How accurate are the prices AI quotes for brands?

We read 100 randomly drawn checkable quotes by hand, with the crawled page text and the brand's current pricing. 79 were real price quotes. The other 21 were dollar figures that are not a brand price, such as a futures margin, an account minimum, or a cost estimate for another product.

We counted a quote within 10% of the listed price as right, as the automated check does. Of the 79 real price quotes, 65 were right (82.3%), 8 were wrong (10.1%), and 6 could not be settled (7.6%). Four of the 8 wrong quotes gave an older price, two gave a price below the cheapest listed plan, and two mixed up a per-GB and a per-TB price.

The automated check rarely called a wrong price right: 54 of the 56 real quotes it matched were confirmed. It was much less reliable on mismatches. Of the 23 real quotes it flagged, 8 were wrong, 11 were right prices that sat on another page or were not readable in the crawled text, and 4 could not be settled.

Searchable's brand-graded study put AI's error rate on pricing facts at 13.5%. Our hand check lands close to that, at 10.1%. A plain automated match against pricing pages would have suggested 37%, which overstates the AI pricing error by more than three times.

Across 358,902 brand price quotes, 7,594 were checkable, or 2.1%. The other 351,308 (97.9%) could not be checked. For 92.8% of all quotes, we had no crawled pricing page on the brand's own website. Another 2.9% named a brand whose crawled pricing page did not list at least three prices in its text, and 2.2% named a brand whose pricing page was last crawled more than 60 days from the answer.

Most quotes are product prices in shopping answers: 74.3% had no billing unit or price word next to the amount, such as "$29.99" after a product name. Those products have no pricing page to check. The checkable brands are mostly software, phone plans and financial services, such as Trello, Verizon and Charles Schwab.

From September 22 to October 6, 2026, 22,900 of 91,525 ChatGPT Search answers contained a dollar figure, or 25.0%. Over the full window from May 24, the rate was 26.5%. Google AI Mode put a dollar figure in 13.5% of its answers.

Not every dollar figure is a brand price. A brand price quote appeared in 8.3% of ChatGPT Search answers and 4.8% of Google AI Mode answers. The older engines, measured from March 9 to 29, 2026, show the same order: ChatGPT quoted a brand price in 11.0% of answers and Google AI Overviews in 5.8%.

We counted the share of answers in each market that quoted at least one brand price, on ChatGPT Search and Google AI Mode, for the 413 markets with at least 500 answers. Across those markets the share was 6.9%.

The top markets are consumer shopping categories. Occasion Dresses & Formalwear and Online Eyewear and Vision Care both quoted a brand price in 43.3% of answers, and Home Coffee Makers and Grinders in 42.3%. In 118 markets, fewer than 1% of answers quoted a brand price. Most of those are business and enterprise software markets, such as Contract Lifecycle Management (CLM) Software.

ChatGPT Search matched the brand's own pricing page on 66.5% of 4,479 checkable quotes. ChatGPT matched 61.9% of 409, Google AI Mode 57.4% of 1,804, and Google AI Overviews 55.3% of 902.

ChatGPT Search had a higher price-match rate than ChatGPT, although the two were measured in different months. Part of that lead comes from citations: ChatGPT Search cited the brand's own pricing page far more often than Google AI Mode, as the next section shows.

When the same answer cited the brand's own pricing page, 72.7% of 2,463 checkable quotes matched. When it did not, 58.0% of 5,131 matched. The gap held on both current engines: 72.3% vs 60.4% on ChatGPT Search, and 78.9% vs 56.3% on Google AI Mode.

Of the checkable quotes, ChatGPT Search cited the brand’s pricing page 2,304 times; Google AI Mode did so 90 times. Quotes backed by a pricing page were more likely to match.

When the pricing page was crawled within 7 days of the answer, 68.7% of 3,267 checkable quotes matched. At 8 to 30 days it was 60.5% of 2,709, and at 31 to 60 days 54.6% of 1,618.

Prices change, so a check is only as good as the date of the page it uses. Some of the mismatches at 31 to 60 days are price changes between the answer and the crawl, not AI errors.

A hedged price comes with a qualifier: an approximation ("around $20", "~$20"), a range ("$20 to $40"), a floor ("starting at $20") or a cap ("up to $20"). Across all 358,902 brand price quotes, 37.8% were hedged.

Google AI Mode hedged 74.7% of its quotes and Google AI Overviews 66.9%. ChatGPT Search hedged 30.8% and ChatGPT 8.8%.

Hedged quotes matched less often. Plain prices matched 68.1% of the time (3,922 checkable quotes). Approximate prices matched 59.7% (1,575), "starting at" prices 57.4% (1,012) and ranges 53.0% (1,022). A hedged price was less often an exact listed price.

Of the 2,828 checkable quotes that did not match, 72.1% were amounts that appeared nowhere on the crawled page. 16.6% equaled 12 times a listed price or one twelfth of it, which is a monthly vs annual mix-up. 6.9% named a plan whose listed price was different, and 4.4% matched a price the page showed in an earlier crawl.

Only 9.5% of all brand price quotes stated a billing period, so a reader often cannot tell whether "$240" is per month or per year. The hand check found the same pattern in the real errors: older prices, prices below the cheapest plan, and unit mix-ups.

On ChatGPT Search, 63.3% of 311 checkable quotes matched under gpt-5-5 (May 24 to August 7) and 66.7% of 4,168 under gpt-5-6 (from August 8). That gap comes from a change in which brands were quoted. On the 45 brands checkable under both versions, the match rate was 67.3% under gpt-5-5 and 66.8% under gpt-5-6.

Google AI Mode shows the same flat result on its own 68-brand panel: 56.9% before August 8 and 59.4% after. ChatGPT on gpt-5-3, measured in March 2026, matched 61.9% of 409 checkable quotes.

Among the 65 brands with at least 30 checkable quotes on at least 5 prompts, the median brand matched 74.0% of the time. Three brands matched on every checkable quote: BambooHR, Plutio and AppSheet. Retool matched 99.0% of 101 quotes, Trello 92.1% of 242 and Datadog 93.8% of 210.

The crawled pricing pages of these brands listed their plan prices as text, and the AI quotes repeated those prices. Brand pricing in AI answers is most reliable when the pricing page states it plainly.

At the other end, fewer than 25% of checkable quotes matched for ten brands, including Xero (4.3% of 163), HubSpot (19.7% of 346) and PagerDuty (3.6% of 56).

This is mostly a page problem, not an AI problem. The Xero quotes we read gave US plan prices of $27, $59 and $97 a month after an October 1, 2026 change, and $25, $55 and $90 before it. Xero's live US pricing page lists $27, $59 and $97 on October 6, 2026. The crawled Xero page text did not list those plan prices. The same was true of the RunPod, HubSpot and Buildium pages we inspected: their plan prices were not in the crawled text.

We started with every AI answer with a dollar figure to an organic prompt: 344,276 answers out of 1,662,814. ChatGPT and Google AI Overviews were measured from March 9 to 29, 2026, and ChatGPT Search and Google AI Mode from May 24 to October 6, 2026. We split each answer into sentences and table rows and kept a dollar amount only when a detected brand was named before it in the same sentence or row, or was the only brand named just after it. We dropped amounts in thousands, millions or billions, non-US dollar amounts, zero amounts, and amounts next to words like funding, minimum, margin, deposit or salary.

A brand's own pricing page is a page at /pricing, /plans, /price or /prices on the website that is the brand's main address in the Parse index. We only check quotes where that page was crawled within 60 days of the answer and its text listed at least three prices between $1 and $10,000. Most uncheckable quotes, 92.8% of all quotes, named a brand with no crawled pricing page on its own website. The 60-day window and the three-price floor removed another 5.0%.

We computed the headline two ways. A separate in-database rebuild, which reads each brand's stored mention sentence instead of our own sentence split, gave 60.5% matched on 6,160 checkable quotes, against 62.8% here. A literal text search of the crawled page for 2,000 sampled quotes agreed with the automated exact-match verdict 95.0% of the time. The 100-case hand check is in the second section. These figures are an observed cut of Parse data over the stated windows, not a live measure.

Put your prices in the text of one pricing page. Every brand that AI quoted right more than 9 times in 10 had plan prices readable as text on its pricing page. Where the crawled page text did not show the prices, AI still quoted prices, often correct ones, but they could not be checked against the brand's own page.

State the billing period next to every price, and say which plan each price belongs to. Monthly vs annual mix-ups were 1 in 6 mismatches, and only 1 in 10 quotes stated a period.

Check your own brand pricing in AI answers monthly, against a fresh crawl of your page. Match rates fell from 69% to 55% as the crawl aged, and 4 of the 8 real errors in our hand check were older prices. When you change prices, expect AI answers to lag.