AnswerPlume

AI-search citation snapshot (Nov 3–6, 2026)

A public dataset of 48 “best … 2026” buying queries and 240 citation rows (exactly 5 source links per query) collected by AnswerPlume.

This is a single-engine, single-run, single-location snapshot, not a ranking study, and not a comparison of ChatGPT, Perplexity, Gemini, or Google AI Overviews. Those engines were not tested.

Every number on this page was counted from the CSV columns below. Page titles and full URL paths were not stored, so they are not used here.

Download the CSVs

The four runs-*.csv files are the original collected tables (no cells edited). The combined file is the same 240 rows with query and first_brand_recommended filled down, plus run_date, vertical, and source_order.

License: CC BY 4.0. Attribution: AnswerPlume, “AI-search citation snapshot (Nov 3–6, 2026),” https://answerplume.venturesites.workers.dev/data/

Method

On four weekday mornings (America/Los_Angeles) we ran 12 buying questions per day through one AI web-search answer tool: the search tool built into our research agent. Each answer came with 5 source links. We recorded the cited domain, labeled who owns that domain (domain_type), and recorded the first brand the answer recommended.

Run dateApprox. local timeVerticalQueriesCitation rows
2026-11-03~9:00 AM PTb2b-software1260
2026-11-04~8:30 AM PTdtc-consumer1260
2026-11-05~8:10 AM PTlocal-services1260
2026-11-06~8:25 AM PTsaas-product1260

The 11-03 set is 11 software-buying questions plus one consumer question (best electrolyte drink mix brand 2026). Queries were written in the form “best [category] 2026” (local-service queries use “best [service] [city] 2026”). One run per query, one engine, one location.

domain_type is our classification, not a label from the engine. The vocabulary is not the same in every run, so “vendor” from 11-03 should not be added to “brand” from later runs without noticing the scheme change:

A domain may appear more than once in the same query. Those repeats are kept. The CSVs contain no personal data.

Key findings (counted from the CSVs)

Percentages are rounded to the nearest whole number. Fractions are exact. “Brand/vendor-owned” below uses each run’s own label: vendor + vendor (platform) on 11-03, and brand on 11-05 and 11-06. The 11-04 file has no such label.

Brand sites vs third-party sources

Vertical (run date)Brand/vendor-owned citationsAnswers with at least one brand/vendor source
Consumer DTC (2026-11-04)0 / 60 (0%)0 / 12
Local services (2026-11-05)17 / 60 (28%)9 / 12
SaaS products (2026-11-06)28 / 60 (47%)10 / 12
B2B software (2026-11-03, vendor + platform)32 / 60 (53%)10 / 12
All four runs77 / 240 (32%)29 / 48

The other 163 of 240 citations (68%) are third-party under those same labels (publishers, aggregators, independents, retailers, directories, agencies, nonprofits, and the remaining codes). 19 of 48 answers cited no brand/vendor-owned source, including all 12 consumer DTC answers.

On 11-03, setting aside the two queries with zero vendor citations (SOC 2 compliance automation, and electrolyte drink mix) leaves 32 of 50 (64%) vendor or platform citations. Being cited is not the same as being recommended: the files still show answers whose cited domains are not the first brand named.

Domains cited most often

Across all 240 rows there are 193 distinct cited domains. Ten domains appear three or more times:

Cited domainCitationsWhere they appear
forbes.com8DTC 5, local services 2, SaaS 1
nytimes.com5DTC 5
goodhousekeeping.com4DTC 4
exa.ai4Local services 4
shopify.com3B2B 2, SaaS 1
learn.g2.com3B2B 2, SaaS 1
wired.com3DTC 3
consumerreports.org3DTC 3
expertise.com3Local services 3
technologyadvice.com3SaaS 3

18 further domains appear twice; the remaining 165 appear once. By vertical:

Each of the 48 answers named a different first-recommended brand (48 distinct values of first_brand_recommended). That column is the brand named in the answer text, not a count of brand-owned citations.

Limits (read these before citing)

Write-ups and citation

Companion narrative posts (method plus discussion; not part of the raw table) are on the research page. A few page-title counts in those posts cannot be re-derived from these CSVs, because page titles were not stored in the collected files.

Suggested citation: AnswerPlume. 2026. AI-search citation snapshot (Nov 3–6, 2026). 48 queries, 240 citation rows, one engine. https://answerplume.venturesites.workers.dev/data/