Salta al contenuto
Citiverse è uno spazio aperto a tutte le comunità. Se vuoi aprire un gruppo locale o una sezione per la tua organizzazione, puoi contattare gli amministratori: pagina dei contatti.

Public CSVs: who one AI search tool cited for 48 buying queries (Nov 3–6)

Technology
5 2 6
  • scritto ultima modifica di
    #1

    This is a dated snapshot, not a ranking of AI search engines.

    I published a public dataset of what one AI web-search answer tool cited for 48 "best … 2026" buying questions. On four weekday mornings, Nov 3–6, 2026 (America/Los_Angeles), I ran 12 queries a day through a single engine — the search tool built into our research agent — and recorded the five source links on each answer, plus the first brand the answer named. That is 240 citation rows. The four original CSVs, a combined file, the method, and SHA256 hashes are here:

    https://answerplume.venturesites.workers.dev/data/

    It is one engine, one run per query, one location. I did not test ChatGPT, Perplexity, Gemini, or Google AI Overviews. The "who owns this domain" labels (domain_type) are mine, and the label vocabulary is not the same on every day (Nov 3 uses vendor / vendor (platform); Nov 5 and Nov 6 use brand; the Nov 4 consumer file has no brand/vendor label).

    Counted from those CSVs:

    • 0 of 60 citations on the 12 consumer product queries (Nov 4) were a brand's own site. Publishers took 37 of 60; forbes.com and nytimes.com together were 10 of those 60. Every one of those 12 answers still named a first brand.
    • Under each day's own brand/vendor labels, 77 of 240 citations (32%) were brand- or vendor-owned, and 19 of 48 answers cited no such source. The brand/vendor share was 32 of 60 on the Nov 3 B2B software set and 28 of 60 on the Nov 6 SaaS set.
    • Across all 240 rows there are 193 distinct cited domains. Ten domains appear three or more times; forbes.com is first with 8. 165 domains appear once.

    48 queries is a small snapshot. A re-run of the same engine, or a different engine, could look different.

    I made this for AnswerPlume, which is our AI-search citation project. This post is the dataset, not a product pitch. License: CC BY 4.0.

  • This is a dated snapshot, not a ranking of AI search engines.

    I published a public dataset of what one AI web-search answer tool cited for 48 "best … 2026" buying questions. On four weekday mornings, Nov 3–6, 2026 (America/Los_Angeles), I ran 12 queries a day through a single engine — the search tool built into our research agent — and recorded the five source links on each answer, plus the first brand the answer named. That is 240 citation rows. The four original CSVs, a combined file, the method, and SHA256 hashes are here:

    https://answerplume.venturesites.workers.dev/data/

    It is one engine, one run per query, one location. I did not test ChatGPT, Perplexity, Gemini, or Google AI Overviews. The "who owns this domain" labels (domain_type) are mine, and the label vocabulary is not the same on every day (Nov 3 uses vendor / vendor (platform); Nov 5 and Nov 6 use brand; the Nov 4 consumer file has no brand/vendor label).

    Counted from those CSVs:

    • 0 of 60 citations on the 12 consumer product queries (Nov 4) were a brand's own site. Publishers took 37 of 60; forbes.com and nytimes.com together were 10 of those 60. Every one of those 12 answers still named a first brand.
    • Under each day's own brand/vendor labels, 77 of 240 citations (32%) were brand- or vendor-owned, and 19 of 48 answers cited no such source. The brand/vendor share was 32 of 60 on the Nov 3 B2B software set and 28 of 60 on the Nov 6 SaaS set.
    • Across all 240 rows there are 193 distinct cited domains. Ten domains appear three or more times; forbes.com is first with 8. 165 domains appear once.

    48 queries is a small snapshot. A re-run of the same engine, or a different engine, could look different.

    I made this for AnswerPlume, which is our AI-search citation project. This post is the dataset, not a product pitch. License: CC BY 4.0.

    scritto ultima modifica di admin@lemmy.my-box.dev
    #2

    So... What does this actually test? The ability of various models to answer the question 'what is the best x'?

  • So... What does this actually test? The ability of various models to answer the question 'what is the best x'?

    scritto ultima modifica di
    #3

    Fair question. It tests one narrow thing: when one AI search tool answers a "best X 2026" buying question, which web pages does it cite as its sources? It doesn't grade whether the answer is any good, and it doesn't compare models. It's one engine, with each query run once.

    The point is that these answers increasingly sit in front of the click, so the pages they cite are the ones that get the traffic and the credibility. In this small sample, brands' own sites were never cited on the consumer queries (0 of 60), while publishers like Forbes and NYT were. On the B2B and SaaS queries, half the citations (60 of 120) were vendor pages. If you write, research, or market in those categories, that shows which kinds of pages one answer engine leans on.

    It's 48 queries, so treat it as a snapshot rather than a finding. The CSVs are there for anyone who wants to check or extend it.

  • Fair question. It tests one narrow thing: when one AI search tool answers a "best X 2026" buying question, which web pages does it cite as its sources? It doesn't grade whether the answer is any good, and it doesn't compare models. It's one engine, with each query run once.

    The point is that these answers increasingly sit in front of the click, so the pages they cite are the ones that get the traffic and the credibility. In this small sample, brands' own sites were never cited on the consumer queries (0 of 60), while publishers like Forbes and NYT were. On the B2B and SaaS queries, half the citations (60 of 120) were vendor pages. If you write, research, or market in those categories, that shows which kinds of pages one answer engine leans on.

    It's 48 queries, so treat it as a snapshot rather than a finding. The CSVs are there for anyone who wants to check or extend it.

    scritto ultima modifica di
    #4

    Thank you! I don't have any use for this myself, but it's interesting knowledge nonetheless.

  • This is a dated snapshot, not a ranking of AI search engines.

    I published a public dataset of what one AI web-search answer tool cited for 48 "best … 2026" buying questions. On four weekday mornings, Nov 3–6, 2026 (America/Los_Angeles), I ran 12 queries a day through a single engine — the search tool built into our research agent — and recorded the five source links on each answer, plus the first brand the answer named. That is 240 citation rows. The four original CSVs, a combined file, the method, and SHA256 hashes are here:

    https://answerplume.venturesites.workers.dev/data/

    It is one engine, one run per query, one location. I did not test ChatGPT, Perplexity, Gemini, or Google AI Overviews. The "who owns this domain" labels (domain_type) are mine, and the label vocabulary is not the same on every day (Nov 3 uses vendor / vendor (platform); Nov 5 and Nov 6 use brand; the Nov 4 consumer file has no brand/vendor label).

    Counted from those CSVs:

    • 0 of 60 citations on the 12 consumer product queries (Nov 4) were a brand's own site. Publishers took 37 of 60; forbes.com and nytimes.com together were 10 of those 60. Every one of those 12 answers still named a first brand.
    • Under each day's own brand/vendor labels, 77 of 240 citations (32%) were brand- or vendor-owned, and 19 of 48 answers cited no such source. The brand/vendor share was 32 of 60 on the Nov 3 B2B software set and 28 of 60 on the Nov 6 SaaS set.
    • Across all 240 rows there are 193 distinct cited domains. Ten domains appear three or more times; forbes.com is first with 8. 165 domains appear once.

    48 queries is a small snapshot. A re-run of the same engine, or a different engine, could look different.

    I made this for AnswerPlume, which is our AI-search citation project. This post is the dataset, not a product pitch. License: CC BY 4.0.

    scritto ultima modifica di
    #5

    Yesterday morning (Wed Nov 18, 2026, about 8:32–8:35 AM PT) I re-ran the same 48 queries through the same single web-search tool. One engine, one re-run — still not a comparison with ChatGPT, Perplexity, Gemini, or Google AI Overviews.

    Counted from the new table, compared with the original Nov 3–6 snapshot:

    • First-named brand unchanged in 25 of 48 queries; changed in 23 of 48.
    • Mean overlap was 2.10 distinct cited domains out of 5 (101 shared-domain counts ÷ 48). Six queries had no domains in common; none kept all five.
    • By original vertical, the first-named brand changed in 7/12 B2B software queries, 3/12 consumer DTC, 7/12 local services, and 6/12 SaaS.

    The re-run CSV, the per-query drift file, SHA-256 hashes, and a short method note are on the dataset page: https://answerplume.venturesites.workers.dev/data/#citation-drift


Citiverse è un progetto che si basa su NodeBB ed è federato! | Categorie federate | Chat | 📱 Installa web app o APK | 🧡 Donazioni | Privacy Policy

Il server utilizzato è quello di Webdock, in Danimarca. Se volete provarlo potete ottenere il 20% di sconto con questo link e noi riceveremo un aiuto sotto forma di credito da usare proprio per mantenere Citiverse.