Do Google Results and AI Answers Cite the Same Domains? An Exploratory Pilot on URL and Domain Overlap for B2B Software Buying Prompts

Buyers shortlist business software with Google and with AI answer engines, and sellers ask whether a place in one goes with a place in the other. Published overlap figures measure different units and rarely separate buyer intent. This exploratory pilot describes, for B2B software buying prompts, how far the sources cited in AI answers coincide with Google's top ten organic results at URL and domain level, by buyer intent, and how stable each measure is across repeated runs. Under a protocol frozen by SHA-256 file hash before collection, we ran 60 prompts (6 categories, 5 intents) three times each on one date (1 November 2026) through the DataForSEO ChatGPT and Gemini scrapers and Google results pages. Overlap is the Jaccard index against the union of three same-day Google passes (two for five prompts). For the verbatim prompt, domain-level mean Jaccard overlap was 0.03 to 0.11 by intent for both scrapers (0.06 to 0.15 against a keyword-form Google reference). URL-level overlap was near zero but rests on the 427 of 1,532 Google results that exposed a destination address. The Google passes agreed only partly with each other (mean pairwise Jaccard 0.339 at domain level, 0.062 at URL level), and repeat-run variation was a sizeable fraction of cell means. Trust-validation answers were the most concentrated and leaned on review and marketplace domains, as classed by a language model and not checked by a person. An exploratory recovery of 130 Google results with no resolvable domain changed cell means by at most 0.006, while a stress case that ignores the recovery reached 0.24. Separate code reproduced all 720 run-level overlap values from the raw responses. The pilot has six categories, one date and scrapers rather than consumer products. It supports no general claim and does not show whether a Google rank predicts an AI answer's sources. The author's company sells search and AI-visibility services; language models assisted with code, data audit and drafting (section 12).

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-13
DOI
https://doi.org/10.5281/zenodo.24190544
Primary Topic
Information Retrieval and Search Behavior
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Do Google Results and AI Answers Cite the Same Domains? An Exploratory Pilot on URL and Domain Overlap for B2B Software Buying Prompts

Ankur Shrestha
3 links
Zenodo (CERN European Organization for Nuclear Research)
Information Retrieval and Search Behavior
preprint

Do Google Results and AI Answers Cite the Same Domains? An Exploratory Pilot on URL and Domain Overlap for B2B Software Buying Prompts

Ankur Shrestha
preprint en

Abstract

Buyers shortlist business software with Google and with AI answer engines, and sellers ask whether a place in one goes with a place in the other. Published overlap figures measure different units and rarely separate buyer intent. This exploratory pilot describes, for B2B software buying prompts, how far the sources cited in AI answers coincide with Google's top ten organic results at URL and domain level, by buyer intent, and how stable each measure is across repeated runs. Under a protocol frozen by SHA-256 file hash before collection, we ran 60 prompts (6 categories, 5 intents) three times each on one date (1 November 2026) through the DataForSEO ChatGPT and Gemini scrapers and Google results pages. Overlap is the Jaccard index against the union of three same-day Google passes (two for five prompts). For the verbatim prompt, domain-level mean Jaccard overlap was 0.03 to 0.11 by intent for both scrapers (0.06 to 0.15 against a keyword-form Google reference). URL-level overlap was near zero but rests on the 427 of 1,532 Google results that exposed a destination address. The Google passes agreed only partly with each other (mean pairwise Jaccard 0.339 at domain level, 0.062 at URL level), and repeat-run variation was a sizeable fraction of cell means. Trust-validation answers were the most concentrated and leaned on review and marketplace domains, as classed by a language model and not checked by a person. An exploratory recovery of 130 Google results with no resolvable domain changed cell means by at most 0.006, while a stress case that ignores the recovery reached 0.24. Separate code reproduced all 720 run-level overlap values from the raw responses. The pilot has six categories, one date and scrapers rather than consumer products. It supports no general claim and does not show whether a Google rank predicts an AI answer's sources. The author's company sells search and AI-visibility services; language models assisted with code, data audit and drafting (section 12).

Zenodo (CERN European Organization for Nuclear Research)
Information Retrieval and Search Behavior
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon