A search system can hold an accurate record of who you are and still never put you forward when someone describes the problem you solve. Ask it your name and the answer comes back correct. Ask it the question a prospective client would actually type, and you are not in the room at all. These are two different states of visibility, and the distance between them is where most of the current work on professional presence is done.
Oleksandr Koliesnikov advises local businesses in Canada and the United States on how they are found by search engines and by the assistants that increasingly answer instead of them. He measured that distance on himself before he tried to close it, and then published the measurement. The decision is less ordinary than it sounds, because a starting measurement in this field exists to record the moment when nothing was working yet.
The measurement he ran on himself
On 27 July 2026 he put 12 customer questions to ChatGPT, Perplexity, Gemini and Google AI, in English, Ukrainian and Russian. The questions were the ordinary commercial ones a buyer asks before hiring anybody: who offers this service in Toronto, who can help a Canadian business get recommended by an assistant, whether a local company should invest in one discipline before the other. The run produced 48 answers across the four systems, of which 44 were commercial and carried no brand name in the question.
Neither his name nor his site appeared in any of them.
The branded question behaved differently, and that contrast is the useful part of the exercise. Asked directly who he was, three of the four systems returned a recognisable description of him; one selected a different person with the same surname, and another blended a namesake into the profile. The record existed and was largely accurate. It simply was not attached to any question a client would ask.
Known by name, not connected to the work
What the systems held was a fair summary of a career: digital marketing since 2010, paid advertising, lead generation, local search, automation. What they did not hold was the narrower commercial claim he now makes, which is that this experience is currently pointed at one thing. The older and broader description kept winning because it was better supported everywhere the systems looked.
Being findable by name is a record. Being named in answer to a commercial question is a conclusion, and a conclusion has to be assembled from sources the system trusts.
He describes the underlying condition as ordinary rather than exotic. His work had accumulated across client sites, a separate service, social profiles and several transliterations of his own surname, and it was legible to any human being who already knew the context. A machine collecting facts from unconnected pages had no reason to resolve them into one practitioner with one current specialism. The domain koliesnikov.com had existed since June 2024, so the gap was never about the absence of an address.
What changed after the measurement
The rebuild came after the reading, not before it, and that order matters more than any single change he made. He rebuilt the site twice, settled on one canonical spelling of his name, connected his official profiles to it, added structured data describing the relationship between the person, the site and the services, and gave the cases and the speaking work a fixed place in that structure. Older capabilities stayed on the site as supporting experience rather than as competing headlines.
His own account of the mistake behind the first attempt is worth repeating, because it is the common one. He tried to preserve the full breadth of what he could do while establishing a narrower commercial focus, and the breadth kept obscuring the focus. A business does not need to erase its history, but it does need to make the current priority unmistakable to a reader who is assembling facts mechanically.
The same habit shows up earlier in his career. After several unsuccessful advertising launches for a windows and doors company he stopped treating a campaign as a self-contained object and began examining geography, offer, landing page, lead quality, analytics and response speed together, because a weak result at the end of the chain never points to the link that produced it. The system built afterwards eventually ran at more than 1000 leads a month.
First appearances, and what they do not yet prove
On 12 August 2026 he ran spot checks in a fresh session and recorded two results. Google’s AI Overview listed him second among the people it named for a query about an AI visibility specialist in Toronto, and cited his site among its supporting sources. ChatGPT included him in an answer about a Toronto marketing consultant who speaks Ukrainian, though its description still led with search advertising and lead generation rather than his current specialism.
He states plainly that these were two spot checks and not a repeat of the original run. That distinction is the reason the case is worth reading at all. Two favourable answers are an early signal about direction; they are not a measurement, because they do not cover the same questions, the same systems or the same conditions as the reading they are being compared against.
The check that would settle it
The test that would settle it is available to anyone, and it is deliberately unglamorous. Take the commercial questions your buyers actually ask, put them to the same systems in the same conditions, count how many answers name you, and then repeat the identical run after the work is done. Keep the branded question separate from the commercial ones, because a system that recognises your name tells you nothing about whether it will recommend you.
Koliesnikov holds a defensible position on what a consultant can promise here: current visibility can be measured and changes in it can be observed, while the answer a given system produces on a given day stays outside anyone’s control. He now has the one thing that makes such a claim checkable by a stranger, which is a published starting point with a date on it. What he does not yet have is the second reading.
Updated 25 August 2026.
