Most “AI is eating search” coverage treats search as a uniform substance that’s shrinking by some percentage per year. That framing is wrong in a way that matters.
Search isn’t shrinking uniformly. Specific kinds of queries are leaving Google for AI assistants, and the kinds that are leaving first are the ones with the highest commercial intent. The categories that move slowly (navigational queries, brand-name searches, simple informational lookups) are still mostly on Google. The categories that move fast (comparison queries, multi-constraint product searches, “best X for Y” queries, planning queries that span multiple products) are meaningfully shifting in some buyer cohorts.
This matters because not all queries are commercially equal. The query “ChatGPT vs Claude for coding” has a much higher chance of producing a paying customer than “what time does Target close.” If the high-intent queries are the ones leaving Google fastest, the share of Google traffic that converts is going to fall faster than the share of Google traffic overall. The merchants who haven’t done this math yet are about to.
This post is the structural argument behind the thesis. What’s leaving Google, what isn’t, why the structural fit drives the split, and what it means for merchant acquisition allocation.
What’s leaving Google fastest
Three query shapes account for most of the “leaving Google” pattern in commercial categories:
Comparison queries. “Best X for Y.” “X vs Y.” “What should I buy if I want both A and B.” These queries have a structurally bad fit with the SERP. The buyer wants a recommendation with reasoning. Google gives them ten links. AI assistants give them a synthesized answer. The shift in query share for this shape is the largest of any category we observe, and the rate is climbing. Younger buyers move first; the demographic curve is steep.
Multi-constraint queries. “Under $200, ships before Tuesday, available in size M, leather alternative.” These queries have always been awkward on Google because the buyer is asking the surface to filter against a set of constraints simultaneously. The SERP doesn’t filter. The buyer has to. AI assistants do the filtering. The shift toward agents for this shape is roughly as fast as for comparison queries and probably accelerating, because the constraint-stacking gets harder as the buyer’s preferences become more specific.
Planning queries. “Help me put together an outfit for a fall wedding.” “I need everything to set up a home office for $1500.” “Plan a backpacking trip in Glacier.” These queries weren’t really a Google category. The buyer had to break them into sub-queries and assemble the answer themselves. AI assistants can handle the whole task. The category is growing, not just shifting from Google, because the demand for “help me plan this” has been latent and the agents are unlocking it.
These three shapes together account for a substantial share of the high-commercial-intent queries in most categories. They’re also the shapes that convert best when they do route through agents.
What’s not leaving Google (yet)
The categories that have stayed on Google are the ones where Google’s surface still has a structural advantage:
Navigational queries. “Acme Outfitters” or “Carrigan jacket.” The buyer knows what they want and is using search to navigate. Google’s surface is fast and the answer is a link, which is exactly what the buyer wants. Agents don’t add value here.
Brand-and-product queries. “Acme Outfitters return policy” or “Carrigan jacket size guide.” The buyer wants a specific page on a specific brand’s site. Google’s surface is still the fastest path. Agents can answer these but the buyer’s habit and the agent’s added latency don’t favor the shift.
Local queries. “Pizza near me.” “Coffee shops open now.” Google’s local pack and Maps integration are still strong. The agents that can answer these well (Google’s own products, primarily) are integrated tightly with the same surface that Google already owned.
Pure informational queries. “When did the Civil War end.” “What’s the population of Indianapolis.” Google’s snippet and Knowledge Graph give the answer immediately. Agents add some friction without much benefit.
The pattern across these categories is that Google’s existing surface fits the query well. Where it fits well, it stays. Where it doesn’t fit well (comparison, multi-constraint, planning), it loses ground.
The structural fit argument
The structural reason for the split is that Google was built for a specific kind of query and AI assistants are built for a different kind of query.
Google’s surface is optimized for “find me the best page that answers this.” It ranks pages, presents them, and lets the buyer navigate. The unit of value is the page. The reward function rewards pages that satisfy the buyer well enough that they don’t go back to the SERP.
AI assistants are optimized for “give me the answer.” They synthesize across many sources, present a recommendation, and let the buyer act on it. The unit of value is the answer. The reward function rewards answers that the buyer trusts and acts on.
Both reward functions are valid. They’re optimal for different query shapes. The mismatch happens when buyers use Google for a query whose answer-shape doesn’t fit a ranked list, or when they use an AI assistant for a query whose answer-shape is a single canonical link they already know.
In 2010, the buyer mostly didn’t have a choice. Both kinds of queries went to Google because Google was the option. In 2026 the buyer has a choice. The choice is being made rationally: the surface whose answer shape fits the query wins. Comparison queries and multi-constraint queries fit AI assistants. Navigational and brand queries fit Google. The split is consequential and the trend is durable.
What this means for conversion
The agentic surface tends to convert at higher rates than search-referred traffic for the query shapes that have shifted to it. The mechanism is that the agent has already done the filtering. The buyer who clicks an agent referral has been told by the agent why this product is being recommended. They arrive on the merchant’s page with a meaningful share of the purchase decision already made.
The conversion lift is largest for multi-constraint queries, where the agent has done filtering work the buyer would have done themselves on a SERP. The buyer lands on a product that already matches their constraints, which is structurally a much higher-intent visit than a generic search referral.
The conversion picture on planning queries is harder to summarize because the buyer is often buying multiple products from multiple merchants in the same session. The per-merchant lift is real but the right metric for this query class is total spend across the planned purchase, which the industry doesn’t yet have clean attribution for.
The aggregate picture: where the queries are leaving Google, the conversion economics on the new surface tend to be better, not worse. The merchants that capture agent referrals are getting higher-quality traffic. The merchants that haven’t built for the new surface are losing both the volume of these queries and the conversion premium they carry.
What this predicts
If the structural fit argument is right (and we think it is), the queries that will leave Google next are the ones with the next-worst Google fit. The candidates we’re watching:
Sequenced decision queries. “I’m furnishing my apartment, where do I start.” “I want to learn to bake, what do I need.” These have some of the planning-query shape but they’re sequential rather than simultaneous. We expect them to migrate as agents get better at holding state across sessions.
Compatibility queries. “Will this part fit my car.” “Is this skincare ingredient safe with my medication.” These have very specific answers that agents can synthesize from structured data faster than Google can rank a forum post. The shift hasn’t happened yet because the trust bar on these queries is high. Once the agents prove themselves on the safety-critical version, the shift accelerates.
Subjective preference queries. “Which book should I read after [other book].” “What movie should I watch on a rainy Sunday.” The agent-vs-Google fit on these is contested. Google has community-curated recommendations. Agents have synthesis. We expect a split, with the share moving slowly toward agents as their personalization improves.
The category that we’re not predicting to move much is navigational and brand-specific queries. Those queries are about getting to a specific destination, and Google is faster than any agent will be. Brand campaigns that rely on the navigational query will keep working on Google. Comparison campaigns that rely on the merchant being in the consideration set will not.
What to allocate against
The acquisition allocation question that matters: how much of your budget is sitting against the share of queries that are leaving the surface vs. the share that’s staying?
If you sell something where the buyer journey is shaped like comparison, multi-constraint, or planning queries, your acquisition budget should be migrating toward the new surface as fast as your measurement allows. If you sell something where the buyer journey is shaped like navigational queries (brand-loyal buyers coming back, branded campaigns driving direct-search traffic), the allocation shift is slower and your existing Google work has more runway.
Most brands are in the first category and treating themselves like they’re in the second. The high-intent share of their commercial queries is migrating off Google and they haven’t rebalanced. The brands that rebalance early in 2026 will get the leverage. The brands that wait until BFCM 2026 makes it politically obvious will be six months behind their fastest competitors.
Not all queries are leaving Google. The high-intent ones are. Worth knowing which category you’re in.