I Actually Think
Every audit I run now starts with a fetch, not a browser.
I pull the raw HTML of the three pages that decide most B2B deals. The pricing page. The comparison page against the two competitors buyers name. The product page for the flagship use case. Then I search that raw text for numbers. Not numbers in the design. Numbers in the bytes. When the bytes have no digits, I know what the answer engine will say before I ask it.
Most of the time the bytes are thin. A headline, a form, a script tag.
The part teams get wrong is the response. They treat AI search as a publishing problem and answer it by shipping more articles. More articles do not fix a page that renders on the client. You can double your output and still lose every pricing question in your category, because you never supplied the sentence the engine needs to quote.
I think of it as a rendering test rather than a content test. Marketing teams dislike that framing because it hands work to engineering. That is exactly why it works. An engineer can fix rendering in a sprint. Marketing cannot outwrite a page that never arrives. I have watched content teams add forty posts in a quarter while the pricing page stayed a form.
Your pricing page is a text file before it is a design. Most teams spend the quarter on the design and ship a text file with no price in it.
The Five-Step Pricing Page Audit
Run curl against your pricing URL and read the text that comes back. If the price is missing from that output, stop here. Nothing downstream matters yet.
What does the product cost? Is there a free tier? How does it compare to the two competitors buyers name? Will it fit a 40-person team? Log the answers word for word.
Count how many of the four answers contained your number, your unit and your segment. Two out of four is normal. Four out of four is rare and it is the goal.
Render a starting price, the billing unit and the minimum contract as static text. Keep the calculator for buyers who want precision. The calculator is an enhancement, not the source of truth.
One page per named competitor, with the differences a buyer checks, in text. That is the format 62.3% of comparison answers pull from.
Three of those five steps need an engineer for a day. None of them need a new content calendar, which is why they keep slipping behind the article queue.
What the Crawler Sees vs. What You Shipped
| Signal | What the crawler receives | What to ship instead |
|---|---|---|
| Starting price | Nothing until the script runs | Base price and unit as static text |
| Annual discount | A toggle with no text value | Both prices stated in one sentence |
| Seat tiers | A slider with no readable values | The tier table as HTML |
| Custom pricing | A form with no range | The floor and what moves it |
| Competitor comparison | Not built | One page per named competitor |
Two of those rows are the entire job at most companies.
The Transparency Trade Is Real, and It Is Worth Making
The transparency debate usually collapses into lead capture. HockeyStack’s numbers complicate the usual answer. Non-transparent pricing pages converted at 4.6% and transparent pages at 2.8%. Transparent pages held visitors longer: 4.26 pages per session against 2.57, and three minutes twenty-five seconds against two minutes thirty-one.
Fewer form fills per visit, far more exploration. That trade looks rational when the buyer researches before contacting anyone. Gartner’s March 2026 release puts 67% of B2B buyers in a rep-free preference. A gate on the number is a gate on the buyer who already decided not to call.
I do not read that as an argument for publishing a full rate card with every enterprise tier in it. I read it as an argument for publishing the shape of the price: floor, unit, and what changes it. That is enough for an engine to answer, and it is far less than the number your sales team wants to protect.
The comparison work goes further than the page itself. When you publish the comparison, you join the set of pages an engine can cite for every question that names you and a rival. Skip it and the rival’s comparison page becomes the answer about your product. That is the demo gate repeating one layer down the funnel. It costs more, because the buyer is already deciding.
The Move This Quarter
Pick the pricing page. Fetch it raw. If the number is missing, you have found a leak that costs less to fix than a single campaign.
The compounding part is the comparison set. Every quarter it stays unbuilt, someone else’s page answers for your name. “A company that best frames a problem is the company that often comes to define and take the category. Put another way: winning companies today market the problem, not just the solution.” That line is from Play Bigger, and it is the whole argument for publishing a price.
A pricing answer is a frame. Supply it, or the engine supplies another brand’s. The citation rules for AI search reward the same discipline: specific, structured, and plain enough to lift.
Which of the four buyer questions does your site answer in plain text today?
AI search does not reward the team that publishes the most. It rewards the team whose highest-intent pages survive a fetch.


















