A 30-day study found AI engines misquoting pricing in 38% of answers. How to write prose, billing units, and a pricing FAQ so ChatGPT and Perplexity cite your real price.
Half of B2B software buyers, 51%, now start their research with an AI chatbot more often than Google, up from 29% eleven months earlier, and 71% use one somewhere in the process, according to G2's March 2026 survey of 1,076 B2B decision-makers. Some of what they're asking is comparison shaped, "X vs Y," which we've covered in GEO for comparison pages. A lot of it isn't. It's "how much does X cost," "is X free," "X enterprise pricing," a fixed family of budget-intent phrasings that has almost nothing to do with "best" or "vs" fanout, and that your pricing page has to answer correctly on its own.
GEO for pricing pages, generative engine optimization applied to the one page type a buyer reads right before deciding, means writing the price itself so an assistant extracts it correctly: a full sentence with the number and the billing unit, a stated reason for anything custom, and an FAQ in the buyer's own words. This post covers why cost queries behave differently from comparison queries, what a 30-day monitoring study found AI engines actually get wrong when they quote a price, and how to structure tiers, add-ons, and FAQs so they stop.
A cost query and a comparison query look like cousins, both are pricing related, but an assistant processes them as separate jobs. "Salesforce vs HubSpot" triggers a fanout into brand-specific sub-searches and a synthesized verdict. "How much does Salesforce cost" doesn't fan out into anything. It's a lookup: find the number, find the unit it's measured in, quote both correctly. There's no verdict to synthesize, only a fact to extract, which means the entire query lives or dies on whether your page states that fact in a form a model can lift cleanly.
That family is narrow and repeats in the same shapes across almost every SaaS buyer: how much does it cost, is there a free plan, what's the difference between the plans, do you charge per seat, what does enterprise pricing look like. None of those phrasings ask you to be the best. They ask you to be correct. Tim Sanders, G2's Chief Innovation Officer, frames the underlying shift plainly: "Now, AI chatbots are compressing it into a single answer. Buyers have moved from reference to inference," he said, describing how buyers now trust a chatbot to return a shortlist in one prompt instead of gathering and synthesizing sources themselves, per the same G2 release. A cost query is the sharpest version of that compression: one number, extracted once, believed on the spot.
It's worth naming what a budget question is actually asking for, because it isn't only the number. A buyer typing "under $500 a month" into an assistant needs cost boundaries, sure, but also deployment reality, seat limits, and what's included at that tier, the same point made in guidance on GEO for B2B SaaS pricing pages. A budget question is a fit question wearing a dollar sign. Answer the number without the surrounding shape of what it buys, and you've technically answered the query while leaving the buyer to guess the part that actually decides the deal.
Transparent pricing has been B2B tech buyers' number one wish-list item for vendors for four straight years running, since TrustRadius started tracking it in 2023, per TrustRadius's 2026 B2B Buying Disconnect Report, a survey of 1,862 buyers and 444 vendors run in January 2026. That same report found 63% of buyers used AI somewhere in their purchase journey, but 94% of them fact-check what the AI tells them at least some of the time. Rajat Bhatnagar, SVP of Growth at HG Insights, quoted in the same report, put the buyer's posture this way: "Buyers are using AI to move faster, not to think less. They want speed, but they still want the efficiency of AI synthesis and the confidence of verified sources."
That fact-checking habit is the good news hiding in a scary-looking stat. A wrong price quoted by an assistant usually costs you a shortlist slot, not a silent, permanent loss, because a fair share of buyers will go check it against your actual page before they act on it. But "usually caught" isn't "always caught," and the exposure runs the other way too: 69% of buyers in G2's survey chose a different vendor than they originally planned to, based on what a chatbot told them, and 85% think more highly of a vendor an assistant mentions favorably. A pricing query answered wrong is a coin flip you didn't need to take: sometimes it costs you the slot before a human ever checks, sometimes it just adds friction the buyer has to route around to reach you at all.
Here's the part that should be reassuring: the errors aren't exotic. AEO agency MaxAEO ran a 30-day monitoring case study on one of its own mid-market B2B SaaS accounts, tracking a fixed set of 40 pricing prompts, how much does [brand] cost, is [brand] free, [brand] enterprise pricing, [brand] vs [competitor] price, against eight AI assistants. In week one, 38% of pricing-intent answers contained at least one inaccuracy. After roughly three weeks of fixes to the underlying source pages, that fell to about 9%, according to the agency's published write-up. It's a single account, not a peer-reviewed study, so treat the exact percentages as directional rather than universal, but the fix pattern lines up with everything else in this post: reporting bad answers to the assistants didn't help. Fixing what the assistants were reading did.
The error breakdown from that study tells you exactly where to spend effort:
| Error type | Share of misquotes |
|---|---|
| Stale tier | 34% |
| False "free" claim | 19% |
| Phantom enterprise number | 15% |
| Wrong billing basis | 12% |
| Currency/region mismatch | 8% |
| Competitor comparison error | 8% |
| Anchor/bundle confusion | 4% |
Stale tiers alone are a third of all errors, and the study's own explanation is mechanical, not mysterious: "stale tiers dominate because pricing pages change faster than assistants re-crawl them." That's the same freshness gap we cover generally in pricing page SEO, the discipline of keeping a plain-text price current and re-checked on a schedule; we won't re-derive the crawl-and-JavaScript mechanics here. What's specific to the cost-query family is the next three rows: a false free claim, a phantom enterprise number, and the wrong billing basis together account for 46% of misquotes, and none of those three are freshness problems. They're structure problems. The page was current. It just wasn't written so a model could quote it right.
A styled pricing card with "$79/mo" in a big font next to the plan name looks obvious to a human and is close to useless to a model trying to lift a quotable fact. The fix is writing the price as a complete sentence next to the card, something like "the Team plan is billed at $79 per seat each month when you pay annually," rather than relying on the card's visual hierarchy to do the explaining, a pattern documented in GEO guidance for SaaS pricing pages, which notes plainly that models extract prose more consistently than design-heavy layouts. The card can stay, for the human reading the page. The sentence next to it is what gets quoted correctly.
This is the general engine-mechanics point running under everything below: how ChatGPT, Perplexity, and Claude retrieve and cite differently still applies to a pricing page, the same divergent per-engine habits that shape a comparison page's citations shape a cost query's too. A full sentence is the one format that survives all three retrieval styles, because it doesn't depend on any engine correctly parsing a table's visual layout.
Each of the three biggest non-freshness error categories, wrong billing basis, phantom enterprise numbers, and the false free claim buried in anchor/bundle confusion, has a specific structural fix. None of them require new tooling. They require writing the number differently.
"Wrong billing basis" is 12% of misquotes in the study above, and it's the easiest of the three to fix because the omission is usually just missing words. "$79" alone doesn't say per seat or flat, monthly or annual. "$79 per seat each month, billed annually" says both in the same breath. Do this for every tier, not just the flagship one, and do it in the sentence itself rather than a separate legend or footnote a model might not connect back to the number.
A phantom enterprise number is 15% of misquotes, and it exists because a blank field is an invitation, not a null. When your enterprise tier just says "Contact us," a model answering "what's [brand] enterprise pricing" has nothing true to extract and will sometimes fill the gap with a plausible-sounding figure pulled from a competitor's page or a stale third-party listing. Replace the blank with a sentence explaining what actually drives the quote: seat count, usage volume, SSO and compliance requirements, dedicated support. "Enterprise pricing is based on seat count, usage volume, and whether you need SSO or a dedicated support SLA, contact us for a quote" gives a model a true thing to say, custom pricing, and why, instead of a number to invent.
A pricing FAQ works because it mirrors the exact question-and-answer shape an assistant is already trying to produce. Write the questions the way a buyer actually types them, not the way marketing would phrase a header, and answer each in two or three plain sentences directly under the question, per guidance on optimizing pricing pages for AI search. Mark it up with FAQPage schema once the prose is right, and keep the schema text word-for-word identical to what's on the page, so there's no drift between what a model reads and what it can quote. We're consistent about this stance across the blog: schema is hygiene, not a citation lever, and a pricing FAQ is no exception. The FAQ earns citations because the answers are right and specific, not because a <script type="application/ld+json"> tag exists.
Commercial-intent queries, the research-and-compare-before-buying phase a pricing FAQ lives in, make up 8% of Google AI Overviews' cited volume against 3% for ChatGPT, with pure transactional intent the smallest bucket on both platforms, per BrightEdge's AI Hyper Cube analysis from May 2026. That's a small slice of total cited volume on any single query type, and it's exactly the slice a buyer sits in right before they decide, which is why getting it right is worth disproportionately more than its share of traffic would suggest.
Anchor/bundle confusion and currency or region mismatch together are 12% of misquotes, the smaller end of the breakdown but the easiest to let slide because they feel like edge cases. They aren't, once your buyers span more than one region or your plans include add-ons that change the effective price. If a seat add-on or an overage fee changes what someone actually pays past the sticker price, say so in the same sentence as the base price rather than in a separate add-ons table a model might never connect to the tier it modifies. If pricing differs by currency or region, state both explicitly, "$99 USD per user per month; £79 GBP for UK accounts," rather than showing one number and letting a geo-redirect handle the rest silently. A model reading a single snapshot of your page has no way to know a redirect exists.
Getting a pricing page's cost queries right is the same discipline as getting its comparison queries right, stated facts, checked against a real source, kept current, just aimed at a narrower, higher-stakes question: one number, one billing unit, one honest reason when the number is custom. Lyra's pricing verification runs on exactly this logic. When you set a pricing_page_url, she fetches that page with WebFetch before her writer or reviewer cites any number, checks the figure against what she reads there, and requires a disclaimer that the price is current as of the post's date, the same mechanism covered in pricing page SEO. A number she can't verify against your own page gets flagged, not guessed, the same fact-checking standard she holds every other claim to before a post ships.
If you want the autonomous writer to write your next pricing-adjacent post with that verification built in, or you'd rather talk through what a pricing_page_url setup looks like for your account, see the plans or talk to the founder.
A pricing page that gets quoted wrong costs you a shortlist slot before a human ever sees it. Lyra checks every number she cites against your own pricing page before a post ships, and flags what she can't verify instead of guessing.
FAQ
No. Stale tiers are the single biggest error in a 30-day monitoring study of eight AI assistants, 34% of misquotes, but the other three cost-specific error types, false free-plan claims, phantom enterprise numbers, and wrong billing basis, add up to 46% and have nothing to do with freshness. A page can be current and still get misquoted if the price isn't written as a complete, quotable sentence with the billing unit stated.
Yes. A cost query is a fixed, budget-intent phrasing, how much does it cost, is it free, what's enterprise pricing, that an assistant answers by extracting a specific number and its billing unit from your page. A comparison query fans out into multiple brand-specific sub-searches and rewards off-page reputation. Cost queries reward exactly one thing: a page that states the number in a complete, quotable sentence.
Their own line, stated explicitly next to the base price rather than left to a separate table or a silent geo-redirect. Anchor/bundle confusion and currency or region mismatch together are a smaller share of misquotes, about 12% in the study above, but a model reading one snapshot of your page has no way to know an add-on fee or a regional redirect exists unless the sentence says so.
State a rationale instead of leaving the field blank. Say what actually drives the custom quote, seat count, usage volume, support tier, SSO and compliance requirements, so a model has a reason to describe pricing as variable instead of a number to guess. A blank enterprise tier is exactly the gap a phantom-number error fills.
Questions in the buyer's own words, not marketing language: how much does it cost, is there a free plan, what's the difference between the plans, do you charge per seat. Answer each in two or three plain sentences directly under the question, and keep the on-page text identical to whatever FAQPage schema you generate from it, so there's no drift between what a model reads and what it can quote.
Built by the tool you're reading about
Lyra finds the topics worth ranking for, writes them in your repo's voice, fact-checks every claim, and opens a pull request scored and ready to merge. You review and hit merge. Want to see what she'd write for you? Start free with three posts, no card.
Keep reading
AI referral traffic 2026, with real numbers: ChatGPT drives 92% of it, Claude grew 64x, and why the session count still undersells GEO's payback for SaaS blogs.
ChatGPT hallucinating about your company? Fix the cited source, force a recrawl with schema and llms.txt, then verify the correction holds across engines.
Original research content earns an 82% AI citation rate, the highest of any format. Here's how SaaS teams turn first-party data into a citable, dated post.