Pricing is where AI gets companies wrong most often. Where the errors come from
Profound's new analysis finds pricing and billing claims are the most error-prone, seeded by third-party and owned pages alike. A fix list for both.
The short answer
Pricing errors in AI answers come from two places: your own pages, when old plans and undated price tables stay live, and third-party pages, such as resellers, marketplaces, review sites and old press coverage. Profound's August 2026 analysis found pricing and billing the most error-prone topic. Fix your pricing page first, then the outside listings engines quote.
Key takeaways
- Profound's analysis of claims checked by its fact-checking product found pricing and billing to be the topic AI engines get wrong most often.
- Errors clustered in claims citing competitors' pages and earned media, and brands' own pages contributed too, so a clean pricing page alone does not fix the problem.
- Inaccuracies were spread across many different domains, which means the fix is a list of listings, not one outreach email.
- Prices, offers and eligibility change often, so every price on the web needs an effective date an engine can read.
- Track pricing claims per engine, because engines differ in how often they get them wrong.
In this article
What Profound found
Profound, which sells an AI visibility platform, published an analysis on 17 August 2026 of claims that AI engines make about brands. It grouped equivalent claims, sorted them by topic and by the type of source behind them, and checked each against the facts.1 Its conclusion: pricing and billing is the topic engines get wrong most often, and the errors cluster in claims citing competitors' pages and earned media, with brands' own pages contributing too.
We cite the study without its figures because we have not been able to check its sample and method independently. The direction is clear enough to act on, and it matches what anyone who has audited AI answers about a software or services company will recognise. Prices change; old prices linger on the web; engines repeat whichever version they find.
Our view: pricing errors cost more than most errors. A wrong founding date is embarrassing. A wrong price changes who contacts you, what they expect to pay and how the first sales call goes.
Where do pricing errors in AI answers come from?
They come from both sides of the fence: your own pages and other people's. Profound reports errors traced to brands' own content, to competitors' pages and to earned media, spread across many different domains rather than concentrated on a few. It also notes that pricing, offers and eligibility change frequently, which makes them likely to go out of date.1
Your own site seeds errors in predictable ways. Old plan pages stay indexed after a pricing change. Promotional prices sit on landing pages after the promotion ends. The pricing page shows prices in one currency without saying so, or shows a monthly figure with annual billing in small print. Some prices live only in an image or behind a calculator, so the engine finds an older text version somewhere else.
Third parties add the rest. Reseller and marketplace listings carry the price from the day they were set up. Comparison and review sites quote a plan you retired. Old press releases announce a launch price. Competitor comparison pages describe your pricing in the least flattering way they can defend.
A fix list for owned and third-party pages
Start with the pages you control, because they are the fastest to change and engines lean on them heavily. Then work through the outside listings in order of how often they appear as sources in answers about your pricing.
| Source | Typical error | Fix |
|---|---|---|
| Your pricing page | Undated, image-only or currency unclear | HTML table, currency, billing terms, effective date |
| Old plan and promo pages | Retired prices still indexed | Redirect to the current page or mark as ended |
| Help centre and docs | Plan limits from an older version | Link limits to the pricing page; review on every change |
| Resellers and marketplaces | Price set at listing time, never updated | Update listings; add pricing to partner agreements |
| Review and comparison sites | Retired plans quoted as current | Request corrections with a link to the dated page |
| Old press releases | Launch prices read as today's | Add a dated note pointing to current pricing |
Our fix list. Illustrative: order the third-party rows by how often each appears as a source in your own answer checks.
The effective date matters more than it looks. An engine comparing two prices for the same plan has no reliable way to pick the newer one unless the page says when the price took effect. Our own pricing page states every price as text in two currencies, and our review of published GEO prices shows how rarely the category publishes prices at all.
Why sourcing errors persist
Engines often cite a source that does not support the claim next to it, which makes pricing errors hard to trace. The EBU and BBC studied AI news answers collected in May and June 2025, data now more than a year old. In that study, 45% of answers had at least one significant issue and 31% had sourcing problems.2 That study covered news, not pricing, but the mechanism is the same: the cited page and the stated fact do not always match.
So when an engine states a wrong price, check every cited source, then search for the wrong figure itself. The page that seeded it may not be among the citations. Our guide to fixing what ChatGPT gets wrong about your company walks through that trace, and our summary of the accuracy studies covers what else they measured.
Why the error reaches your sales team
Because buyers bring it with them. In Gartner's survey of 645 B2B buyers from August and September 2025, data now close to a year old, 69% said they prefer to validate AI-generated insights with sales reps.3 A wrong price in an AI answer becomes a question on the first call, or a lead that never calls because the price looked out of range.
Our view: give sales a short note on what engines currently say about your pricing, updated after each check. A rep who knows the engine quotes last year's plan can correct it in one sentence instead of losing the first five minutes.
How to monitor pricing claims
Write ten to fifteen pricing questions a buyer would ask: what a plan costs, what is included, whether there is a free tier, how billing works. Run them across engines on several days and record every price stated and every source cited. Repeat after each pricing change, because the old figure can take weeks to fade.
Track the results per engine. Profound's analysis found inaccuracy rates differ between engines, so a fix that clears one may leave another unchanged.1 At Sigzen AI, our AI answer monitor covers six engines: ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI. It uses official APIs where offered and a licensed AI-answer data provider otherwise, and our audits run each prompt on three separate days.
