Research · Banking · 23 Oct 2026
A bank’s organic traffic fell a third: what explains it and what not A study on a large bank’s data: seven hypotheses, a check of each, and linked reviews
A large bank’s organic visits fell by a third in a year, brand dropped more than non-brand, and conversion rose. We went through the analytics hypothesis by hypothesis and separated what the data confirm from what is merely consistent with them.
- Organic, May 20251,460
- Organic, May 2026−32%1,000
- Brand, May 20251,020
- Brand, May 2026−38%632
- Non-brand, May 2025448
- Non-brand, May 2026−18%368
Internal data of a large bank: web analytics, Search Console, Wordstat and a rank tracker, January–May 2026. The name, competitors and URLs are removed. Values are anonymised: absolute figures are altered and rounded, and relative changes are recomputed from the numbers shown.
In short
- The bank’s organic traffic fell by a third in a year (−32%), brand more than non-brand (−38% vs −18%). From January to May conversion rose from 0.61% to 0.71%, yet the number of leads from organic still dropped 15%.
- Of seven hypotheses one is refuted by the data, three are partly supported, two are consistent with the data but not measured directly, and one was not tested. The label “confirmed” in the source analytics means “does not contradict the data”, not “cause proven”.
- The main methodological finding: one report holds three different bases of “organic” (708k, 590k and 1.00M in May). Percentages inside a base are comparable; sums across bases are not. Reconcile the bases first, then look for a cause.
What happened: four numbers and one trend
We compare May 2025 with May 2026. “Total site traffic” is taken from the source as is: its composition is not explained there and it is not comparable with the channel summary (where paid and direct each exceed this total). Organic means visits from search engines.
| Metric | May 2025 | May 2026 | Change |
|---|---|---|---|
| All site traffic | 1.85M | 1.14M | −38% |
| Organic | 1.46M | 1.00M | −32% |
| Brand (within organic) | 1.02M | 632k | −38% |
| Non-brand (within organic) | 448k | 368k | −18% |
The non-brand share grew from 31% to 37%, not because non-brand is growing but because brand fell faster. In the five-month data the topical share of organic also declined from February.
- January966
- February871
- March920
- April844
- May−16% vs April708
The monthly summary shows the decline before the May holidays. From January to April organic in the channel summary lost 13% (966 → 844k), and May added another 16%. Caveats: January includes the holidays, December is not in the data, and March is above February. Holidays explain the dip inside May but not the whole trend; see the review on seasonality.
Quality metrics rose, but this may be an effect of the traffic mix and a change in measurement. Organic bounce fell from 17.3% to 12.8%, pages per visit rose from 2.25 to 2.86, and conversion to a full application rose from 0.61% to 0.71% (January → May). But there were fewer qualified leads: from 5,920 to 5,050 a month, −15%. The rise in conversion is arithmetically consistent with losing visits that converted worse (not proven), and the April jump in bounce should be checked as a possible measurement change; details in the review on conversion vs volume.
Seven hypotheses: what the data actually say
The source analytics gave each hypothesis a label. We kept it and added our own assessment: what is measured and what merely does not contradict the hypothesis.
| Hypothesis | Label in the analytics | Our assessment | Review |
|---|---|---|---|
| Fall in search demand | Confirmed | Partly: for deposits demand fell more than traffic, for debit cards traffic fell more than demand | Demand vs traffic |
| AI answers in results | Confirmed | Consistent: visibility holds, visits fall, but there is no query-level click data | AI answers and definition pages |
| Shift to chatbots | Confirmed (indirectly) | Not tested: no measurement of visits from assistants | Chatbots |
| Local loss of visibility | Partly confirmed | Partly: falling in some sections, growing in others | Sections |
| Cannibalisation by paid | Refuted | Refuted in the form “traffic moved to ads”: paid fell by the same share | Paid and direct |
| Outdated content | Confirmed | Consistent: pages that fell were last updated 8–19 months earlier, but there is no control group | Outdated pages |
| Seasonality | Partly confirmed | Partly: holidays explain April → May but the decline started earlier | Holidays and calendar |
We use four labels: “refuted” (data contradict), “partly” (part of the effect is measured), “consistent” (the picture fits but the cause is not measured) and “not tested” (no measurement). For budget decisions the last two matter most: that is where money gets spent on things that do not work.
Bases first: three different “organic” numbers in one report
The source analytics gives May 2026 organic traffic as three different numbers. The samples are not disclosed in the source, so the numbers cannot be added or compared with each other.
| Where in the report | What it shows | May 2026 |
|---|---|---|
| Channel summary (web analytics) | Visits from search engines | 708k |
| Split by two engines: brand + topical | Sum of brand and topical in Yandex and Google | 590k |
| Brand / non-brand year over year | “Organic” in the May vs May comparison | 1.00M |
One possible explanation: part of the visits (about a sixth) fall into neither brand nor topical — across the two engines these cuts cover about 84% of their visits. The source does not say, so we do not claim a cause; we only record the 1.4x gap between the first and third number.
A practical rule: write down the base of every number — source, filters, period, device. When two numbers about the same thing disagree, reconcile the bases first and only then look for a cause in the site or in search. The same happened with site sections: their sum did not match the channel summary; see the section breakdown. Also check whether a conclusion rests on growing segments: the subset trap.
What is known worldwide and what it does not prove
Two outside studies show that an AI block in Google results lowers the share of clicks on ordinary links. Both cover another market and other topics, and neither proves the cause of one bank’s decline.
| Source | What was measured | Result | Caveat |
|---|---|---|---|
| Pew Research Center, 2025 | Browsing of 900 US adults in March 2025, Google only | A click on an ordinary link in 8% of visits with an AI summary and 15% without | Tracked panel, US, not a financial topic |
| Ahrefs, 2026 update | 300k keywords, Search Console, December 2023 and December 2025 | Position-one CTR (desktop) 58% lower when an AI Overview is present | Correlation; the baseline is a forecast, not an observation |
The conclusion for a bank is cautious: the direction of the effect is supported, the size for your site is not. It has to be measured on your own queries — here is how. Third-party estimates for competitors need the same caution: how to read third-party benchmarks.
Can the decline be explained yet: six questions
Answer “yes”, “no” or “don’t know” about your site. This tests readiness for a diagnosis, not the cause: until the answers are “yes”, any cause stays a guess.
- Are the same months compared year over year, with holidays and day swaps accounted for?Calendar
- Are brand and non-brand split by one fixed list of brand queries?Query set
- Is traffic checked against demand for the same queries (Wordstat, Trends)?Demand
- Are impressions and position in Search Console read separately from clicks?Results page
- Have paid and direct channels been checked for the same movement?Neighbouring channels
- Is the base written down for every number: source, filters, device, period?Base reconciliation
This study’s own working scale, not an industry standard. Any “don’t know” answer makes the page show “Gaps remain” and no verdict.
Where to start when organic falls
- Write down the base of every number and reconcile the gaps, as in the bases section. Start with the section breakdown.
- Split brand and non-brand and check both lines against demand and traffic; compare engines only with the query mix in mind (Google and Yandex).
- Separate calendar and events from structural decline: holidays, deadline-driven pages.
- For definition and reference pages, measure the effect of AI blocks on your own queries rather than relying on others’ studies (AI answers and definition pages); check falling impressions with rising position separately (impressions and position).
- Update pages when facts change, not on a calendar (outdated pages, pages with a year in the title).
- Only then plan the work and how its effect will be checked (plan).
FAQ
Why did conversion rise while traffic fell?
Possibly the lost visits were less valuable: lost leads came to about 0.34 per 100 lost visits against January’s conversion of 0.61% (a ratio of two differences, not the conversion of the departed visits; not proof: a change in goal tracking is also possible). But the number of leads still fell 15%. Details in the review on conversion vs volume.
Can this study be used to judge my own site?
No, it is the data of one site. The checks and the list of questions transfer; the numbers and the proportions between causes do not.
Why are there no bank names, competitors or links?
The data are anonymised: absolute figures are altered and rounded, and relative changes are recomputed from the numbers shown, so they may differ from the source analytics by 1 percentage point. For third-party banks the values are altered more heavily, so their percentages are illustrative.
What is known about AI blocks worldwide?
Pew and Ahrefs show fewer clicks when an AI summary is present in Google results, but neither is about banks or this market, and Ahrefs reports a correlation. The size for your site has to be measured on your own data.
Sources and method
- Pew Research Center: Google users are less likely to click on links when an AI summary appears — clicks with and without an AI summary, US, March 2025
- Ahrefs: AI Overviews reduce clicks (update) — position-one CTR, correlation, 300k keywords
- Search Console Help: impressions, position and clicks — how clicks and position are counted
- Internal data of a large bank: web analytics, Search Console, Wordstat, rank tracker (January–May 2026)
Method: data of one large bank for January–May 2026 (May vs May and April vs May comparisons). Values are anonymised: absolute figures are altered and rounded, and relative changes are recomputed from the numbers shown; for third-party banks the values are altered more heavily, so their percentages are illustrative. The assessments “consistent / partly / not tested” are ours; the “confirmed” labels come from the source analytics. The data were not re-checked inside the systems. As of 23 Oct 2026.