On this page
Here is an illustrative version of a complaint that comes up constantly: a SaaS company raises its starter plan from $29 to $39, updates the pricing page the same afternoon, and three weeks later a prospect pastes a ChatGPT answer into Slack still quoting $29. Nobody on the team delayed the update. The page is correct, has been correct for weeks, and the assistant is still wrong. The instinct is to assume the AI is simply broken or lazy. It is neither. It is reading from somewhere that is not your live page, and figuring out where narrows the fix from “get AI to notice us” to a short, specific list of things to check.
Two different staleness problems, not one
The complaint sounds like a single bug, but it is usually one of two unrelated problems wearing the same symptom. The first is a timing problem: your page is correct, but no AI system has re-read it since you changed it, so every engine is still working from the version it saw before. The second is a consensus problem: your page is correct, but it is not the page, or even the majority of pages, that mention the number. A review site, a comparison article, a reseller listing and a two-year-old Reddit thread all still say $29, and that crowd of older mentions outvotes your one updated page. The fix for the first problem is patience and a few technical nudges. The fix for the second is chasing down other people’s pages, which is slower and has nothing to do with anything you control on your own site.
The gap between your update and the next crawl
AI engines do not reopen your page live for every answer. ChatGPT’s search feature, for instance, runs on OAI-SearchBot, a crawler that is separate from GPTBot and whose only job is building the index ChatGPT search draws citations from. OpenAI’s own documentation for its crawlers confirms the bots exist for genuinely different purposes, but it does not publish a crawl schedule or say how often a given page gets revisited, only that changes to a site’s crawl permissions can take about a day to take effect. In practice that means there is no published number to aim for, and the honest range practitioners report is anywhere from a day to several weeks depending on how often a domain is crawled already. A page that already gets crawled often for other reasons, home page, blog, anything linked from elsewhere, tends to refresh faster than a pricing page three clicks deep that nothing else links to. Our guide to checking whether AI crawlers are visiting your site covers how to see, rather than guess, whether a bot has actually been back since your change.
The number you fixed is still winning a vote somewhere else
The slower, more common version of this problem has nothing to do with your own crawl lag. When you raise a price, the old number does not disappear from the internet, it just stops appearing on your page. It is still sitting on a listing you did not write: a “best tools” roundup from eight months ago, an affiliate comparison table, a forum answer someone gave in good faith when it was accurate, a cached PDF of your old one-pager. None of those pages are going to update themselves, and most AI engines are not distinguishing “your page, which is the source of truth” from “a page about you” with any special weight. If five pages across the web say $29 and one says $39, the $39 page does not automatically win just because it is yours and current. Our piece on the sources AI engines trust goes into how engines pick among competing mentions, and the short version is that consistency and repetition count for a lot, which is bad news if the old number is the one that is consistent and repeated.
When the mismatch is on your own page
Before blaming the rest of the internet, rule out a simpler bug: your own page disagreeing with itself. A discount or plan rename often gets applied to the visible price a customer sees but not to the structured data sitting underneath it, the schema.org Offer markup in a page’s JSON-LD block that is meant to state the price in a machine-readable way. If that block still says $29 while the visible text says $39, you have shipped two different facts on the same URL, and there is no reliable way to know which one a given engine will trust. Audit the raw page source, not just the rendered page, whenever a price or spec looks stuck. Our AI-readiness audit guide walks through checking structured data against visible content as part of a broader crawlability pass, and it is worth doing that check specifically on whatever page just changed, not just once at setup.
What actually narrows the gap
Nothing here makes an engine recrawl on your schedule, but a few things shrink the window and shrink how much the old number matters while it is still floating around:
- Keep one canonical pricing or spec URL, and make sure the visible copy and any JSON-LD on it agree, word for word and number for number, the moment you make a change.
- Say the new number in plain sentences, not just a price table a parser might not resolve cleanly, the way you would write it for a person skimming the page.
- Look up where the old number already lives, review sites, directories, old guest posts, and ask for the update rather than assuming it will fall off on its own. It usually will not.
- Check what AI engines are actually saying now, not what you assume they are saying from a complaint one customer forwarded you. Our guide to seeing which sources and citations AI relies on covers how to pull the current answer across engines on a schedule instead of waiting for the next forwarded screenshot.
A system problem, not just a you problem
This is not a small-company quirk. The UK’s Department for Business and Trade found Google’s AI Overview telling people a charity could be set up for £13 online or £40 by post, pulled from an old, unmaintained GOV.UK page, when the real cost by then was £100 online or £124 by post. The department’s response was not a plea to Google, it was an internal audit: they checked for GOV.UK pages that had gone unused for five years with fewer than 11 views and found 150 of them feeding exactly this kind of answer, then redirected or retired them and put the remaining pages on a six-month review cycle with the review dates shown on the page itself (The Register, 2026). A national government with a dedicated content team had the same problem a five-person SaaS company has: old, correct-at-the-time pages kept answering questions they were never updated to answer, long after anyone was looking at them directly.
Check it instead of guessing
The reliable move is to stop inferring staleness from one customer’s screenshot and go look. Run the price or spec through the engines your buyers actually use, side by side, and see whether the old number shows up once, consistently, or not at all anymore. A free Visibility Check gives you that same-day read across engines, and if the old number is still showing up in more than one of them, treat it as a research task, find out whether it is a crawl-lag problem on your own domain or a dead-but-indexed listing somewhere else, before you spend time re-publishing a page that was never the actual problem.