Singapore GEO Citation Probes, September 2026 (ChatGPT and Google AI Mode)

This dataset records which web pages two AI answer engines cited when asked buyer questions from five Singapore industries: dental and aesthetics, family law, tuition, B2B SaaS and e-commerce. It was collected while writing the GEO Playbook v1.0 (https://doi.org/10.5281/zenodo.23005022). Mirrors: https://github.com/Canlah-AI/ai-citations-sg-2026 · https://huggingface.co/datasets/CanlahAI/ai-citations-sg-2026 · https://canlah.ai/data/ai-citations-sg-2026/ Collection. Both rounds were run on 2026-09-23 (UTC+8). ChatGPT was queried through the OpenAI Responses API (gpt-5.5 with the web search tool, approximate user location Singapore, no system prompt). Google AI Mode was queried through a third-party SERP API (gl=sg, hl=en). Round 1: 60 questions (12 per industry, 6 in Chinese) about buying decisions, run on both engines: 526 citations, 307 from ChatGPT and 219 from AI Mode (474 distinct URLs). Round 2: 50 questions (10 per industry, 5 in Chinese), one per question family (definition, procedure, statistic, calculation, nearby, single-business review, complete guide, rules, checklist), run on ChatGPT only because the SERP API quota was used up: 241 citations (225 distinct URLs). A citation is one distinct URL within one answer. For 30 round-2 questions a second, independent ChatGPT call is included as supplementary data. Files. queries.csv (110 questions with our intent labels); citations.csv (917 rows: cited URL, domain, order in the answer, engine, round, run, rule-based site type, forum/social flag, and for dental a manual page-type label); chatgpt_answers.jsonl (141 ChatGPT answers; official runs stored only the first 600 characters, the supplementary re-runs are full text); page_annotations.csv (our annotations of 92 cited pages from round 1 and 49 sample pages from round 2: tables, FAQ, estimated position of the copied sentence); page_types.csv (46 cited-page types with family and evidence tier); stats.md (recount and findings with denominators); README.md (datasheet). Selected findings. In round 1, 0 of 307 ChatGPT citations were forum or social sites, against 30 of 219 AI Mode citations and 114 of 556 Google organic top-10 results for the same questions. For the same question, the two engines shared 19 of 507 cited URLs. Only 22 of 307 ChatGPT citations were in the same question's Google top 10. When 30 round-2 questions were asked twice, the two runs shared 53 of the 222 URLs they cited between them. Not included. Google AI Mode answer text, Google organic result lists (only aggregates are reported) and copies of third-party pages (only URLs and our annotations). Limits. Each question was sampled once in the official runs; round 2 has one engine only; positions of copied sentences were estimated by eye; samples are small; single market (Singapore); results come from APIs, not consumer interfaces; site types are rule-based. The page annotations were drafted by an AI multi-agent workflow and were not re-checked line by line. 中文摘要:新加坡五个行业(牙科医美、家事法、补习、B2B 软件、电商)的买家问句在 ChatGPT 与 Google AI Mode 上的引用记录。第一轮 60 句、两个引擎、526 条引用;第二轮 50 句、只测 ChatGPT、241 条引用。含问句、每条引用的网址与顺序、ChatGPT 答案、92 个被引网页的标注与 46 种页型。不含 Google AI Mode 答案原文、Google 排名原始列表和第三方网页原件。数据说明书是 README.md(中文)。