# What AI answer engines cite when Malaysian SMEs ask for marketing help: data package

Version 1.0. The paper and these files are at https://studio.aliqgroup.com/en/research/ai-answer-citations-malaysia-2026/

## Contents

- `README.md`: this file.
- `trials.csv`: the anonymised trial-level dataset (described below).
- `protocol.md`: the study protocol as frozen on 26 September 2026, with the log of every later change.
- `codebook.md`: the coding rules for source type, publisher country and page type.
- `tables.md`: the tables the paper draws on, generated from the run logs and the coding sheets.
- `prompts.json`: the question panel, version 1.0.
- `literature-search.md`: the searches behind the paper's statement about earlier work.
- The paper itself, as a PDF.

## Licence

The paper and this package are published under the Creative Commons Attribution 4.0 International licence (CC BY 4.0, https://creativecommons.org/licenses/by/4.0/). You may copy, share, adapt and build on them for any purpose, including commercially, as long as you give appropriate credit, link to the licence and say if you changed anything.

Suggested credit: Yik Jinn, ALIQ Studio (2026). What AI answer engines cite when Malaysian SMEs ask for marketing help, v1.0. https://studio.aliqgroup.com/en/research/ai-answer-citations-malaysia-2026/. CC BY 4.0.

## What `trials.csv` covers

576 trials: every English and Bahasa Malaysia question in the panel, asked of Gemini and of ChatGPT through their APIs, three times in each of two windows (26 to 27 September and 4 October 2026). Each cited page in a trial is one row, 2,846 citation rows in all; a trial with no citations has one row with an empty `citationIndex`. The columns are listed in section 12 of the protocol. Google's own AI answers (an exploratory sub-panel) are reported in the tables, not in this file.

## What is withheld, and why

- **Answer text.** The engines' answers name competing agencies, so they are not published.
- **Agencies.** Every agency or freelancer appears as `[agency]`, in both the website and the address columns, so no provider's rows can be matched or followed across rows. ALIQ Studio, the publisher, is treated the same way.
- **Unnamed websites.** Websites that sell marketing, web, creative or advertising services but are coded otherwise appear as `[unnamed]`.
- **Page addresses.** Only the website is kept for most pages, with the path shown as `[path withheld]`. Full addresses are kept for Google's own pages, software and platform vendors' pages, and government or association pages, except where the address is about one business or carries an anonymised name.

## Reproduction limits

The shares by source type, publisher country and page type can be recomputed from `trials.csv`. Measures that follow individual websites cannot be recomputed for agency rows, because every agency carries the same label: the overlap between the two windows and the concentration of citations on a few websites. The tables give those results as computed on the internal data.

## Version

1.0. Later versions keep the same address and list their changes.