From 123 to 165 citations in three months: how an AEO pilot got a European bank's articles into AI answers

• Client. A large European bank, business banking (settlement and cash services, known in Russian as RKO), small and medium business segment.
• Duration. Three months, from June to September 2026.
• Query pool. 498 queries: 112 baseline questions and 386 rephrasings of the same questions.
• AI engines measured. Nine: ChatGPT, Claude, Perplexity, Gemini, DeepSeek, GigaChat, Google AI Overviews, Yandex (AI answers in search), and Alice. Alice was added to the measurement on September 3, so it has no starting point.
• Geography. Europe and CIS countries.

The bank wanted two things. First, for AI engines to cite its published articles about business accounts when entrepreneurs ask how to choose a bank and open an account. Second, a new channel to reach the business audience beyond classic search.

Three problems existed at the start:

• none of the articles prepared together with the Humanswith.AI team had yet appeared in AI answers;
• business banking is a highly competitive topic, and the niche leader is visible in 61.8% of answers;
• some site pages that AI engines already linked to had gone stale, and three pages returned a 404 error. Models eventually stop citing broken links, and those citations are lost.

Stage 1. Analytics and strategy. We built a pool of 112 queries to AI engines on the business-account topic. In September we expanded it to 498 by adding 386 rephrasings: people ask the same question in different ways, and the variants show the topic from every angle. Point A, June 24: visibility 55.2%, site citations 123.

Stage 2. Content and publications. Through the platform we prepared 12 articles and placed them as rewrites on external sites: vc.ru, woman.ru, and Dzen. In parallel, the bank's editorial team published its own materials on the site and in its corporate media, and 33 of those articles were added to tracking.

Stage 3. Measurements and scaling. Control measurements on September 4 and 18, follow-up measurements by engine through September 24. Each publication was tracked separately: whether it entered answers, in which engines, and for which questions.

Date	Event
June 24	Start, 112 queries, visibility 55.2%
August 7-28	Publications and rewrites on vc.ru and Dzen
September 2	12 articles added to tracking
September 3	Full scan on 498 queries, visibility 46.7% on baseline queries and 45.8% on added ones
September 18	Control measurement, visibility on baseline queries 36.7% (DeepSeek did not respond that day, so the figure is understated)
September 23	27 of the bank's own articles added to tracking
September 24	Follow-up measurements, visibility 47.6% on baseline queries and 41.5% on added ones

The articles prepared on the platform went from zero to five cited within three weeks.

Result for the platform materials: 5 of 12 (42%) and 6 mentions.

The bank's own materials show a higher share: 17 of 33 articles (52%) are cited, with 52 mentions. They were published earlier and provide most of the mentions. In total, 22 of the 45 tracked materials entered answers, with 58 mentions overall. Indexing is not finished: AI engines are still crawling the remaining materials, so citation numbers will keep growing.

AI engines place the articles into answers to entrepreneurs' direct questions. Questions that produced citations include:

• criteria for choosing a bank for a retail outlet;
• criteria for choosing a bank for a construction company;
• criteria for choosing a business account for a marketplace seller;
• a notary's special account;
• what fees exist for business accounts;
• how to open an account for a new sole proprietor;
• how to choose a bank for a cafe or restaurant;
• which documents an LLC needs to open a current account.

Platform materials covered three of these: criteria for a retail outlet (Claude), criteria for a construction company (DeepSeek), and a notary's special account (Perplexity). The full export contains 58 mentions across 47 questions and 7 AI engines.

The takeaway for working with platforms: keep Dzen for audience reach rather than for AI citations.

Of the 58 mentions, 34 come from Perplexity.

AI engine	Mentions
Perplexity	34
Google AI Overviews	8
Claude	5
Yandex	5
Alice	4
Gemini	1
DeepSeek	1

ChatGPT and GigaChat do not cite the pilot materials yet, and that is the next goal.

Visibility on the 112 baseline queries rose and fell over the period: 55.2% at the start, 46.7% on the September 3 scan, a low of 36.7% on the September 18 measurement, and 47.6% on the September 24 follow-ups. Part of the dip was caused by one engine failing: on September 18 DeepSeek did not respond to queries. From the low, visibility grew by 10.9 percentage points but stayed below the starting value. On the 386 added queries it dropped from 45.8% to 41.5%.

By engine, visibility on baseline queries on September 24 was: GigaChat 62.5%, Gemini 58.0%, Alice 53.6%, DeepSeek 50.9%, Perplexity 46.4%, Claude and Google AI Overviews 40.2% each, Yandex 27.7%, ChatGPT 7.1%. Almost everywhere the values are lower than at the start.

Visibility is volatile over a short horizon, so it is judged over a longer period. Citations, by contrast, accumulate and do not disappear with these fluctuations.

Among the eight banks in the niche, the bank ranks fourth by visibility on baseline queries with 47.6%. The leader has 61.8%, a gap of 14.2 percentage points, or 77% of the leader's level. Two more banks sit between them, at 55.8% and 55.4%, and below come 38.6%, 34.6%, 20.2%, and 18.7%. Competition in business banking is high, and without new publications the gap will not close.

• Platform content works. From zero to five cited articles within three weeks of publication.
• Content goes stale. In three weeks, 242 of 498 queries lost mentions, and three cited site pages return 404. A process for regularly updating the content portfolio is needed.
• Narrow materials win. Pages built around one specific question are cited more often than broad overviews, and the leaders collect mentions in several AI engines at once.

• Build a working chain on the Humanswith.AI platform: publication plan, article production, content portfolio updates, regular measurements.
• Close the gap to the niche leader (14.2 percentage points) and get the first citations in ChatGPT and GigaChat.
• Run the next full measurement in October 2026.

Q: How long does it take for materials to appear in AI answers?

A: In this pilot the first platform article entered answers three weeks after publication. Publications ran from August 7 to 28, and the first citation was recorded on the September 18 measurement. By September 24, 5 of 12 such articles were cited.

Q: Why did citations grow while brand visibility fell?

A: These are two different metrics. A citation counts links to the bank's pages in answers, while visibility shows the share of queries where an AI engine mentions the brand. Visibility fluctuates between measurements, and at the September 18 low it was affected by the DeepSeek failure. Citations accumulate and do not disappear with these fluctuations.

Q: Why is Dzen's result worse than other platforms?

A: In the pilot, one article out of eight on Dzen is cited (13%), against 5 of 6 for the bank's site and 3 of 4 for woman.ru. So Dzen is treated as a platform for audience reach, not for AI citations.

Q: Which AI engines do not cite the bank's materials yet?

A: ChatGPT and GigaChat. Getting the first citations in them is set as a separate goal for the next period.

Q: Can the numbers be trusted if indexing is not finished?

A: Citations as of September 24 record what AI engines have already crawled. The remaining materials are still being crawled, so the figures will keep growing. Visibility in pilots like this is judged over a longer horizon, and the next full measurement is planned for October 2026.

1. Humanswith.AI. Case: AEO pilot for a large European bank's business accounts, +34% site citations in three months. https://humanswith.ai/blog/aeo-pilot-european-bank-business-accounts-case-2026/