The State of AI Visibility in Sri Lankan Banking

How AI describes Sri Lanka's banking sector

Between November 2025 and June 2026, SiteSignal tracked how ChatGPT, Gemini and Perplexity answer questions about Sri Lankan banks, logging every institution named, every source cited, and every fact checked against the truth. This is the richest single dataset in SiteSignal's monitoring archive.

• 178 distinct institutions were named by an AI platform over the monitoring window, the six largest licensed commercial banks anchor nearly every month's leaderboard, with a long tail of finance companies, digital banks and international comparators filling out the rest (Section 03).
• Citations lean heavily on a bank's own website, roughly two-thirds of all citations point to owned domains, a notably higher share than in other sectors SiteSignal tracks, consistent with banking being a category where regulatory and product-specific information has few independent substitutes (Section 05).
• Mention rates for a tracked bank were remarkably stable across funnel stages, unlike categories where AI visibility concentrates sharply at the decision stage, Sri Lankan banking questions surface a bank's name at a broadly similar rate whether the question is educational or a direct comparison (Section 07).
• ChatGPT and Perplexity are both far more willing to commit to a specific answer about a Sri Lankan bank than the global average SiteSignal sees elsewhere, insufficient-data rates under 2% for ChatGPT, though which platform is more often wrong shifts over the monitoring window (Section 08).
• Technical crawlability is the clearest actionable gap: in the sites SiteSignal audited, every single audit failed the check for allowing known AI crawlers (Section 10).

Six terms, defined once, used throughout

Mention

An AI platform said an institution's name out loud in its answer, in any context, a recommendation, a comparison, or a passing reference in a list.

Visibility share

One institution's mentions as a percentage of every mention across the whole tracked market that period. A 10% share means roughly 1 name-drop in 10 went to that institution.

Citation

The specific webpage an AI platform pointed to as the source behind a claim, separate from which institution is being discussed.

Source type

What kind of page got cited: Owned (the bank's own website), Editorial (independent coverage, news, comparison sites, forums), Documentation (rate sheets, regulatory filings), Review, Social, or UGC (user-generated content).

Authority & reusability

Two quality scores applied to every citation: authority is how credible/established the source looks; AI reusability is how easily a model could lift a fact straight off the page without interpreting it.

Funnel stage

Every tracked question is tagged by where a customer would be in their decision: Awareness ("what is a fixed deposit") · Consideration ("options for a savings account") · Evaluation ("which bank should I use").

Fact-check status

When a model states a specific claim about a bank, SiteSignal checks it against ground truth: Match (correct) · Partial match (roughly right) · Mismatch (wrong) · Insufficient data (the model declined to give a specific answer).

One methodology note: tracking ran continuously from 4 Nov 2025 through 24 Mar 2026, then resumed 19 May – 24 Jun 2026 after a platform migration, there is a real six-week gap in April 2026 with no data. Every chart and table below marks that gap explicitly rather than interpolating across it. No client account is named or identifiable anywhere in this report.

Eight banks, tracked month by month

The chart follows the eight highest-volume institutions across the monitoring window. The dashed segment marks the six-week gap in April with no tracking data, the lines either side of it are real, the connector across it is not.

• Sampath Bank and Bank of Ceylon peak hardest in January (794 and 748 mentions respectively) before settling back, the sharpest single-month spikes among the six domestic majors.
• DFCC Bank is the most volatile of the six large domestic banks month to month, swinging from 265 mentions in January to 639 in February, no other bank in the top six moves that sharply in consecutive months.
• After the April gap, June's volumes for the two international banks in this set (Standard Chartered, HSBC Sri Lanka) recover close to their pre-gap levels, while several domestic banks, Bank of Ceylon and People's Bank especially, return far below where they left off, still climbing back as of June.

Leaderboard by quarter

Q4 2025 (Nov–Dec) · 4,722 tracked mentions

Q1 2026 (Jan–Mar) · 17,816 tracked mentions

Q2 2026 (May-Jun, partial: April untracked) · 1,723 tracked mentions

Q2 2026's share figures come from a smaller, post-migration prompt set weighted toward direct comparisons, treat the share percentages as directional, not comparable one-to-one against Q4/Q1's larger sample. International fintech and accounting platforms (Xero, and others further down the list, Chase, Mercury, Block, BlueVine) begin appearing here for the first time, a genuine widening of what gets compared against Sri Lankan banks in AI answers, not a data artifact.

A consolidated top six, a shifting order beneath it

• The same six licensed commercial banks, Sampath, Bank of Ceylon, DFCC, Seylan, People's Bank and Standard Chartered, place in the top eight in every quarter tracked. Below that settled group, HSBC Sri Lanka, Nations Trust Bank, National Savings Bank and Hatton National Bank rotate through the remaining places.
• DFCC Bank shows the widest swings of the six majors, its visibility share more than doubled from Q4 2025 to Q2 2026 (10.5% to 45.4%, largely a post-migration prompt-set effect worth reading cautiously per the note above), while its raw month-to-month mention count also swung more than any peer's.
• International names entering the conversation is the real Q2 story: Xero, Chase, Mercury and Block/Square all cleared the visibility bar for the first time once tracking resumed, Sri Lankan banking questions are increasingly being answered with reference to global digital-banking alternatives, not just domestic peers.

Every institution the monitoring surfaced: 178, in full

Licensed banks, finance companies, digital-lending platforms and the international comparators AI increasingly names alongside them, ranked by total tracked mentions.

A sector that still runs on the bank's own website

40,900+ citation events were classified by source type. Banking looks different here from other sectors SiteSignal tracks:

• 65% owned-site citations is a markedly higher share than the roughly 45–50% typical of other categories SiteSignal tracks at a comparable point in their monitoring, consistent with banking being a category where rates, product terms and account details genuinely only live in one authoritative place: the bank's own site.
• Unlike categories where owned-site share erodes steadily as independent commentary grows, DFCC's monthly figures held in a tighter 60–80% band throughout, without a clear directional trend either way.
• This is itself a useful contrast for a bank's content team: in this sector, a well-structured product and rates page is disproportionately likely to be exactly what a model cites, there is less of an independent-commentary ecosystem to compete with than in more consumer-discretionary categories.

Tone, full period

LinkedIn is the quiet constant; everything else is noise-level

*June's figures jump sharply because Facebook, X/Twitter and Wikipedia citations also entered the mix for the first time post-migration (Facebook alone reached 71 citations, more than any single social platform in the entire Nov–Mar window), read as an early post-migration signal, not a settled trend, the same caution applied to the equivalent finding in SiteSignal's other sector reports.

• Unlike the sharp, sustained Reddit surge SiteSignal has documented in other sectors, Sri Lankan banking shows no comparable single-platform breakout through the tracked window, LinkedIn is consistently the largest social citation source, but never exceeds roughly 1% of total citations pre-gap.
• The post-gap jump in Facebook, X and Wikipedia citations is the one data point worth watching going into the next quarter, if it holds through July and August, it would mark banking joining the broader pattern of AI leaning more on social platforms as a citation source.

A flatter funnel than most categories SiteSignal tracks

• Even educational, category-level questions name a specific bank more than half the time in this sector, a sharp contrast to categories like uniform & workwear, where awareness-stage questions almost never surface a specific supplier.
• The gap between education and decision stage is real (51.5% to 66.4%) but far gentler than the multi-fold jumps SiteSignal has found elsewhere. Banking appears to be a category where even generic questions ("what is a fixed deposit") get answered with reference to specific institutions' actual products, likely because rate and product specifics are genuinely bank-specific rather than generic.
• The practical read for a bank's content strategy: unlike categories that reward decision-stage content almost exclusively, category-education content in banking is already a meaningful visibility channel, it doesn't need to be abandoned in favor of comparison pages alone.

Both major platforms commit far more often here than SiteSignal's cross-sector average

• ChatGPT's 1.1% "insufficient data" rate is the lowest SiteSignal has recorded for this platform in any sector, it almost never declines to state a specific fact about a Sri Lankan bank, and is right (match + partial) nearly 80% of the time it does.
• Perplexity hedges more here than ChatGPT does but far less than its own behavior in other sectors SiteSignal tracks, 23.5% insufficient-data, against rates north of 30% seen elsewhere for the same platform.
• Gemini's small sample (only tracked post-migration) shows a hedging pattern broadly consistent with its behavior elsewhere, insufficient data is its single largest outcome, though the sample here is too small to weight heavily.
• The practical read: banking appears to be a category where the major platforms have unusually strong, specific training signal, plausibly because banks publish extensive, well-structured public information (rates, terms, regulatory filings) that gives models more to work with than in less-documented categories.

Where AI gets a bank's facts wrong, and one place it only looks that way

1,272 fact-checks across the monitoring window, checked against ground truth. One finding needs a caveat before the data, because it would otherwise read backwards:

Founding year, country of registration and headquarters are highly reliable, 68–85% match. A published contact email is wrong more often than any other field checked, in either direction (support email 23.6% mismatch, general public contact email 56.9% mismatch).

• Contact-detail fields are, again, the least reliable category, exactly the same pattern SiteSignal found in its UK uniform & workwear report. Public contact email addresses are wrong outright 56.9% of the time, the single worst rate recorded for any field in either sector study.
• Terms URL is unusually unreliable for a banking-specific reason: reading the raw answers, several "mismatches" are the model correctly pointing to a current terms page while ground truth held an older or differently-structured URL, banks appear to restructure these pages often enough that even accurate answers can look wrong against a static reference.
• Identity basics that don't change often (founding year, country, HQ) stay reliable across the board, the same structural pattern as every other sector SiteSignal has studied: hallucination concentrates in fields that require current, specific, frequently-changing information.

A sharper gap than in other sectors SiteSignal has audited

SiteSignal ran 3 infrastructure audits within the monitoring window. The sample is small, but the pattern is stark enough to flag with confidence.

"Trusted AI crawlers allowed" checks whether a site's robots.txt explicitly permits GPTBot, ClaudeBot, PerplexityBot and similar crawlers, every audit in this small sample failed it outright. Combined with a 33% pass rate on having a valid robots.txt at all, this points at a genuine, fixable crawlability gap rather than a close call.

Five things that held across the window

1. A settled top six, with real month-to-month volatility inside it

Sampath, Bank of Ceylon, DFCC, Seylan, People's Bank and Standard Chartered anchor every quarter's top eight. Which one leads which month moves substantially, this is a market with a stable membership and an unsettled pecking order.

2. Banking leans on owned content more than other sectors

65% of citations point to a bank's own site, meaningfully higher than the roughly 45–50% typical elsewhere in SiteSignal's coverage, consistent with banking being a category with few good independent substitutes for rate and product information.

3. The funnel gap is real but gentle here

Mention rate climbs from 51.5% at awareness to 66.4% at evaluation, present, but nowhere near the multi-fold jump seen in more discretionary categories. Category-education content still carries real visibility weight in banking.

4. Both major platforms commit unusually often

ChatGPT's 1.1% insufficient-data rate and Perplexity's comparatively moderate 23.5% both sit well below what SiteSignal has measured for the same platforms elsewhere, banking appears to be a well-documented category that gives models more to work with.

5. Hallucination concentrates in the same kind of field, again

Contact details are the least reliable fact type, in this sector as in every other SiteSignal has studied. Identity basics (founding year, country, HQ) stay reliable everywhere.

What this report is, in one place

• This report covers Sri Lankan banking specifically. Citation mix, funnel behavior and fact-accuracy rates plausibly look different in a different market or category, the uniform & workwear comparisons drawn throughout are offered as cross-sector context, not a claim that these patterns generalize.
• Tracking ran 4 Nov 2025 – 24 Mar 2026, then 19 May – 24 Jun 2026 after a platform migration, with a six-week untracked gap in April 2026 disclosed wherever it affects a number.
• No client account is named or identifiable anywhere in this report. Every figure describes the wider market a monitored account's tracking surfaced, not that account's own performance.

The clearest actionable finding in this report is also the simplest: Sri Lankan banking is a sector where AI answers still lean heavily on a bank's own website, and where, in the sample audited, that website isn't yet configured to let the crawlers behind those AI answers in. Closing that gap is a smaller, more concrete project than most AI-visibility recommendations, and this sector has more to gain from it than most.