Connected Apps in AI Mode: When Search Acts Inside Software You Do Not Own

The Search Box Just Grew Hands

On July 16, Google started rolling out connected apps in AI Mode, and it is a bigger deal than the muted announcement suggested. The first partners are Instacart, Canva, and YouTube Music. Link an account once, and AI Mode can read from and act inside that app while it answers your query: build a grocery cart, pull design templates, generate a playlist, all without leaving Search.

I have been saying for two years that search was moving from answering to doing. Connected apps in AI Mode is the moment that shift stops being a slide in a keynote and starts touching your revenue. Search no longer just recommends a product and sends a click. It can now complete the transaction inside a third party's software, on a screen you never see.

Numbers over noise: this is not another SERP feature to chase. It is a structural change in where the buying decision happens. Let me explain what actually changed, why it matters, and what to do about it before your category gets connected.

What Actually Changed

Three things are new here, and only one of them is the apps.

One: the selection moment moved. When AI Mode assembles a grocery cart in Instacart, something chooses which brand of olive oil goes in. That decision now happens inside a connected app, in a context you cannot observe with your analytics and cannot influence with a landing page. The picking, not just the answering, is the game now.

Two: it stacks on Personal Intelligence. Earlier this year Google launched a personalization layer that lets AI Mode tap a user's own Google apps, like Gmail and Photos, for tailored responses. Connected apps ride on top of that. A barbecue on your calendar becomes a shopping list; a purchase history shapes which products get recommended. The answer is personalized before your brand ever enters the conversation.

Three: the funnel compressed to one surface. Awareness, consideration, and conversion used to be separate steps across separate pages. Now they can collapse into a single AI Mode session. This is the same dynamic I wrote about in Google completing the task instead of sending the click, except it now reaches into commerce apps you have no login to.

Worth being honest about the limits. This is a US rollout, YouTube Music is Google's own product, and the commercial terms with the true third parties are undisclosed. Treat the economics as an open question, not a settled one. But the direction is not ambiguous, and Google said more partners are coming.

Why This Should Worry the Digital Shelf Crowd

If you sell through platforms you do not own, you already understand the core problem: your visibility depends on someone else's ranking logic. That is the whole story of the digital shelf. Connected apps in AI Mode extend that problem into the search box itself.

Here is the uncomfortable part. When AI Mode builds a cart inside a connected app, you are competing on:

• The connected app's own ranking and recommendation logic
• The user's purchase history and stated preferences
• Whatever product data the app holds about you, accurate or not
• Google's synthesis on top of all of it

None of those levers are your website. Your beautifully optimized product page might never get rendered. This is the zero-click problem with a checkout attached. Traffic does not just disappear; the sale can close entirely off your property, and you find out from a settlement report weeks later.

The Connected Answer Audit

You cannot optimize a surface you cannot see. But you can make sure your brand is structured to be selected wherever the selection happens. Here is the checklist I am running with teams right now. I call it the Connected Answer Audit, and it has five parts: FEEDS.

F: Feeds first. Your product feed is now a ranking asset, not a compliance chore. Wherever a connected app pulls inventory, price, and attributes, that feed is your storefront. Audit it for completeness, freshness, and attribute coverage before you touch anything on your own site. A stale feed is an invisible brand.

E: Entities, not just pages. AI Mode reasons about things, not strings. If your brand, products, and their relationships are not clearly defined as entities, the model has to guess. Tighten your structured data and your knowledge graph presence. I laid out the mechanics in optimizing for things, not strings, and it matters more now than when I wrote it.

E: Endpoints and eligibility. Find out how a brand becomes available inside each connected app. On Instacart that means retailer and catalog relationships. On other partners it will mean API connections, product feeds, or partner programs. The brands that get connected early will own the default answer while everyone else waits for a rollout.

D: Data you actually own. The more this moves off your property, the more your first-party data becomes the thing you can still control and measure. Email, loyalty, direct relationships: these are how you reach a customer the model chose for someone else. If you have not built that muscle, start with the post-cookie first-party data playbook.

S: Signals from the assist. You will not see the in-app selection, but you can still measure the assist. Watch branded search, direct app engagement, and post-session conversion patterns. This is measurement in a world where the click is not the event, which I broke down in measuring SEO when the clicks fall.

Run all five. The order matters: feeds and entities are table stakes, endpoints decide whether you are eligible at all, and data plus signals are how you survive not owning the surface.

What I Would Not Do

A few reflexes to resist, because I am already seeing them.

• Do not panic-build an "AI Mode strategy" deck. This is your existing product data and entity work, done properly, aimed at a new surface. It is not a new discipline. It is the old discipline with higher stakes.
• Do not assume your category is safe. Groceries, design, and music are the beachhead, not the map. If your product can be listed, carted, booked, or subscribed to, plan as if you are next.
• Do not confuse presence with control. Being available inside a connected app is not the same as controlling how you are presented. You will influence the inputs, not the output. Build accordingly.
• Do not neglect the destination. When a user does break out of AI Mode to finish on your site or app, that experience has to be flawless. The fewer sessions you get, the more each one has to convert.

The Real Shift

For twenty years, SEO meant earning a position in a list of links and then earning the click. Connected apps in AI Mode quietly retire that model for a growing set of queries. The new job is to be selectable: structured cleanly enough, fed accurately enough, and connected widely enough that when an AI assembles the answer inside someone else's software, your brand is the obvious pick.

That is harder than ranking, because you are optimizing for a decision you never witness. But it is the same core work I have argued for all along: clean data, clear entities, real first-party relationships, and honest measurement of the assist rather than the click. The teams that already did that work will be fine. The teams still counting sessions as their only proof of value are about to have a rough couple of quarters.

If you are trying to figure out where your catalog sits in this shift, and whether your feeds and entities are ready to be selected instead of just ranked, the channel is open by introduction. Bring your product data. That is where the real conversation starts.