![\"On](\"https:\/\/www.lukaswojcik.com\/blog\/wp-content\/uploads\/diagrams\/advisor-sprache-en.png\")
Two names for one thing<\/h2>\n
The announcement calls it Analytics Advisor<\/em>, paired with an Ads Advisor<\/em> on the campaign side. The help page calls it Ask Advisor<\/em> and marks it Beta.<\/p>\n This is a small thing that costs real time. A search for the announced name finds blog coverage and the launch post; a search for the documented name finds the page that states the restrictions. Anyone trying to establish what the feature actually does, rather than what it was said to do at launch, needs the second name, and nothing in the first one points to it.<\/p>\n The documentation states that the assistant is available in properties with English as the selected language, with expansion to other languages planned. That single sentence is the whole availability rule, and its subject is worth reading twice.<\/p>\n It is not the browser language. It is not the interface language of the person asking. It is not the language of the site being measured or of the event names inside it. It is the language selected on the property<\/em>.<\/p>\n For a property configured in German or Polish, the assistant is therefore not present at all \u2014 not empty, not limited, absent. And because the property language is a shared setting rather than a personal one, the obvious workaround has a shared cost: switching it to English to gain the assistant changes the interface for everyone who opens that property, along with the language of scheduled exports and anything else that follows the setting.<\/p>\n That trade is worth stating plainly rather than discovering. On a property with one owner it is a preference. On a property that several teams or several agencies work in, it is a change that belongs in a conversation before it happens.<\/p>\n The documented range is broader than a query builder and narrower than an analyst.<\/p>\n It is reachable from an icon in the top right of the property and from the search box, which means it sits alongside navigation rather than inside a report.<\/p>\n The documentation is explicit that the assistant processes data solely at the property level. Everything answerable is answerable from inside one property, and nothing crosses the boundary to another.<\/p>\n For a single-property setup this is invisible. For a group running one property per market or per brand, it is the limit that decides how much of the routine reporting it can absorb: any question of the form “which of our markets moved” has to be assembled outside it, in exactly the way it was before.<\/p>\n The page states that chat activity may be used to improve the product, and that use is subject to Google’s terms, AI use policy and privacy policy. That is a normal disclosure and an unremarkable one \u2014 but it applies to a surface where the input is typed prose rather than a dimension picker, and that changes what ends up in it.<\/p>\n A report request contains dimension names. A question contains whatever was typed: a client’s name, an unannounced campaign, an internal target, the reason a figure is being checked. The data in the property is governed by the property’s own agreements. The question about that data is a separate artefact with a separate disclosure attached.<\/p>\n The habit that follows is simple and costs nothing: ask about the metric, not about the circumstances. Why did key events fall last week<\/em> gets the same answer as the version that names the client and the reason for asking, and only one of the two puts those into a chat log.<\/p>\n The label has outlasted the rollout by three quarters of a year, which suggests it describes the state of the capability rather than the state of the release. The page says as much in its own way, noting that the assistant may have capability limitations without listing them.<\/p>\n An undocumented limitation set is the practical problem here, not the label. A question that cannot be answered will be answered anyway, in prose, because prose is the output format. The check is therefore the same one that applies to any generated figure: a number worth acting on is worth reproducing in a report before it is acted on. That is not distrust of the tool. It is what makes it usable at a point where nobody can say precisely where its edges are.<\/p>\n","protected":false},"excerpt":{"rendered":" The Gemini assistant went to all English accounts in December 2025. Nine months on the help page still says Beta, calls it by a different name than the announcement did, and names one setting that decides whether it appears at all.<\/p>\n","protected":false},"author":1,"featured_media":15515,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[4],"tags":[91386,91258,91252],"class_list":["post-14882","post","type-post","status-publish","format-standard","hentry","category-digital-analytics","tag-marketing-cyfrowy-pl","tag-google-analytics-4","tag-web-analytics"],"yoast_head":"\nThe setting that decides whether it exists<\/h2>\n
What it answers where it exists<\/h2>\n
Generic performance \"How is the site doing\"\n\nFully specified a named metric over a named\n dimension and period\n\nDiagnostic \"why\" questions - what drove a\n rise or a fall\n\nHow-to questions about the product's\n own features\n\nConfiguration retrieving how the property is\n currently set up\n\nRecommendations optimisation suggestions<\/code><\/pre>\nThe boundary is the property<\/h2>\n
The sentence about chat activity<\/h2>\n
What Beta means after nine months<\/h2>\n