Method. This is a single-engine, single re-run, not a ranking study and not a comparison of ChatGPT, Perplexity, Gemini, or Google AI Overviews. Those engines were not tested. On four weekday mornings, Nov 3–6, 2026 (America/Los_Angeles), we ran 48 “best … 2026” buying queries (12 per vertical) through one AI web-search answer tool — the search tool built into our research agent — and recorded the five source links on each answer. On Wed Nov 18, 2026, about 8:32–8:35 AM PT, we re-ran the same 48 query strings through the same tool. Each run produced 240 citation rows (48 × 5). Domains are the URL host with a leading www. stripped; other subdomains are kept. Repeated domains in the same query are kept and count separately. domain_type is our classification, not a label from the engine, and the vocabulary differs by vertical (see the method note). Every number below was counted from the public CSVs: runs-2026-11-03.csv, runs-2026-11-04-dtc.csv, runs-2026-11-05-local-services.csv, runs-2026-11-06-saas.csv, runs-2026-11-18-rerun.csv, and drift-2026-11-18.csv. The earlier posts on this site each cover one vertical’s original snapshot. This post asks a different question: which domains showed up most often across both runs, and which ones appeared or dropped out.
Findings (counts from the CSVs)
vendor where SaaS and local services use brand).Citation counts are raw row counts out of 240, including repeats in the same query. Combined rank is original + re-run.
| cited_domain | original citations (of 240) | re-run citations (of 240) | change | domain_type in the CSVs |
|---|---|---|---|---|
| forbes.com | 8 | 12 | 4 | publisher |
| nytimes.com | 5 | 6 | 1 | publisher |
| goodhousekeeping.com | 4 | 4 | 0 | publisher |
| blog.hubspot.com | 2 | 3 | 1 | brand, vendor |
| consumerreports.org | 3 | 2 | -1 | independent |
| expertise.com | 3 | 2 | -1 | aggregator |
| technologyadvice.com | 3 | 2 | -1 | aggregator |
| theknot.com | 2 | 3 | 1 | directory |
| bestinhood.com | 2 | 2 | 0 | aggregator |
| capterra.com | 2 | 2 | 0 | aggregator |
| exa.ai | 4 | 0 | -4 | aggregator |
| g2.com | 2 | 2 | 0 | aggregator |
Shown only where the absolute change is at least 2. A domain can change by 2 from two citations in one query (for example theknot.com was cited twice, then three times, on the same wedding-photographer query — that +1 change is not in this table). learn.g2.com was labelled independent on the original B2B rows and aggregator on the SaaS rows; both labels from the CSVs are listed.
| cited_domain | original citations | re-run citations | change | domain_type in the CSVs |
|---|---|---|---|---|
| forbes.com | 8 | 12 | 4 | publisher |
| exa.ai | 4 | 0 | -4 | aggregator |
| wired.com | 3 | 0 | -3 | publisher |
| learn.g2.com | 3 | 1 | -2 | aggregator, independent |
| shopify.com | 3 | 1 | -2 | brand, vendor (platform) |
| aarp.org | 2 | 0 | -2 | publisher |
| ecommerce-gold.com | 2 | 0 | -2 | independent |
| elliottbayah.com | 0 | 2 | 2 | brand |
| getjobber.com | 0 | 2 | 2 | vendor |
| local.yahoo.com | 0 | 2 | 2 | aggregator |
| mailercloud.com | 2 | 0 | -2 | brand, vendor |
| meetingnotes.com | 0 | 2 | 2 | independent |
| outfy.com | 0 | 2 | 2 | vendor |
| pango.ai | 2 | 0 | -2 | vendor |
| stackscout.net | 2 | 0 | -2 | independent |
| teamwork.com | 0 | 2 | 2 | brand |
| tsbpe.texas.gov | 0 | 2 | 2 | other |
| uk.pcmag.com | 0 | 2 | 2 | publisher |
| womenshealthmag.com | 0 | 2 | 2 | publisher |
Each vertical is 12 queries and 60 citation rows per run. Unique-domain counts treat a domain once per vertical even if it was cited more than once.
| vertical | unique domains original / re-run | in both runs | only original | only re-run | original domain_type mix | re-run domain_type mix |
|---|---|---|---|---|---|---|
| B2B software (original Nov 3) | 55 / 54 | 20 | 35 | 34 | vendor 30/60, independent 16/60, publisher 6/60, agency 5/60, vendor (platform) 2/60, marketplace listing 1/60 | vendor 32/60, independent 18/60, publisher 5/60, agency 4/60, vendor (platform) 1/60 |
| Consumer DTC (original Nov 4) | 40 / 38 | 25 | 15 | 13 | publisher 37/60, independent 16/60, nonprofit 3/60, retailer 3/60, professional 1/60 | publisher 39/60, independent 14/60, professional 3/60, nonprofit 2/60, retailer 2/60 |
| Local services (original Nov 5) | 50 / 52 | 25 | 25 | 27 | aggregator 26/60, brand 17/60, publisher 7/60, directory 6/60, other 3/60, nonprofit 1/60 | brand 29/60, aggregator 19/60, publisher 4/60, other 4/60, directory 4/60 |
| SaaS products (original Nov 6) | 56 / 52 | 24 | 32 | 28 | brand 28/60, aggregator 26/60, other 3/60, publisher 3/60 | brand 30/60, aggregator 22/60, publisher 6/60, other 2/60 |
Most-cited domains in each vertical, original then re-run. Where several domains share a count, all of them are listed:
Limits: one engine, one original snapshot, one re-run, 48 queries, with our own domain_type labels. A third run, a different engine, or a different query list could look different. Mean per-query domain overlap (2.10 distinct shared domains out of 5) and first-named-brand changes (23 of 48) are in the citation-drift section and the method note; this page does not repeat those tables.
Raw data: dataset page (CSVs, SHA-256 hashes, method note). If you want the same one-engine method run on up to three of your own queries, request a free one-page snapshot.