ChatGPT Priced All 10 Vendors With No Website Price When Asked by Name

When asked about them by name, ChatGPT put a price on all 10 AI customer service vendors we tested that show no price on their own websites. On shortlists, it almost never priced them, while rivals' published prices showed up. Sales-led companies keep prices off the website so the first number comes from a sales call. In ChatGPT, that number comes from somewhere else. The real choice is whose number it is.

The short version

• Named, they got a price anyway: ChatGPT priced 10 vendors with no price on their own websites in about 4 in 5 direct answers (38 of 49).
• On shortlists, they rarely got a price: When the request named a budget, vendors with no price on their websites got a figure in under 1 in 10 shortlist mentions (7 of 86), against 6 in 10 for vendors that publish (232 of 386).
• Their own listings were named as the source: All 5 with a marketplace price had that listing named as the source of ChatGPT's first figure (14 of their 20 priced answers).
• Published prices came back right: For 8 of the 10 vendors that publish, at least 97% of ChatGPT's checkable figures matched the pricing page.
• A budget brought prices in: With a budget, 9 in 10 shortlist answers stated a price, against 4 in 10 without (91 and 39 of 100).

What to do with this

1. If the price is hidden, treat marketplace listings as the public price. Each listing under the brand, on AWS Marketplace, Salesforce AppExchange, the SAP Store, or Shopify, should state the unit, what's included, and the term. Remove placeholder figures.
2. If the price is hidden, ask ChatGPT what the product costs and log where each number comes from. Use five questions: cost, unit price, price against a named rival, entry price, and whether pricing is public. Ask each with and without a budget. Where the figure comes from a third-party estimate, publish a starting price, so ChatGPT has a first-party number to quote.
3. If the price is published, show one current price per plan. The two vendors below 97% showed a second number on the same page. Where an offer price sits beside a regular one, say in text which applies to whom.

ChatGPT priced these vendors in 38 of 49 answers about them, but rarely on shortlists

We asked five price questions about each of 10 AI customer service vendors that show no plan price on their own websites, including Decagon, Sierra, Ada, and Cognigy. Every vendor got a figure in at least 3 of its answers.

For example, Cognigy has no pricing page, yet ChatGPT gave its entry price as "approximately $43,080 per year," the Basic package on its own AWS Marketplace listing. Ada shows the shortlist gap on its own: a figure in 4 of 5 direct answers, 3 of 33 budget-shortlist mentions, and 0 of 50 without a budget.

Our read: We think ChatGPT fills the gap when someone asks about one vendor, and leaves it empty when it compares several. The hidden price surfaces where it helps least, alone, without the rivals' prices beside it.

A day later, the same direct questions drew a figure in 43 of 50 answers, and 29 got the same first figure on both days.

This is one niche on one engine over two days. Shortlist mentions were mostly Ada, Decagon, and Sierra, and 8 of the 10 vendors without a price are enterprise AI agents, so the gap with vendors that publish is a pattern, not proof.

All 5 vendors with a marketplace price had that listing named as the source

Five of the 10 show no price on their own website but, by ChatGPT's account, list one on a software marketplace: Ada, Cognigy, Maven AGI, Parloa, and Gladly. For each, ChatGPT named that listing as the source of its first figure in at least 2 direct answers. ChatGPT said Ada's AWS and AppExchange listings show $30,000 to $35,000 a year.

The vendors' own pricing pages rarely backed these figures. ChatGPT cited them in 10 of 38 priced direct answers, all on pages that show no price, against 49 of 49 for vendors that publish.

Our read: We think listings win because they are often the one vendor-made number ChatGPT can find. That includes careless ones. ChatGPT said Sierra's AWS listing shows a $1 million figure, which it called a "private-offer pricing placeholder."

The 4 vendors with no marketplace listing (Decagon, Forethought, Kore.ai, and Kustomer) got third-party estimates, procurement data, or, for Kustomer, an older copy of its own prices.

All 4 priced Kore.ai answers gave $0.20 per session, a rate that may sit in Kore.ai's own developer documentation; we did not confirm it. Third-party estimates often came with no source named, so only the marketplace share is firm.

For 8 of the 10 vendors that publish, at least 97% of figures matched the pricing page

Of 865 checkable figures, 5 were wrong or out of date, spread over Freshdesk, Gorgias, and Agentforce. The two vendors below 97%, Intercom and Agentforce, fell short because ChatGPT quoted the other number their pages carried.

On the test day, Intercom's page showed its Essential plan at $19 a seat, and that evening it showed $29 struck through beside $19. ChatGPT gave $29 in 59 answers and $19 in 1.

Our read: We think the misses start on the vendor's page, not in the model. A price the page shows two ways comes back as either. Salesforce's Agentforce rate sat in the page code, not on the visible card, and ChatGPT quoted it.

We loaded the pages from Europe, so we can't tell which version the US answers saw.

A budget put a price in 91 of 100 shortlist answers, up from 39

We wrote 100 shortlist requests in two versions, identical except for one budget sentence. One came from a 15-agent support team at a project management SaaS, with about 8,000 conversations a month.

Without a budget, ChatGPT recommended five vendors and gave one price. With "Our budget for AI is about $4,000 a month," it priced Intercom, Zendesk, and Freshdesk and dropped Sierra and Decagon. It kept Ada but wouldn't "assume it will fit a $4,000 monthly budget without a direct quote."

Our read: We think a budget turns a shortlist into a pricing question. In 52 of the 100 requests, only the version with a budget stated a price, and in none did only the version without one.

The shortlists moved less. A budget cut contact-sales vendors' recommendation rate by 2.8 points and lifted publishing vendors' by 0.8, a gap small enough to be chance.

The budget version always ran first, and each prompt ran once.

Methodology

Question. When a SaaS vendor shows no price on its own website, what price does ChatGPT give, and where does it come from?

Data. 300 prompts about 20 AI customer service vendors ran once on ChatGPT through Atomic AGI's tracking (US run location, single turn) on October 20, 2026; 299 completed. On October 21, 100 ran again: the 50 direct questions about vendors without a price and 25 shortlist pairs.

Sample. 10 vendors state a price on their pricing page and 10 show none (pages captured October 20, 2026); 5 of the 10 list a price on a software marketplace. 100 direct questions (5 per vendor) and 100 shortlist requests, each with and without a budget sentence. Budgets were tight (27 pairs, median $500 a month), moderate (45, median $4,000 a month), or generous (28, median $66,000 a month).

Coding. Each answer was coded with AI-assisted reading against written rules. A second reading of 40 random answers met every preset threshold. A figure matches when it is within 10% of a list price in the same unit; 155 of the 1,020 figures for vendors that publish had no list price in that unit. The source of each first figure is the one the answer named or, where none was named, the coder's reading of the answer. A second, blind coding of 36 of the 38 priced answers agreed on the source class in 28 (kappa 0.71) and on all 13 marketplace calls; the other 2 were checked by hand. Figures matching another version of the page (76, all at Intercom and Agentforce) were sorted into that group after the matching rule was set.

Analysis. Vendors are the independent unit for any comparison between the two groups; the recommendation gap was tested by shuffling the labels across the 20 vendors (p 0.19). The one planned budget-level check, generous budgets only, went the other way (contact-sales vendors +4.3 points, publishing vendors -1.8; p 0.14), also no clear difference.

Limitations. One engine, one niche, one run location, two days, and model settings not recorded. A cited page doesn't prove where a figure came from. Only Cognigy's marketplace listing was captured; the others are as ChatGPT described them. Atomic AGI, which ran the prompts, sells AI search tracking.

Reproducibility. Aggregates are in each chart's data table. Prompt wording and answer text stay internal, apart from the quotes above.

Sources

1. AWS Marketplace, "NiCE Cognigy: AI Agents for Enterprise Contact Centers," pricing section, captured October 20, 2026. https://aws.amazon.com/marketplace/pp/prodview-6c25zypcl2fro

Omnius articles this note builds on

• SaaS Pricing Transparency: ChatGPT Puts a Price on Contact Sales