AI answers within integrated business promotion

What generative search research tells us

Is an owned website enough to appear in AI answers? This is a question about where prospects find company information. There is no universal answer without understanding the questions tested and how results were recorded.

The author studied Yandex generative answers through an API returning source links. It measures one system, not all AI or search methods.

How measurement was conducted

The sample covered online services for entrepreneurs, dentistry in Saratov and production-control programmes for catering businesses. The author prepared 40 questions for each topic and reports 160 requests including repeated tests.

The author saved answers, sources and the internal search formulations used by the system, then calculated metrics programmatically. This enables analysis of the recorded answers and links but does not make the sample exhaustive.

The author repeated all 40 online-service questions. Source lists matched in 28 pairs, according to the study. This shows stability within part of a sample, not completely predictable results. The other pairs did not match completely.

Limits of the conclusions

The study covers one search system, three topics and one city in its regional sample. Treat its figures as that measurement, not independently verified statistics for all Russian businesses.

The author warns that 40 questions do not cover a company's entire search demand. The study is a starting point for testing AI presence hypotheses. It is too early to conclude that an owned website is always sufficient or no longer needed.

Why an owned website's role varies by sector

Sources differed across the sectors studied. External publications mattered in some, directories in others and provider websites elsewhere. These observations do not establish a single promotion formula for every business.

Three sectors with different source patterns

For online services, the author counted 250 pages across 81 domains. Ten domains accounted for 34% of sources. Reviews and specialist publications featured alongside providers' websites. Within this sample, the service's information environment extended beyond its owner's material.

The dental sample differed: 101 of 131 page mentions referred to third-party platforms, while 30 linked to clinic websites. Ten domains accounted for 73.3% of sources, including maps, aggregators and directories. The author observed much greater concentration among a small group of external resources.

In the third category, the researcher counted 112 links across 66 domains. The ten most represented domains accounted for 35.7% of sources, and providers' own websites featured prominently. No maps or aggregators appeared in the recorded sources. This describes one measurement, not proof that such platforms never contribute to answers on the topic.

Implications for allocating effort

Editorial inference from a limited sample: integrated promotion requires checking which source types appear for your sector. A fixed split between website, directories and external articles may ignore important differences.

If reviews feature prominently, assess external content. If directories recur, check your listings. If provider websites dominate, examine which pages are cited. These are analytical leads, not a ready-made budget formula.

A high share of external sources does not make your website unnecessary or justify abandoning SEO. Dental findings also cannot be generalised to every local business. Integrated digital marketing aligns channels with business needs rather than replacing one universal recipe with another.

Being cited and being named are different objectives

Finding a page does not mean citing it, and using its content does not mean naming the business. Availability, use as a source and company visibility are separate outcomes.

The study's API data distinguished these levels:

• The page was found. It was found among documents, but may not have been used as an answer source.
• The page was used. The content was used as an answer source.
• The company was mentioned. Its name appeared in the answer, a separate outcome from the previous two.

The author reports using an average of 2.8 pages out of every ten found. This describes the sample, not a fixed selection rule. Some answers used a corporate website without naming its owner. Company names were absent from 48% of answers in the online-services part of the study.

How a question becomes a search query

The export showed search queries that differed from users' original questions. In the author's examples, a long conversational prompt became a short query retaining the service, city, opening hours or selection task. Assess a page not only against the original wording, but also by how clearly it addresses the underlying need. This is an editorial inference, not a proven explanation of the algorithm.

What to check in page content

The author suggests identifying the content owner and author and using structured data. The source offers no before-and-after comparison. Clear headings, separating service details from reference information and linking content explicitly to the company are hypotheses to validate, not proven methods of answer inclusion.

Changes may improve clarity, but cannot guarantee the company name remains in an AI answer. A corporate link and a named mention are different metrics.

Generative search within integrated digital marketing

Connect generative answers with content, external platforms and analytics. The sequence below is an editorial recommendation for testing hypotheses, not a proven method for increasing sales.

1. Collect customer questions. Use sales conversations, messages and search wording to identify pre-purchase questions. Keep a fixed question set for comparable future tests.
2. Keep the baseline measurement. Record the answer, date, time and links for each question. The author saved timestamps and processed data programmatically. A log separates observed change from impressions.
3. Separate the two indicators. Track source pages separately from companies named in the text. Extract mentions from answers rather than memory. Automated processing in the study found five omitted participants; adding them moved one metric by four percentage points. An incomplete list can distort assessment.
4. Define each platform's role. Review source types and domain frequency. Form hypotheses about owned content, company information and external publications, then combine the tasks into one plan rather than isolated activities.
5. Check again. Repeat the same questions after changes and compare sources and mentions. One appearance does not establish lasting visibility.

Visibility is not a commercial outcome. The study measures answers and sources, not enquiries, revenue, profit or return. Attribution to generative search requires separate traffic, lead and sales evidence.

The next step is to audit channels and measurement: what is tracked, what is missing and how new observations fit ongoing marketing.