This is the predictions post. Five bets on what 2026 looks like for agentic commerce, each one we’ll grade ourselves against in June.
The format on each: the bet, the reasoning, the signal to watch for, and what to do this quarter if you agree. We’ll come back to score these honestly mid-year. Some will land. At least one won’t.
1. AI assistants will overtake Google for high-intent product queries
The bet: By Q4 2026, the share of “best X for Y,” “X vs Y,” and multi-constraint product queries answered primarily through an AI assistant (ChatGPT, Perplexity, Claude, Gemini, Copilot) will exceed the share answered through Google’s blue links. AI Overviews don’t count for this. We’re talking about queries that resolve inside a chat surface, not on a SERP with a generated summary above the links.
The reasoning: Comparison-shaped queries are structurally better served by a recommendation with justification than by a ranked list of pages. Google has won them for fifteen years because the alternatives were worse. The alternatives are no longer worse. The query-mode mix on ChatGPT and Perplexity is already heavily comparison-shaped, and BFCM 2025 showed conversion rates on agent referrals in those categories matching or exceeding paid search.
The signal to watch: Quarterly query-share reports from Adobe, eMarketer, and the agentic-search trackers. The crossover point will be category-specific. Electronics, beauty, and home will tip first. CPG and apparel will lag because of buyer-experience requirements (touch, try, return) that don’t move as fast.
This quarter, do this: Map your top 20 commercial queries to query shape. The ones that are comparison-shaped or multi-constraint are the ones where agentic share is climbing fastest. Audit how the major agents answer them today. If you’re not in the response, find out why before BFCM 2026 makes it urgent.
2. Agentic checkout will become default for mid-sized baskets
The bet: By year-end 2026, a meaningful share of purchases that originate from an AI assistant conversation will complete inside the conversation, without the buyer ever loading the merchant’s storefront. The share will be larger for mid-sized baskets than for higher-consideration purchases, because higher-consideration purchases pull the buyer back to the site for the look-around step.
The reasoning: ChatGPT Instant Checkout, ACP-compliant Shopify and BigCommerce storefronts, and Apple Intelligence’s Siri commerce path all collapse the steps between “the agent recommends a product” and “the purchase is complete.” The friction reduction is large enough that the behavior change happens quickly. Buyers who’ve completed one purchase this way will complete the next one this way. The category that takes the longest to convert is the buyer who hasn’t tried it yet.
The signal to watch: Stripe’s quarterly transaction-type breakdown, Shopify’s Plus-tier merchant data, and the cohort behavior of buyers who’ve completed at least one in-conversation purchase. The repeat rate is the leading indicator.
This quarter, do this: If you’re on Shopify and the merchant-side ACP toggle isn’t on, turn it on. If you’re on a custom stack, publish an ACP manifest at /.well-known/acp/manifest.json. The merchants who treat this as a Q3 priority are going to find the leaderboard already set when they get there.
3. llms.txt mainstreams and becomes a job-description line item
The bet: By Q4 2026, “Owns the llms.txt strategy” appears as a bullet point on job descriptions for senior e-commerce, content, and SEO roles at mid-market and enterprise brands. The role isn’t new. It’s the same role that owned the editorial calendar or the structured-data implementation. The line item is new because the file became too important to be nobody’s job.
The reasoning: Grok already fetches llms.txt actively. ChatGPT and Claude reference it when present. The agents that find a clear, accurate llms.txt index your catalog with higher confidence. Once that becomes measurable in merchant reporting, the file moves from “someone should write that” to “someone owns that.” This is the trajectory the robots.txt file took in the early 2000s. The trajectory llms.txt is taking is faster because the visible commercial impact is larger.
The signal to watch: Job postings on LinkedIn and the major retail-industry boards. Watch the senior-content and SEO lead descriptions in Q2 and Q3.
This quarter, do this: Decide which role owns your llms.txt today. If the answer is “nobody,” you have your action item. The first version should be honest, matter-of-fact, and version-controlled. The second version, which you’ll write after watching three months of agent query patterns against it, will be the one that compounds.
4. “Agent SEO” emerges as a discipline distinct from traditional SEO
The bet: By mid-2026, a recognized “Agent SEO” practice exists, with its own job titles, its own consultants, its own playbooks, and its own conference track at the major SEO events. Traditional SEO practitioners who don’t add Agent SEO to their skill set will start losing engagements to those who have.
The reasoning: The signals agents reward are different from the signals search engines reward. Honesty bar is different. Structured-data weight is different. Citation patterns are different. Speed-to-answer matters in a way it didn’t for search. The practitioners who study these signals empirically, build their own measurement, and update their playbooks against the agents’ moving behavior will produce better outcomes for their merchants. The practitioners who keep optimizing only for Google blue links will, depending on category, lose meaningful share of their addressable acquisition.
The signal to watch: The agendas of SearchEngine Land’s SMX, BrightonSEO, and the new “agentic search” conferences that will appear in Q2 and Q3. The talks that fill the largest rooms tell you where the discipline is consolidating.
This quarter, do this: Run an audit on your current SEO efforts. What share of your effort and budget addresses signals that matter to agents, vs. signals that matter only to Google? The mismatch is your roadmap.
5. Walled gardens vs. open commerce becomes the year’s defining tension
The bet: By year-end 2026, the two paths for agentic commerce diverge enough that brands will have to pick a posture. Path one: optimize for the walled gardens (Amazon’s agent surface, Walmart’s agent surface, Apple’s Siri commerce path, eventually Meta’s). Path two: build for the open protocols (ACP, llms.txt, the broader open-web agent surface). Most brands will end up doing some of both. The leverage is asymmetric, and the brands that recognize where their unit economics live will allocate accordingly.
The reasoning: Marketplaces extract margin. They always have. The agent-mediated path inside a walled garden is, structurally, a marketplace dynamic. The buyer asks Amazon’s agent. The agent recommends from Amazon’s catalog. The margin Amazon takes for placement is real, and it’s negotiated from a position the brand doesn’t control. The open protocols (ACP, llms.txt, the broader open-agent web) are the path where the brand owns the relationship. The margin structure is different. The compounding is different. The merchants who recognize this in early 2026 will allocate their integration budgets accordingly.
The signal to watch: The take-rate data from the walled-garden agent surfaces. Amazon, Walmart, and Apple don’t publish it cleanly, but the inference from earnings calls and merchant feedback will get clearer through the year.
This quarter, do this: Run the unit economics for two scenarios. Scenario A: a significant share of your agent-mediated revenue comes through walled-garden paths at their take rates. Scenario B: the same share comes through ACP and other open paths at your direct take rate. Pick a share that matches your category’s plausible trajectory. The delta tells you how much it’s worth to invest in the open-protocol path now.
We’ll grade these in June
We will write a mid-year retrospective in June that scores these five bets honestly. The ones that landed get a victory lap. The ones that didn’t get an explanation of what we got wrong. We’d rather be wrong publicly and update than be vaguely right in private.
If you run a store, the question isn’t whether all five of these land exactly as written. It’s whether the direction is right enough that the moves you make this quarter compound. We think it is. We’ll find out together.