SiteSignal
When people ask AI engines how to contact a Sri Lankan bank, the email address they get is wrong more often than it is right. Public contact email addresses were wrong outright 56.9% of the time, the worst rate SiteSignal has recorded for any field in any sector it has studied.
The finding comes from a seven-month study of how ChatGPT, Gemini and Perplexity answer questions about Sri Lankan banking, covering 178 institutions.
Key findings
- Contact emails are the weak spot: wrong 56.9% of the time, while basic identity facts such as a bank's name and headquarters were far more reliable.
- ChatGPT rarely says "I don't know": it declined to state a fact only 1.1% of the time, the lowest rate SiteSignal has measured for ChatGPT in any sector, and was right, fully or partly, nearly 80% of the time it did.
- Banks' own websites still lead: 65% of citations pointed to a bank's own site, compared with roughly 45–50% in other sectors SiteSignal tracks.
- A settled top six: the same six domestic banks led throughout, but their monthly visibility swung sharply. DFCC Bank moved from 265 mentions in January to 639 in February.
- Visible early and late: banks were mentioned in 51.5% of early-stage questions and 66.4% of comparison questions, a much flatter gap than in most categories.
What it means for banks
"Customers now ask AI how to reach their bank, and more than half the time the email they get is wrong," said Niraj Appuhamy, founder of SiteSignal. "Banks already publish good information on their own sites, and AI uses it. The gap is in the small, changeable details, and those are exactly what customers act on."
About the data
SiteSignal asked ChatGPT, Gemini and Perplexity questions about Sri Lankan banking through each engine's official API over seven months (November 2025 to June 2026, with April untracked) and fact-checked the answers against each bank's published details. Full methodology and data: The State of AI Visibility in Sri Lankan Banking.
Press contact: hi@sitesignal.app