TL;DR
- AI answers misquote SaaS prices mostly by repeating the old price.
- Removed plans are the riskiest change, misquoted in 11.5% of answers.
- Gemini misquoted 12.8% of answers on changed tools, ChatGPT 5.7%.
- Five tools hold 41 of the 71 stale answers.
- Retire old pricing pages and get third-party articles updated.
ChatGPT and Gemini get SaaS pricing wrong mostly by repeating a price the vendor has already dropped. On 77 tools that changed their pricing, Gemini gave a stale or wrong price in 12.8% (59 of 462 answers) and ChatGPT in 5.7% (26 of 460 answers). Old prices make up 83.5% of those errors (71 of 85 answers), and removed plans are the most misquoted change at 11.5% (65 of 563 answers).
The two engines fail in different ways. 70.3% of Gemini’s stale claims (52 of 74 claims) appear in a third-party article it cited, while 82.9% of ChatGPT’s stale claims (34 of 41 claims) appear on no page it listed. ChatGPT hedges in 51.9% of its answers (359 of 692), and 78.6% of its stale or wrong answers carry that hedge next to the bad price (22 of 28 answers). Three vendors, Proposify, Beautiful.ai and UptimeRobot, still host their own pages with the old price.
Gemini misquoted 12.8% of answers on changed tools, ChatGPT 5.7%
Both engines were asked what 116 SaaS tools cost, from a US location on 12 October 2026. 77 tools had changed their pricing and 39 had not. The 39 unchanged tools are the baseline: they still drew a wrong price in 2.6% (12 of 466 answers).
| Engine | Tools that changed pricing (77) | Tools that did not (39) |
|---|---|---|
| ChatGPT | 5.7% (26 of 460 answers) | 0.9% (2 of 232 answers) |
| Gemini | 12.8% (59 of 462 answers) | 4.3% (10 of 234 answers) |
| Both engines | 9.2% (85 of 922 answers) | 2.6% (12 of 466 answers) |
Overall, 91.1% of answers (1,265 of 1,388) carried a correct price and no stale or wrong one. Both engines nearly always name a price: 1.0% of answers (14 of 1,388) gave none. That makes price accuracy part of any GEO program for a B2B SaaS brand, since the engine will state a number whether it is current or not.
The changed group was picked because its pricing moved. Its error rate describes recently repriced tools and does not carry over to SaaS in general. Engine choice changes the answer here, as it does for the brands each AI engine recommends.
83.5% of errors on changed tools are old prices
A stale answer states a price that matches the vendor’s old pricing page and no live price. On changed tools, 83.5% of error answers were stale (71 of 85 answers).
- ChatGPT: 96.2% of its errors on changed tools were stale (25 of 26 answers).
- Gemini: 78.0% were stale (46 of 59 answers). Gemini also gave 13 wrong prices on changed tools, against 1 for ChatGPT.
- Unchanged tools cannot have stale answers, because their old and live prices are the same. All 12 of their errors were wrong prices.
Removed plans are the most misquoted change
A tool counts under every kind of change it made, so the rows overlap.
| Change | Tools | Stale or wrong answers |
|---|---|---|
| A plan was removed | 47 | 11.5% (65 of 563 answers) |
| A listed price moved | 40 | 5.4% (26 of 479 answers) |
| A plan was only renamed | 25 | 5.4% (16 of 298 answers), pattern only, too small to recommend |
By engine, plan removal gave ChatGPT 6.4% (18 of 281 answers) and Gemini 16.7% (47 of 282 answers). Both engines still list plans the vendor no longer sells.
Gemini repeats third-party articles, ChatGPT’s old prices sit on no page it lists
Each stale claim is matched against the pages the answer listed as sources. A claim counts under every kind of page that shows it, so the shares in a row overlap.
| Engine | On no listed page | In a third-party article | On a vendor page | On a review site |
|---|---|---|---|---|
| ChatGPT | 82.9% (34 of 41 claims) | 0.0% (0 of 41 claims) | 17.1% (7 of 41 claims) | 0.0% (0 of 41 claims) |
| Gemini | 21.6% (16 of 74 claims) | 70.3% (52 of 74 claims) | 23.0% (17 of 74 claims) | 21.6% (16 of 74 claims) |
Gemini’s stale prices usually come from a page it cited, most often a third-party article that still carries the old price. Its wrong prices follow the same route: 62.7% of Gemini’s wrong claims (32 of 51 claims) appear in a third-party article. For Gemini, a stale price is a wrong AI answer about your brand with a URL attached, and the pages Gemini cites decide what it says about your price.
ChatGPT’s stale prices mostly appear on no page it listed, and the study cannot say where those figures came from. No stale or wrong claim from either engine traced to a Reddit thread or a video. The figure an engine repeats often lives on pages the vendor does not run, the same problem as pricing that other sites publish for you.
ChatGPT warns the reader, then states the old price
ChatGPT hedges in 51.9% of answers (359 of 692): it tells the reader to check the vendor’s page or says prices may change. Gemini hedges in 2.0% (14 of 696 answers).
- 78.6% of ChatGPT’s stale or wrong answers carry a hedge (22 of 28 answers). For Gemini it is 1.4% (1 of 69 answers).
- ChatGPT’s hedged answers carried a stale or wrong price more often than its unhedged ones: 6.1% (22 of 359) against 1.8% (6 of 333).
The hedge sits next to the stale price, and the answer still states the price. A buyer who skips the warning leaves with the old number.
Proposify, Beautiful.ai and UptimeRobot still host their own old prices
These three vendors carried the old price on their own domain on 12 October 2026. Any of them may have changed the pages since.
Proposify
11 of 12 answers were stale (ChatGPT 6, Gemini 5). The old Basic plan at $29 monthly or $19 annual and Team at $49 or $41 appear on the old Proposify pricing page, on the blog post Is Proposify right for you and in the subscription plans and pricing help article. The live Proposify pricing page shows Build at $49 monthly and $35 annual.
Beautiful.ai
11 of 12 answers were stale (ChatGPT 5, Gemini 6). The Beautiful.ai pricing-plans page shows Pro at $12 a month billed annually. The Beautiful.ai pricing page shows Pro at $14.50. Four ChatGPT stale claims traced to the pricing-plans page. A Beautiful.ai blog post, Is Beautiful.ai free to use, also carries the $12.
UptimeRobot
2 of 12 answers about UptimeRobot were stale, both from Gemini. The UptimeRobot knowledge-hub page lists Solo at $7 a month. The live UptimeRobot pricing page shows Solo at $12 annual and $13 monthly.
A few tools carry most of the errors
- 25 of 77 changed tools (32.5%) drew at least one stale or wrong answer: ChatGPT on 11 of 77 tools, Gemini on 20 of 77 tools.
- The 71 stale answers sit in 18 tools. Five tools hold 57.7% of them (41 of 71): Beautiful.ai 11, Proposify 11, Profound 7, Framer 6, Pipedream 6.
- Without Beautiful.ai, Proposify and UptimeRobot, the changed-group error rate is 6.9% (61 of 886 answers): ChatGPT 3.4% (15 of 442), Gemini 10.4% (46 of 444).
Read the group rates as an average over a few bad cases and many clean ones. Gemini still misquotes more often than ChatGPT once the three vendors drop out.
What to fix after a price change
Retire or redirect old pricing pages
Proposify’s pricing-old page and Beautiful.ai’s pricing-plans page both still show the old price, and four ChatGPT stale claims traced to the Beautiful.ai page. Point any retired pricing URL to the live one with a 301 redirect. A page with no reason to exist goes on the list when you prune outdated content for AI search.
Update blog posts and help articles that state prices
Vendor pages carry 20.9% of all stale claims (24 of 115 claims). The old prices on vendor domains in this study sit in a Proposify blog post, a Proposify help article, a Beautiful.ai blog post and an UptimeRobot knowledge-hub page, next to the two old pricing pages. Search your own site for every old plan name and old amount. Rewrite each hit as a real content update, with the new price in the text.
Ask third-party articles that rank for your tool to update
70.3% of Gemini’s stale claims (52 of 74 claims) appear in a third-party article it cited. Those are the reviews and best X lists on other sites that rank for your tool. Send each publisher the old price, the new price and the live pricing URL. It is the same outreach as digital PR for AI citations, aimed at fixing a number. When a page will not change, correct the claim at its source with a page of your own that states the current price.
Treat plan removals with extra care
Removed plans drew stale or wrong prices in 11.5% (65 of 563 answers), and Gemini’s rate on those tools was 16.7% (47 of 282 answers). Both engines still list plans the vendor no longer sells. When you drop a plan, remove its name and price from every page you run. Tell the publishers above which plan replaced it.
State prices in llms.txt (a pattern to test)
Tools whose llms.txt file states a price had fewer stale or wrong answers: 4.2% (10 of 240 answers, 20 tools) against 11.0% (75 of 682 answers, 57 tools) without. The group with a price is 20 tools, so this is a pattern only, too small to recommend. Offer schema markup on the pricing page leaned the same way, also from a group too small to recommend. Because errors cluster in a few tools, one or two tools can move either cut. Adding the price to llms.txt costs little, so treat it as a test.
Ask both engines after every change
ChatGPT and Gemini misquote prices through different routes, so check both. Ask each engine the two prompts from the method section, three times, and compare every price to your live page. The same run fits into a free AI brand visibility audit.
Method
- Engines: ChatGPT and Gemini. Google AI Mode is not included.
- Sample: 116 well-known B2B SaaS tools picked for search demand. 77 changed their pricing since an archived pricing page dated between 1 November 2025 and 28 February 2026. 39 kept the same plan names and headline prices.
- Prompts: “How much does {tool} cost?” and “What plans does {tool} offer and what do they cost?”, each asked 3 times per engine.
- Answers: 1,388, collected from a US location on 12 October 2026. 4 ChatGPT answers failed to collect and are missing.
- Correct: the amount matches the live price of that plan in any billing state captured, within $1 or 2%.
- Stale: the amount matches the old price only, or the answer presents a plan that exists only on the old pricing page as current.
- Wrong: the plan is on the live page, its live price is visible for the billing period the answer states, and the amount matches neither the live nor the old price.
- Excluded from errors: add-ons, usage rates, custom quotes, slider prices, third-party figures and prices behind a billing toggle that was not captured.
- Error rate: stale plus wrong answers divided by all answers in the group, including answers with no price we could check. An answer with several prices takes its worst label.
- Floor: a price counts as correct if it matches the live page in any billing state or currency view, so every error rate is a floor.
- Source trace: a page shows a claim when the answer listed it as a source and it carries the amount within 300 characters of the plan name. Pages the answer did not list are not checked.
- Hedged: the answer tells the reader to check the vendor’s page or says prices may change or vary.
- Small groups: any subgroup under 30 tools is marked pattern only, too small to recommend.