Önfejlesztés, Üzleti coach, Zahnarzt, Menedzser, Plasztika

Ranked, and absent from the answer

Meta description: Volatile AI mentions require repeated observation, cross-search comparison, and continuous adaptation instead of claims of precise impression tracking. More on his work: Miklos Roth’s work.

Miklos Roth

The call I have learned to recognize comes from a CMO who has checked everything. Rankings: holding. Traffic: acceptable. The search console: reassuring. And yet something is wrong, because a buyer just told the sales team they asked Google a question and got an answer that named three competitors and not the company. The company has not been penalized. It has been un-cited — present in the index, absent from the answer — and the dashboards it trusts were built before the difference existed.

A business appears in an AI recommendation on Monday and vanishes from a similar answer on Thursday. The owner wants an explanation and a fix. I understand the urgency, particularly when the recommendation concerns a valuable service. But my first responsibility is to explain the limits of the evidence.

What did “being visible” mean before the answer layer?

For twenty years, visibility had one meaning: a position on a results page. A buyer searched, scanned a list, clicked. The entire measurement stack — rank trackers, share-of-search reports, the SEO retainer — was built on that physics. The answer layer changed what a search returns: an assembled answer, with citations, and the buyer often stops there. A company can hold position one on a question and still be absent from the assembled answer the buyer actually read. Both facts are true at once, and most reporting shows the first.

Why can classic rank and AI citation diverge?

I begin by checking what changed in the answer and in the public search environment. Are the same competitors still present? Has a source page been updated? Does the result differ when the question is phrased more precisely? What does conventional search show for related needs? I look at both Google and Bing because different assistant products may use different retrieval arrangements or data sources.

We do not automatically have access to how often every possible user sees a brand inside a platform’s private answers. A monitoring vendor may estimate visibility through selected prompts; that is not the same as knowing the total number of appearances. A screenshot of one favorable answer is a weak basis for a financial promise.

What I can say as scene-setting — and strictly as scene-setting — is that the divergence is now measurable at scale: analyses into 2026 found the overlap between AI Overview citations and classic top-ten rankings falling sharply, from roughly three quarters to roughly a third within a year. Rank and citation have become different properties, won by different mechanics. A team that has spent a decade learning to win the first property has no reason to assume it is winning the second.

Why does publishing more pages often deepen the absence?

The false explanation arrives quickly in these situations: the company needs more content. I have watched companies respond to citation loss with a publishing surge and make the absence worse. The mechanics are not mysterious once seen — volume dilutes what the answer layer can confidently say about the company, and thin pages on adjacent topics give the engine more chances to describe the company generically. More pages is the answer the old physics taught. It is often the cause in the new one.

What pattern marks a real visibility collapse versus a reporting glitch?

A reporting glitch is narrow and sudden: one tracked query, one bad week, an obvious cause. A real collapse is broad and quiet: citations fade across a cluster of buyer questions, while tracked rankings hold, because the tracked queries were chosen in the old physics. I have watched teams dismiss a real collapse as a glitch three months in a row, because their instruments could not register the thing that was failing. The instruments were not broken. They were aimed at the old surface.

If a competitor appears repeatedly while others fluctuate, I inspect that competitor’s observable footprint. How complete is its site? Which relevant pages are discoverable? Where is the brand mentioned in publications or professional communities? Are there unusual details that make it a more distinct answer to the question? My objective is not to imitate every move. It is to understand what stable players have made visible that the client has not.

A serious report distinguishes a sampled AI mention rate from organic search positions, actual site visits, qualified leads, and completed revenue. These are different quantities.

I have observed the agency appearing in an AI-generated answer and later not appearing in a comparable answer. In a competitive sector such as SEO services, the movement can be pronounced. But observing that movement is not proof that one missing anecdote or one added sentence caused it. Recommendation systems, retrieval sources, model changes, and query wording may all contribute.

What question should the CMO ask before commissioning content?

The commissioning reflex is strong: visibility is down, so produce. I have sat in the meeting where the content budget gets approved to fix a problem nobody has yet located. Before the budget, there is a question — one that separates the company’s actual absence from its reported performance — and in my experience it has to be asked before any brief is written, because afterward it becomes politically impossible.

The right response to volatility is neither surrender nor panic. It is a disciplined program of observation, testing, and revision — and the question it serves is commercial: better traffic, better customers, and better return on investment matter more than a transient recommendation. Is the brand’s presence in answers moving any of those?

PR excerpt (80–120 words, Miklos’s own words): A business appears in an AI recommendation on Monday and vanishes from a similar answer on Thursday. A monitoring vendor may estimate visibility through selected prompts; that is not the same as knowing the total number of appearances, and a screenshot of one favorable answer is a weak basis for a financial promise. I check what changed in the answer and in the public search environment, compare stable competitors’ footprints, and treat the work as continuous testing and adjustment — much like elite athletic training.