How AI answers about banks change with the question and service
A six-market banking study found different first-mentioned banks across tested AI routes and a wider field for business questions, suggesting that a single visibility score hides the customer need.
A bank can lead one tested AI answer and be absent from another. That is the decision problem raised by PRecious Communications' AI Authority Index of Southeast Asian banking: a single tested route, question or visibility score can hide differences in the recorded answers.
The agency compared tested answers about banks in Indonesia, Malaysia, the Philippines, Singapore, Thailand and Vietnam. In 706 of 1,628 eligible matched comparisons, or 43%, two tested routes answering the same question named a different bank first. The comparison included only pairs where both answers named a recognised bank. It measures answer order, not a customer's preference, a bank's quality or the likelihood of a sale.
The question changes the field
Business-banking questions named an average of 4.73 banks per answer, compared with 3.52 for consumer-banking questions. Each question group contained 304 answers in the study's question-family analysis. The roughly 34% wider field matters for a regional bank with distinct consumer and corporate offerings: a general brand check may hide the competitors that appear when a buyer asks about trade finance, growth or operating needs.
The practical test is to keep the customer need attached to the answer. Communications teams can compare a small, repeatable set of questions for each service line, record which banks are mentioned and which comes first, then inspect the source evidence behind the answer. A bank that appears often but rarely leads may face a different content problem from one that is absent altogether. Neither pattern alone proves why an AI service answered as it did.
Route and language need their own checks
The first-mentioned bank differed between the English and Thai answers in 15 of 25 eligible Thailand pairs, and between the English and Indonesian answers in 10 of 26 eligible Indonesia pairs. Those are small, market-specific comparisons. They show why an English-only review may miss a different answer in a language customers use; they do not establish a universal effect of Thai or Indonesian wording.
The label on an AI route also matters. The report's main comparisons mostly used APIs under differing search and source conditions. Its Copilot-labelled test was an Azure model with web search, not Microsoft 365 Copilot; its Google-labelled route was a search-grounded Gemini API, not Google AI Overviews. Some Qwen and DeepSeek sets used frozen supplied sources. These settings make the study a diagnostic snapshot of tested answers, not a measurement of what every consumer saw in a live app.
For a Southeast Asian marketing team, the useful next step is an audit matrix: customer question, market, language, tested service and date, followed by the named banks, answer order and cited evidence. Check whether core product facts and credible third-party sources are current, correct and available in the language being tested. Then rerun the same questions after a material change. This will reveal shifts in recorded answers; it will not by itself attribute them to a campaign or forecast market share.
The public report page describes the tested AI-answer findings as an August 2026 snapshot. It covers 13 priority banks and three global-bank examples, rather than a representative census of every bank in the region.
Source: PRecious Communications, AI Authority Index: Southeast Asia Banking 2026, public findings and method summary.
Disclosure: PRecious Communications and Southeast Asia Connect are part of the PRecious Group. PRecious Communications produced the AI Authority Index cited here. SEA Connect selected and edited this article and subjected the final content to separate editorial quality review.
