Google Information Agents: What the Citation Stability Data Says

Google announced "information agents" at I/O 2026. These are persistent background processes that monitor the web on behalf of subscribers and push synthesized briefings -- with source links -- directly to users without requiring a query. They went live June 12 for Google AI Ultra subscribers and are rolling out to AI Pro subscribers through this summer.

Every new AI surface generates the same question from the businesses we audit: do I need a new strategy? Information agents are different enough from Ask Maps, Ask for Me, and AI Mode that the question is reasonable. They operate on push logic rather than pull. No user types a query. The surface delivers citations you didn't know were being requested.

The short answer is: not yet. Here is exactly what would change that -- and what the largest AI citation tracking study ever run tells you about which infrastructure actually holds when new surfaces launch.

What Google Information Agents Actually Are

Information agents are not a new AI search interface. They're closer to a monitoring service built on AI Mode infrastructure. A subscriber sets standing topics -- competitor news, market trends, regulatory changes -- and the agent monitors web content on their behalf, periodically sending a synthesized brief with links to the sources it drew from.

The citation mechanism converts to referral traffic the same way an AI Mode or AI Overviews citation does. A source link in a push notification is a source link. If your business or content appears in an agent's brief, you receive a referral visit. The question is who's receiving those briefs and what topics trigger a local business to appear.

Those two questions are why information agents aren't in Sourcepull's audit scope today.

Why We're Not Auditing This Surface Yet

In our July 31, 2026 investigation of information agent scope (`methodology-recs/2026-07-31-google-information-agents-audit-scope.md`, Scout session 93), we identified three structural reasons this surface isn't actionable for most of our clients' target buyers right now.

First, access is gated. Information agents require a paid Google AI Ultra or AI Pro subscription. The homeowners and business buyers asking "who should I hire for X in Y city" are predominantly free Google Search users. The consumers looking for a plumber, an HVAC contractor, or a family lawyer are not the subscribers using information agents in summer 2026. That may change as paid AI subscription adoption grows, but the addressable consumer population on this surface is small today.

Second, the query format doesn't match. Information agents are designed for standing monitoring topics -- things you want to track over time. Local service discovery is transient: someone needs a plumber this week and won't think about plumbing again for two years. Whether that query type ever becomes an agent topic that triggers local business citations is unknown.

Third, there's no way to test it from the outside. Agent citations are delivered via push notification to individual subscribers based on their standing topics. There is no query we can run to simulate this surface. Our standard B-query methodology doesn't cover it.

The conditions that would change this: free-tier user access to information agents, or client analytics starting to show "google-agent" as a referral source in GA4. When either appears, information agents join the audit model.

What the Largest Citation Tracking Study Says About New Surfaces

New surfaces create the instinct to build new citation strategies. The SISTRIX longitudinal study, documented in our July 30, 2026 methodology rec (`methodology-recs/2026-07-30-citation-drift-audit-validity.md`, Scout session 92), argues against that instinct.

The study tracked 82,619 prompts across 1,548,213 snapshots over 17 weeks, across three platforms and six countries. The headline numbers are more aggressive than most practitioners expect.

Google AI Mode rotates 56% of its source domains every week. ChatGPT Search rotates 74% of its sources weekly. That's not monthly drift -- it's the majority of the citation pool reshuffling every seven days.

Model transitions accelerate this further. When GPT-5.5 launched, it shifted 47% of ChatGPT citations within 48 hours. When Gemini 3 launched in January 2026, it replaced 42% of AI Overviews cited domains overnight. A model update alone -- no query behavior change, no content change on any cited site -- can move nearly half the citation pool in two days.

The finding that matters for infrastructure planning: brand domains held for all 17 weeks in 43% of tracked cases. Co-citations -- the surrounding sources cited alongside brand domains -- rotated at 89% weekly. Brand domain presence was roughly eight times more stable than co-citation presence across the full tracking window.

The implication is direct. Infrastructure that drives brand domain presence survives model transitions and weekly rotation cycles. Tactics that depend on co-citation position -- appearing alongside the right sources at the right moment -- are inherently fragile. The 89% weekly co-citation rotation rate means that position is essentially rebuilt from scratch every seven days.

The Platform Architecture That Determines What Holds

What drives that brand domain stability across platforms? The Yext 2026 AI Visibility study, which analyzed 17.2 million citations across ChatGPT, Perplexity, and Gemini (documented in `knowledge/platform-citation-behaviors.md`), identified the source architecture per platform.

ChatGPT trusts third-party directories for 49% of its citations. Gemini trusts brand-owned sites for 52% of its citations. Perplexity prioritizes niche expert directories specific to the business category. Each platform weights different inputs, but none sources citations primarily from one-time content pushes or surface-specific tactics.

Directories are indexed consistently. Brand sites with structured schema are indexed consistently. These are the inputs that stay in the citation pool when the 56% weekly rotation runs its cycle. They're also the same inputs that would drive information agent citations if the surface expands to free-tier users -- because information agents draw from the same underlying source infrastructure as AI Mode.

The platform may be new. The citation source architecture is not.

What This Means for Audit Cadence

The 56% weekly rotation rate from the SISTRIX study has a direct implication for how often AI visibility results need to be measured.

Our July 30 rec flagged an audit framing problem: a point-in-time snapshot measures a citation pool that is already 56% different from the prior week's pool. An audit result from 90 days ago reflects a citation reality that has mostly turned over. The businesses that appear as "invisible" in that audit may have entered the pool since it ran. The ones that appear as "cited" may no longer be.

This supports a 60-90 day re-audit window for businesses that have made infrastructure changes -- enough rotation cycles for consistent infrastructure signals to establish stable presence, but not so long that the snapshot becomes misleading.

A Sourcepull Signal Check gives you a current read on where your business appears across platforms today. Given the weekly rotation rates the SISTRIX study documented, "today" is the only time period the result is guaranteed to reflect the actual citation pool.