Your Brand Shows Up in One AI Answer and Not the Next. Spot Checks Miss It.

Ask a Head of Ecommerce how they track whether their brand shows up in AI answers and the honest version is usually the same: "the only way we're tracking that is when we're asking for it." Someone types a question into ChatGPT or Google AI Mode, sees a result, and that result becomes the team's picture of AI visibility until someone checks again.

The problem is that AI answers move. A product that is recommended in one answer can be absent from the next, and the shopper never ran the query twice. Google's own Search agents, released in June 2026, make that point in the most concrete way possible: Google now re-runs shopping queries on a loop for its users, without anyone asking.

What Google shipped

Information agents, the first category of what Google calls Search agents, launched in AI Mode for Google AI Ultra subscribers on June 12, 2026. According to 9to5Google's coverage of the release, a user triggers one by including "keep me updated on" or "alert me when" in an AI Mode prompt. The agent then watches for changes tied to that request and sends an update when something relevant happens.

Search Engine Journal reported that the agents monitor topics in the background and send updates with links back to the web, available in all AI Mode languages and markets, with wider access planned for later in the summer. Google's I/O 2026 announcement describes them operating "in the background 24/7," looking across blogs, news sites and social posts, plus real-time data on finance, shopping and sports.

The shopping example Google used is the one brands should sit with. A shopper asks to be kept updated when a favorite athlete announces a sneaker drop. That is a standing demand signal pointed at product data, evaluated repeatedly, against whatever competing products exist on each pass. If your product is not readable when the agent checks, you are not in the alert.

Why your visibility drifts between queries

A one time check captures one moment. Four things move underneath it.

Your own product data changes. Prices update, variants go out of stock, a feed resubmission drops an attribute, a template change buries a spec in an image. None of these feel like visibility events, and every one of them changes what an engine can retrieve.

Competitors change. A rival that fills in fit, materials and use case on its pages this month can take your place in the answer next month. Nobody on your team ran the query that would have shown it.

The models change. Answer engines update how they retrieve and grade, and the same question can produce a different shortlist after an update even when nothing on your side moved.

And the fan-out changes with phrasing. Google describes AI Mode as breaking a question into subtopics and issuing many queries at once. Two shoppers asking for the same thing in different words trigger different sub-questions, and your page may answer one set and not the other. We cover that mechanism in query fan-out for commerce.

How an agent decides which products make the alert

An agent monitoring "alert me when a good running shoe under 150 dollars restocks" does not match on the literal phrase. It breaks the request into facets such as price, category, use case and availability, then checks each against the product data it can reach, on every pass it makes.

Your product has to be retrievable for each facet, not only the headline term. A listing that states fit, materials and use case in plain text can match an occasion based or constraint based request. A listing that says little beyond a title and a price gives the agent nothing to confirm. The completeness of your product content decides whether the agent can include you with confidence when it writes the update.

The operating model has changed as a result. In the old model a shopper ran one query, saw one answer, and your visibility was whatever it was at that moment. A single audit described reality well enough.

With agents in the loop, the query is re-run continuously and the answer is re-evaluated each time. A static audit goes stale between passes. Winning the moment matters less than staying in the answer across every pass.

What tracking looks like when it is not a spot check

You do not need to watch everything. You need a fixed, repeatable read that shows change. In practice that means:

1. A fixed set of shopper questions. Written from how real customers in your category ask, including the "keep me updated on" and "alert me when" phrasings a Google agent would use, and held constant so week to week comparisons mean something.
2. A regular cadence. Weekly reads across ChatGPT, Google AI Mode and Perplexity catch drift early enough to act on it. A quarterly check mostly tells you what you already lost.
3. Three outcomes, not one. For each question, record whether you appear as a product card, as a mention in the text, or not at all. A brand that moves from card to mention has lost something even though it still "shows up".
4. Two or three named competitors on the same questions. So you see the month they pull ahead and can look at what changed on their pages.
5. A before and after for every fix. When a product content or feed change ships, the same questions get re-run, and the result is attached to that change. That is how you prove it worked.

Frequently asked questions

What are Google's Search agents?

Search agents are AI Mode agents that work in the background. The first type, information agents, monitor blogs, news, social posts and real-time shopping, finance and sports data, then deliver a synthesized update when something relevant to the user's request changes.

How does a shopper trigger one?

In an AI Mode prompt, the shopper includes phrasing like "keep me updated on" or "alert me when." The agent then monitors the web and Google's real-time data for changes tied to that request and notifies the user.

Is this the same as Merchant Center AI Performance Insights?

No. AI Performance Insights is a merchant facing report on how your products perform across Google's AI surfaces, which we explain in what Google AI share of voice really means. Search agents are consumer facing and run on the shopper's behalf. Both point to the same need: knowing how often you show up, over time, not once.

How often should we check whether we show up?

Often enough to see change before it costs you. For most brands that is a weekly read against a fixed question set, with a deeper look whenever product data, a feed, or a page template changes. The point is a trend line you can act on, not a single score.

Where Paz fits

Google checks shopping data on a loop. You can see your own visibility on the same cadence without adding a dashboard to someone's week. A Paz expert writes the shopper questions in your category, finds the gaps in your product information where you drop out of the answer, and fixes them from your own sources. You approve each change, and the platform publishes it and tracks what shows up afterwards, week after week. If you would rather know than guess, talk to us.