Google Search can now finish a task instead of pointing at it. On July 16, Google rolled out Connected Apps inside AI Mode, letting a search conversation add groceries to an Instacart cart, pull Canva templates, or save a YouTube Music playlist without the user visiting Instacart, Canva, or YouTube first (The Verge; TechCrunch). That is not a feature update. That is the search result learning to act.
The incumbent way
The incumbent way treats Search as a hallway. A query comes in, a ranked list of doors goes out, and the brand's job is to be one of the doors and make the walk from click to conversion as short as possible. Every landing page, every bid strategy, every attribution model built for the last decade assumes a person clicks through to a website before anything commercial happens.
AI Mode's Connected Apps breaks that assumption at the point of highest commercial intent. Plan a barbecue in AI Mode, connect Instacart, and the ingredients land in a cart without a search result ever loading in a browser tab. Ask for flyer ideas, and Canva templates appear inside the conversation. Build a playlist, and it saves straight to YouTube Music. Google is not sending the user to the merchant's storefront. Google is assembling the cart and handing it off already built (TechCrunch).
The cost of standing still
Miss this and the first cost is invisible measurement. Your analytics stack counts a session, a click, a landing page view. A Connected Apps transaction produces none of those signals for the brand on the other end. Instacart sees the cart. Your dashboard sees nothing, because nothing happened on your property.
Miss it for a full quarter and the second cost is budget misallocation. Paid search bids on the click. If the click volume for a category keeps shrinking because the task finishes inside the AI Mode conversation, a media plan anchored to cost-per-click is bidding on a shrinking fraction of the category's real commercial activity, and nobody notices until the quarter-over-quarter numbers stop making sense against category growth.
Miss it past that and the third cost is structural. The account-level opt-in model that Google built for Connected Apps means a consumer connects an app once and then every future AI Mode session can transact against it without a fresh authorization step (Tech Times). Once your category's buyers have connected their preferred merchant, the switching cost for a competitor to interrupt that flow gets higher, not lower, every week the connection sits unchallenged.
What the evidence shows
Three named partners went live at launch: Instacart, Canva, and YouTube Music. All three are consumer-facing, high-frequency, low-consideration categories, which is exactly where an agent can complete a task without much risk of getting it wrong. Google has not published a timeline for adding B2B, enterprise, or considered-purchase categories, and has not disclosed the commercial terms behind any of the three launch partnerships (AI Insiders).
That gives marketing directors a narrow but real evidence base, not a hypothetical one. The mechanism is confirmed and live in the US as of this week. The category expansion is not yet confirmed. The right response is not to panic about your specific category being absorbed tomorrow. The right response is to treat this as the opening move in a pattern, because the same week Google made this change, it also renamed NotebookLM to Gemini Notebook and extended cloud compute and code execution to AI Pro subscribers, another step in turning Google's AI products from things that answer into things that act (Tech Times). One agentic commerce launch is a pilot. A pilot arriving in the same week as a second agentic capability expansion is a direction.
Layer in the separate claim Google made the next day, on July 17, that AI features in Search now send "billions of clicks to websites every week," offered with no baseline, no methodology, and no comparison to prior periods (Search Engine Journal; PPC Land). Independent measurement cited alongside that claim shows click-through rate declines on pages sitting under AI features, not increases. Google is simultaneously building the mechanism to keep transactions inside its own surfaces and asserting, without data, that the click economy is healthy. Believe the mechanism you can observe. Treat the unverified claim as marketing, because that is what it is.
The binary marketing directors now face
That is not a traffic story. That is a measurement and budget story with two paths.
Path one: keep every KPI anchored to clicks and sessions, and let a growing share of category-relevant commercial activity happen in a system your dashboards cannot see. Your reporting will look stable right up until it doesn't, because the erosion shows up in category growth rate first and in your own funnel numbers last.
Path two: build a task-completion visibility audit into this quarter's search strategy. That means testing whether your product or service category is one an AI agent could plausibly complete a transaction for today, mapping where your brand shows up (or doesn't) inside connected-app flows once your category is added, and pressuring your analytics and attribution vendors now for a way to detect agent-initiated conversions before that detection gap becomes a reporting crisis instead of a reporting question.
Verdict
Search spent two decades as a hallway to your storefront. It became a system that can build the cart itself. The brands that win the next version of search visibility are the ones auditing for agentic task completion this quarter, not the ones waiting for their category to get named in Google's next blog post.
Magnet builds search visibility and measurement systems for the version of search that exists now, not the version from three years ago. If your reporting can't yet see agent-initiated demand in your category, talk to Magnet about a visibility and measurement audit.
