Review · Banking · 23 Oct 2026
“What is…” pages and AI answers: lost clicks or falling demand A bank’s knowledge base: visits −24%, demand −13%, Yandex visibility rising. What that proves and what it does not
A bank’s reference section with short definitions lost a quarter of visits in a year. Demand for its queries fell half as much, and visibility rose in Yandex (and dipped slightly in Google). That is a classic sign of an answer in the results page taking the clicks, but a sign is not proof.
- Visits, May to May−24%
- Demand, May to May−13%
- Visits, April → May−24%
- Demand, April → May−6%
Internal data of a large bank’s reference section, web analytics and a rank tracker, 2025–2026; Pew Research Center (US, March 2025). Values are anonymised: absolute figures were altered and rounded, and relative changes were recalculated from the numbers shown.
In short
- If visits fall much faster than demand while visibility and positions hold, the click share is probably changing. Here: visits −24%, demand −13%, Yandex visibility up, Google visibility down moderately.
- It is a sign, not proof. The data hold no direct CTR measurement for queries with and without a block, and the section’s sharpest drop (−73%) is most likely explained by demand and a news event; whether an AI effect is separated is not visible in these data.
- The size of the effect must be measured on your own queries: 20–30 queries, a dated record of whether a block shows, CTR compared at equal position. Others’ studies give a direction, not a figure for your site.
The signal: visits fall faster than demand, Yandex visibility rises, Google visibility dips moderately
Three indicators combine into one picture. Visits fall 24%, demand for the section’s queries 13%, while the visibility index for tracked queries rises in Yandex and dips slightly in Google.
| Indicator | May 2025 → May 2026 | April → May 2026 |
|---|---|---|
| Visits from search | −24% | −24% |
| Demand for tracked queries | −13% | −6% |
| Visibility index, Yandex | 15.9 → 29.2 (+13.3) | 27.5 → 29.2 (+1.7) |
| Visibility index, Google | 54.3 → 48.5 (−5.8) | 51.8 → 48.5 (−3.3) |
If visibility and positions hold while clicks leave faster than demand, there can be several reasons: an answer block in results, changes among neighbouring widgets, reshuffling within the top 10. The data do not separate them.
The visibility index is computed by the rank tracker, which sets the method. These data hold no query-level CTR, so the conclusion stays “consistent”, not “proven”. On how AI impressions show in Search Console, see the guide on AI impressions in Google.
Which pages lose the most
Definition pages lose differently. The sharpest drop is on a page tied to the news; it is most likely explained by a news spike, and an AI effect is not separated in these data (next section).
| Definition page | April | May | Change |
|---|---|---|---|
| Mortgage term | 3,170 | 1,890 | −40% |
| Payment indexation term | 1,240 | 340 | −73% |
| Car loan term | 656 | 308 | −53% |
| Enforcement proceedings term | 1,500 | 1,190 | −21% |
| Refinancing term | 234 | 75.2 | −68% |
| Consumer loan term | 447 | 342 | −23% |
The range from −21% to −73% shows the cause is not single.
The “indexation” case: demand and an event explain the fall, an AI effect is not separated
The section’s biggest drop was on a generic query that can refer to pensions or to salaries. In April there was a demand spike driven by the news and the site got extra clicks. In May demand fell after the April spike.
- impressions for this query in Google fell by 358k in May against April;
- the average position improved from 4.4 to 3.2;
- so clicks left together with demand, not because of position.
The conclusion: before blaming an AI block, separate event pages. They should be judged on their own calendar (deadline and event pages), not on the average of a reference section.
What studies show worldwide
| Source | Result | Caveat |
|---|---|---|
| Pew Research Center: 900 US adults, March 2025, Google | A click on an ordinary link in 8% of visits with an AI summary and 15% without; a summary appeared in 18% of searches | US, tracked panel, not banking queries |
| Ahrefs: 300k keywords, Search Console, December 2023 and December 2025 | Position-one CTR (desktop) 58% lower with an AI Overview; −51% at position two, −19% at ten | Correlation, baseline is a forecast, other SERP features also cut CTR |
The two sources agree on direction. Neither says how many clicks specific banking definitions lose on a Russian-language results page.
How to measure the effect on your own data
- Choose 20–30 definition queries for which you have pages.
- On the same date, region and language, record whether an answer block appears. Repeat weekly.
- In Search Console take the CTR of these queries and compare “block present” with “no block” at equal average position.
- Match it with demand for the same queries so a fall in demand is not taken for lost clicks.
- Mark event-driven queries separately and exclude them.
Has an answer “eaten” a definition page: a six-question check
Answer “yes”, “no” or “don’t know” for one page or a group. “Yes” answers are signs of clicks lost to an answer in the results.
- Is the page a short definition with no calculation, product comparison or data missing from an answer block?Content
- Do visits fall noticeably faster than demand for the same queries?Gap vs demand
- Is position for the main queries stable, within 2 positions?Position
- Does today’s results page show an answer block (checked by hand on the date)?Manual check
- Are Search Console impressions stable while CTR falls?Impressions and CTR
- Does the page lack a next step: a product link, calculator or comparison?What the reader gets
This review’s own working scale, not an industry standard. Any “don’t know” answer makes the page show “Gaps remain” and withhold the verdict: close the gap before concluding.
What to do with definition pages
- Split pages into event-driven and permanent and do not mix them in one assessment.
- For permanent ones, measure CTR with and without an answer block at equal position.
- Do not delete definitions automatically: check whether they lead the reader on to a calculator or product.
- Add what an answer block lacks: a calculation, product terms, a comparison, a date and a source for rates.
- Do not expect FAQ markup or structured data to guarantee an effect on clicks: it was not measured in these data.
FAQ
Should “what is…” pages be deleted?
No. First check whether they bring readers to a product or calculator and measure CTR. Delete only what gives neither traffic nor a next step.
Is it true that an AI block takes clicks?
Pew and Ahrefs show a lower click share with an AI summary in Google. Pew: 8% vs 15% of visits with a click in the US; Ahrefs: a correlation across 300k queries. The size for your site has to be measured.
How do I know the cause is the AI block and not demand?
Reconcile visits with demand for the same queries and look at CTR at stable position. If demand and position hold while CTR falls, the cause is the shape of the results.
Are AI blocks visible in Search Console?
Do not rely on a separate slice: impressions in blocks may be part of the totals. Details in the guide on AI impressions.
Sources and method
- Pew Research Center: Google users are less likely to click on links when an AI summary appears — 8% and 15%, US, March 2025
- Ahrefs: AI Overviews reduce clicks (update) — CTR by position, correlation
- Search Console Help: impressions, position and clicks — how impressions and CTR are counted
- Internal data of a large bank’s reference section (2025–2026)
Method: data of one reference section of a large bank, May 2025, April and May 2026. Values are anonymised: absolute figures were altered and rounded, and relative changes were recalculated from the numbers shown. There is no direct query-level CTR measurement, hence the “consistent” conclusion. As of 23 Oct 2026.