ChatGPT keeps getting my company's pricing wrong. How do I fix what it says?

ChatGPT builds a pricing answer in one of two ways. Either it answers from what the model learned in training, or it searches the web and summarizes the pages it retrieves. As of October 2026, ChatGPT decides per question whether to search, and OpenAI says it rewrites your question into its own search queries first.

That split decides the fix:

• No sources shown in the answer: it probably came from training data. The likely cause is an old price frozen before the model's knowledge cutoff.
• Sources shown, and one is your site: check whether your pricing page is readable without JavaScript, and whether ChatGPT's search crawler can reach it at all.
• Sources shown, but they're other sites: a review, directory or comparison page is quoting your old price.

Before you start, confirm the error is stale rather than invented. An old price you really charged once is a different problem from a number you never charged. Our comparison of hallucinations and outdated answers covers how to tell them apart. For the broader split between trained knowledge and search, see fixing ChatGPT errors in trained knowledge vs search.

Every model has a knowledge cutoff. OpenAI's model documentation lists February 16, 2026 for the GPT-5.6 family and August 31, 2025 for the Instant model ChatGPT uses through chat-latest. So "ChatGPT" is not one knowledge state. Which model answers can move the cutoff by months.

If you changed your price after the relevant cutoff, an answer given without search can only repeat the old one. Nothing you publish this quarter reaches that answer. It changes when OpenAI ships a model trained on newer data, and no vendor publishes when that will happen.

What you can do:

• Make search answers right. Pricing questions often trigger search, and that answer draws on live pages you can fix today.
• Don't count on GPTBot. GPTBot governs training use only. OpenAI states each crawler setting is independent, and it claims no search benefit from allowing GPTBot.
• Expect a lag. See how long AI takes to update brand information for what the evidence does and doesn't say about timing.

Vercel's crawler study found that ChatGPT's and Claude's crawlers fetch JavaScript files but don't execute them, so they can't read client-side rendered content. Independent testing reported in June 2026 found the same for OAI-SearchBot, ChatGPT-User and the other major AI crawlers. Googlebot is the main exception.

If your plan names and prices load through a pricing widget, a currency toggle or a client-side app, ChatGPT's search may see an empty table. Then it fills the gap from older or third-party sources.

The fix is mechanical:

1. Load your pricing page with JavaScript disabled and read what's left.
2. If the prices are missing, server-render them or prerender the page so they're in the initial HTML.
3. Write each price as a plain sentence too, such as "The Team plan costs $40 per user per month." A specific stated fact gives the engine a line it can quote.

Then check access. OpenAI says a site must allow OAI-SearchBot in robots.txt and confirm its host or CDN allows traffic from OpenAI's published search crawler IP addresses. A firewall can return a 403 to a crawler robots.txt allows, and nothing in your dashboards will show it. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away. Our guide on publishing prices for AI answers covers the trade-offs of listing prices at all.

When ChatGPT cites other sites instead of yours, those pages are its evidence. Review sites, software directories, comparison articles and old press coverage often keep a price for years after it changes.

Look at the sources in the wrong answer and work through them:

• Pages you can edit: marketplace listings, directory profiles and partner pages. Update them yourself.
• Pages you can request changes on: review and comparison sites. Send the editor the current price and a link to your pricing page.
• Pages you can't change: old news stories. Make sure your own page states the current price clearly so a fresher source exists.

The evidence shows a strong association between being named accurately on the sites engines cite and appearing correctly in answers. It shows no guaranteed effect. Treat outreach as fixing the inputs, not as a switch.

One thing to avoid: hidden text or instructions aimed at AI crawlers. Research on search agents shows manipulated content can sway answers, which is exactly why engines treat it as abuse, and it puts your brand at risk.

Be plain with your team about the limits. You can't edit ChatGPT's training data, choose which pages it retrieves, or make it cite you. A critical survey of 45 GEO studies found no technique with a stable, cross-platform causal effect on discoverability. Typing a correction into a chat won't fix it for everyone either, as our page on correcting ChatGPT in a chat explains.

What you control is the evidence and the access. This week:

1. Ask ChatGPT your three most common pricing questions, logged out. Note whether each answer shows sources.
2. Check your pricing page with JavaScript off, and check that OAI-SearchBot can reach it.
3. List every cited third-party page with the old price and contact each one.
4. Re-check monthly, since answers vary by question and by whether search ran.

The free checker on this site gives you a snapshot of how engines answer up to three buyer questions. For the full process, follow the pricing error playbook.

How do I know if ChatGPT searched the web for its answer? Look for sources. When ChatGPT searches, it shows links or citations alongside the answer. If there are none, the answer most likely came from the model's training data. That isn't certain, because the app decides per question whether to search and doesn't always say. Ask logged out with memory off so saved context doesn't change the result, and repeat the question a few times.

Will adding Product or pricing schema fix the wrong price? There's no evidence it will. Google says structured data isn't required for its generative AI search features, and Ahrefs' matched test found adding schema produced no major citation uplift on any platform. Existing markup does no harm, so there's no need to remove it. The better use of your time is putting the price in the server-rendered HTML as plain text.

Should I block GPTBot to stop ChatGPT using my old prices? Blocking GPTBot tells OpenAI not to use your content for training future models. It doesn't remove old prices from models already trained, and it doesn't affect ChatGPT search, which uses OAI-SearchBot. OpenAI states the settings are independent. Blocking OAI-SearchBot would be counterproductive here: OpenAI says sites opted out of it won't be shown in ChatGPT search answers.

Does the same fix work for Gemini, Claude and Perplexity? The causes are similar, but each engine has its own crawler and index. Claude and Perplexity crawlers don't execute JavaScript either, so server-rendering helps across them. Google renders JavaScript, and its AI features need your page indexed and snippet-eligible in Google Search. Check each engine separately, because a fix that works on one tells you nothing certain about another.