citations in AI answers about software prices pointed at the vendor's own site
When an AI engine tells you what software costs, check where the number came from: often the vendor's own page.
One monthly run of 40 answers; it measures where answers point, not whether they are right.
1 November 2026
1 November 2026 · monthly: new run on the 1st
We asked ChatGPT and Perplexity what 20 business tools cost, and logged every source they cited. In roughly three out of four citations, the answer points back at the vendor’s own pricing page — the page whose headline number is a marketing decision.
A pricing page is written to sell. The number in large type is often the annual rate, the promotional rate, or the rate for a plan many buyers will outgrow — and the conditions attached to it can sit in smaller type, in a footnote, or on a different tab. That is not deception; it is how pricing pages are commonly built.
The risk is what happens when an AI engine answers “what does this cost” by reading that page back to you: the headline number may survive the summary while the conditions do not. This study does not measure that — it records where the answers point, not whether they keep the conditions. When 77% of the sources behind those answers are the vendor’s own site, the vendor’s framing is the most likely starting point; a citation of the vendor’s site is not, on its own, evidence of a poor answer.
| Engine | Answers | Without a source | Share |
|---|---|---|---|
| ChatGPT | 20 | 0 | 0% |
| Perplexity | 1 | 0 | 0% |
An answer with no citation is not necessarily wrong, but it cannot be checked. In this edition every answer carried at least one source.
Correction, 2026-10-10: An earlier version of this page gave a Perplexity answer about SaneBox, with an empty source list and a wrong plan, price and trial length, as the example of an unsourced answer. That example came from a test run before our monthly editions began and is not in any published edition; in the 2026-10-01 edition Perplexity’s SaneBox answer cites ten sources and ChatGPT’s two. The example now comes from the edition itself. We also rewrote two sentences that presented lost pricing conditions as a finding: this study measures where answers point, not whether they keep the conditions.
Counting note: 77 citations is not 77 independent observations. They resolve to 76 distinct URLs on 37 domains, and a single vendor’s site plus its help centre can account for dozens of them. “Vendor’s own site” means the cited domain is the vendor’s registered domain in our database; a vendor help centre on a different domain is counted elsewhere unless we have checked it as the vendor’s. Domains that contain the tool’s name but are not its known domain are shown separately as “possibly the vendor”.
| Where the citation points | Citations | Share |
|---|---|---|
| Vendor's own site | 59 | 77% |
| Possibly the vendor (unverified domain) | 0 | 0% |
| Review or comparison site | 3 | 4% |
| Community or social | 0 | 0% |
| Other (not the vendor's own site) | 15 | 19% |
“Not the vendor’s own site” is not the same as independent: a comparison site may earn from the same vendors, and we did not assess editorial independence.
| Domain | Citations |
|---|---|
| airwallex.com | 10 |
| hubspot.com | 5 |
| whatconverts.com | 4 |
| beehiiv.com | 4 |
| kinsta.com | 4 |
| replit.com | 4 |
| g2.com | 3 |
| todoist.com | 3 |
| kit.com | 3 |
| sanebox.com | 3 |
| easydmarc.com | 2 |
| reclaim.ai | 2 |
The pattern holds across the top of the table: the vendor’s own domain, followed by the vendor’s own help centre. 37 distinct domains appear in total, but the distribution is heavily concentrated.
This is one run of 21 answers across 20 tools and two engines, in a US locale, on a single day. It is enough to show a pattern and not enough to quantify it precisely. We are publishing it because the direction is stark and the method is cheap to repeat — not because 21 answers settle anything.
We did not test whether the answers were correct. Measuring that at scale would mean re-verifying every quoted price by hand, which is the work our own corpus already does for 20 of these tools, but not on the engines’ answers.
And the result that is least flattering to us: this site was cited 0 times. We publish verified prices with reading dates, and none of that reached these answers. That is a finding about our own visibility, and it belongs in the report.
Queries of the form [tool] pricing for 20 business tools we research ourselves, submitted asynchronously to the Cloro API against ChatGPT and Perplexity, US locale, on 2026-11-01. Every response was stored with its full source list. Counting rule: a citation counts as “the vendor’s own site” when its registrable domain equals the vendor’s domain in our database, so a help centre on a subdomain of that domain counts as the vendor. The full rule and domain lists are in the study package.
Raw responses, including the full answer text and source lists: in the study package below, per monthly run. The underlying pricing corpus is published as an open dataset at /data/ under CC BY 4.0.
Download everything for edition 2026-11-01 (.zip)
Edition 2026-11-01. Every file below is the analysis input or output of this edition; the checksums let you confirm you have the same bytes we published. The manifest carries the population, period, definitions and counts. A rebuild of this page is not a new measurement: new rounds become new editions, and a changed edition carries a correction note.
| File | Size | SHA-256 |
|---|---|---|
| answers.json | 50,579 bytes | f6119d9979fc66cc… |
| citations.csv | 8,214 bytes | 5cc087c4ecc70625… |
| classification.json | 2,006 bytes | ff1d205b9a3c466a… |
| Edition | Status | Headline | Data |
|---|---|---|---|
| 2026-11-01 | current | 59 of 77: citations in AI answers about software prices pointed at the vendor's own site | manifest |
| 2026-10-01-published | published earlier | 34%: of citations pointed at the vendor's own page; 1 of 40 answers cited nothing | reconstructed from our published page |
| 2026-10-01 | superseded | 73 of 227: citations in AI answers about software prices pointed at the vendor's own site | manifest |
| 2026-09-02-published | published earlier | 46%: of citations pointed at the vendor's own page; 0 of 40 answers cited nothing | reconstructed from our published page |
| 2026-09-01 | superseded | 120 of 269: citations in AI answers about software prices pointed at the vendor's own site | manifest |
| 2026-08-19-published | published earlier | 79%: of citations pointed at the vendor's own page; 15 of 40 answers cited nothing | reconstructed from our published page |
Archived on Zenodo with a DOI: 10.5281/zenodo.23345626 (always resolves to the newest edition). This edition: 10.5281/zenodo.24074865. The Zenodo copy holds the same data files; screenshots of vendor software stay on this page only.
Licence: CC BY 4.0. Our own data, analysis, tables and charts. Quoted vendor text and screenshots of vendor pages remain the vendor's; we reproduce them as evidence. Questions, corrections or a vendor response: daniel@aibuildermarketplace.com · press & data.
AIBuilder Marketplace, “Who do AI engines cite for software pricing?”, Report 02, edition 2026-11-01. Free to quote and reproduce with attribution (CC BY 4.0). The raw response set is in the study package above.