Skip to content

Agentic Commerce: The $5 Trillion Shift Rewriting How Humans Shop

Here’s a bet about Christmas 2025: Right now, somewhere in the US, someone is asking ChatGPT: “I need gifts for my nephew who loves Lego and space, my sister who’s into sustainable fashion, and my mum who wants kitchen gadgets but her counter space is tiny-budget £200 total, everything needs to arrive by the 23rd.”

And ChatGPT isn’t just answering. It’s shopping for them. Building carts. Comparing prices across retailers. Checking delivery windows. Completing purchases. All inside a single conversation thread.

Five Christmases ago, this was science fiction. Last Christmas, it was a beta feature most people hadn’t heard of. This Christmas, it’s infrastructure - live, scaled, and processing millions of shopping conversations daily across ChatGPT, Perplexity, Google’s AI Mode, and Amazon’s Rufus.

The question isn’t whether AI will change how people shop for the holidays. It already has. The question is whether you’re building for the channel where 51% of Gen Z now start their gift research, where 800 million people have weekly conversations, and where $5 trillion in commerce is projected to flow by 2030.

Large Language Models have crossed a threshold: they’re no longer just information retrieval systems. They’re becoming commerce channels. And the numbers suggest this isn’t hype, it’s the fastest commercial transformation since Google turned search into a $300B industry.

The Important Numbers

Let’s start with scale, because that’s where speculation ends and strategy begins:

800 million weekly active users on ChatGPT alone as of October 2025. That’s 10% of the global adult population in a single platform that launched less than three years ago.

51% of Gen Z now start product research in LLM platforms like ChatGPT or Gemini-bypassing Google entirely.

4,700% year-over-year increase in AI agent traffic to e-commerce sites (July 2025).

$1-5 trillion projected market size for agentic commerce by 2030, with some analysts calling it conservative.

And here’s the kicker: 25% of consumers have already made AI-assisted purchases. Not “might consider it.” Already done it. And 29% of non-users plan to adopt in 2026.

We call this a market shift.

What Actually Changed

If you’ve been watching the LLM space, you’ve seen the progression:

Phase 1 (2022-2023): ChatGPT as a curiosity - “look, it writes poetry!”
Phase 2 (2023-2024): Productivity tool - “it helps me draft emails”
Phase 3 (2024-2025): Answer engine - “it synthesizes research better than search”
Phase 4 (2025-now): Action engine - “it just bought my groceries”

That last phase? That’s agentic commerce. And it’s shipping right now across every major platform.

What Platforms Are Actually Shipping

The pattern is consistent across the ecosystem: move from searching pages → to describing intent → to getting an organized guide → to taking action without leaving the chat.

OpenAI (ChatGPT)

Shopping Research: A dedicated experience that produces shopping guides + comparisons from natural-language needs. Not “best vacuum” but “quiet cordless stick vacuum for small apartment with hardwood floors and a cat.”

Instant Checkout: OpenAI is explicitly enabling in-chat purchase flows. Instacart is one of the first visible “end-to-end checkout inside ChatGPT” implementations. You discuss what you need for dinner, build a cart conversationally, and checkout-all without opening a browser tab.

Agentic Commerce Protocol: Built with Stripe, this is the infrastructure layer for merchants to integrate product feeds and enable instant checkout. It’s open, it’s live, and merchants are onboarding now.

The Trust Signal: OpenAI recently disabled “ad-like” in-app promotions after user backlash. This matters enormously, it signals that trust is the core currency for shopping recommendations, and platforms know it.

Target in ChatGPT

Launched November 2025 in beta. You can now say: “Plan for a holiday family movie night” and Target’s AI will curate throw blankets, candles, snacks, slippers, and cocoa mix-then let you purchase multiple items in a single transaction with Drive Up, Pickup, or Shipping options. All inside ChatGPT.

Perplexity

Launched native shopping + purchasing experience (“Shop like a Pro” / “Buy with Pro”) that can complete orders without sending users to a retailer site. Critically, they’re positioning around merchants retaining the customer relationship via PayPal partnership and merchant-of-record frameworks.

Amazon (Rufus)

Amazon’s generative shopping assistant, embedded directly in the Amazon experience. Rufus is oriented around product discovery, comparison, and decision support-trained on Amazon’s catalog, customer reviews, and Q&A data. It’s purpose-built for commerce, not adapted from general chat.

Alexa+ is pushing into agentic behavior: monitor price drops, auto-buy under a threshold, manage subscriptions.

Google (AI Mode / Shopping Graph)

Google is blending Gemini-style conversational answers with the Shopping Graph-price, reviews, inventory - inside “AI Mode.” They’re not abandoning search. They’re augmenting it with synthesis.

Microsoft Copilot

Shipping commerce features like price tracking, comparisons, and buying assistance. Also running a Copilot Merchant Program to onboard product data for visibility and sales.

Shopify (The Distribution Layer)

Here’s the strategic move: Shopify just launched Agentic Storefronts - a control plane to syndicate product catalogs into AI shopping surfaces (ChatGPT, Perplexity, Copilot). Merchants can control where they appear, track attribution, and measure AI referral performance.

This is Shopify saying: “AI platforms are a new channel. You need controls + attribution.” And millions of merchants are gaining instant access to LLM commerce flows.

How Consumer Behavior is Actually Changing

The web is still here. Google still drives traffic. But inside LLM platforms, fundamentally different interaction patterns are emerging-patterns that resemble the web but are augmented in specific, repeatable ways.

1. From Keywords → Full Intent Narratives

Old web: “best running shoes”
LLM platform: “I’m training for a marathon, recovering from plantar fasciitis, need stability shoes under £150, overpronation issues, UK size 9, need them within 2 weeks for race prep.”

Users give context: budget, lifestyle, constraints, preferences, ethics, compatibility, household details. And platforms are designed around this shift-ChatGPT shopping research explicitly prompts for clarifying questions.

2. Conversational Filtering (Iterative Constraint Tightening)

On the web, you filter with facets: price sliders, checkboxes, dropdowns.

In LLMs, you negotiate:

  • “Ok, exclude brands with poor warranty”
  • “Must be under 2kg”
  • “I hate strong fragrance”
  • “Needs USB-C”
  • “No subscription models”

This becomes a back-and-forth requirements elicitation loop like talking to a knowledgeable salesperson who remembers every constraint.

3. Guide-First Over List-First

LLMs increasingly return:

  • A structured guide (what matters, tradeoffs, how to decide)
  • Shortlists + side-by-side comparisons
  • Contextualized recommendations (best for quiet, best value, best warranty, best for allergies)

Not 10 blue links. Not endless scrolling. A curated decision framework.

This is explicitly the product direction for ChatGPT shopping research and Google’s AI shopping answers.

4. Trust Assembled From Multiple Sources

Consumers ask LLMs to reconcile:

  • Customer reviews
  • Expert tests
  • Forum discussions
  • Return policies
  • “Real world” feedback from Reddit, YouTube, blogs

The LLM compresses it. This is why “trusted sources / reviews / community sentiment” become ranking features inside the conversation even when the purchase happens elsewhere.

5. Delegation: Handing Off Chores, Not Just Questions

  • Price tracking
  • Deal hunting
  • Replenishment reminders
  • Gift selection
  • Cart building
  • Compatibility checking

These are agentic tasks. Alexa+ auto-buy/price-drop flows are strong examples. So is ChatGPT building an Instacart cart based on “I need ingredients for pasta carbonara for 4 people, no bacon, I have garlic and olive oil already.”

6. The Single-Thread Funnel

Huge behavioral change: discovery → evaluation → purchase can happen in one continuous conversation thread.

No tab-switching. No context loss. No “let me go check reviews then come back.” The entire journey collapses into a dialogue.

Especially powerful when checkout is native (Instacart in ChatGPT, Perplexity’s flows, Target integration).

7. Post-Purchase Becomes Conversational Too

After buying, users ask:

  • Setup instructions
  • “How do I use this feature?”
  • “Did I buy the right thing?”
  • Returns/exchanges guidance
  • Troubleshooting

Whoever owns that post-purchase loop wins loyalty. Hence Perplexity emphasizing merchant-of-record / relationship retention.

8. Commercial UX Must Be Earned, Not Injected

LLM users are extremely sensitive to anything that feels like ads inside an assistant.

OpenAI turning off “ad-like” promos is a critical signal: trust > monetization in the short term, or the channel collapses entirely.

Users tolerate ads in search results because that’s the established contract. They won’t tolerate them in a conversation with their “assistant.”

The Five High-Frequency Shopping Intents

Every platform is converging on the same intent categories. If you’re building for this channel, these are your use cases:

1. “Help Me Decide” (High Consideration)

Electronics, appliances, fitness gear, baby products, skincare routines-anything where specs + tradeoffs + reviews matter deeply.

ChatGPT explicitly positions shopping research for this “fewer tabs, better choices” use case.

Example: “I need a laptop for video editing under $2000, prioritize screen quality and quiet fans, must run Premiere Pro smoothly.”

2. Gifting

“I need a gift for [person] who likes [interest], under £[budget], arrives by [date].”

Web search is clumsy for this. LLMs are naturally better. They understand context, recipient psychology, and occasion appropriateness.

Example: “Gift for my 4-year-old niece who loves art, creative but not too messy, around £30, birthday is next week.”

3. Deal Confidence

“Is this a good price?"
"Will it drop?"
"Find a cheaper equivalent with similar specs.”

Copilot and Alexa+ both lean into price intelligence / tracking behaviors.

Example: “I’m looking at this KitchenAid mixer for £280-is that fair or should I wait?“

4. Routine Replenishment + Basket-Building

Groceries, household essentials, repeat purchases-where the job is assembling a cart, applying constraints, and finishing checkout.

Instacart’s in-ChatGPT flow is the canonical example.

Example: “I need my usual groceries plus ingredients for Thai curry tonight, serves 4, one person is vegetarian, budget around £60.”

5. Compatibility & Ecosystem Shopping

“What works with my existing [device/system/setup]?”

Phone model, camera mount, smart home hub, skin sensitivity, dietary restrictions.

This is where structured product data matters most-the AI needs to understand compatibility matrices.

Example: “What MagSafe accessories work with iPhone 15 Pro and are TSA-compliant for carry-on?”

What This Changes for E-Commerce

A) Visibility Shifts from SEO Rank → Inclusion in the Assistant’s Shortlist

If the assistant returns 3–7 options, the game becomes: be one of those options, and be framed correctly (best for quiet, best value, best warranty, best for allergies).

This is why platforms are building merchant programs and catalog syndication layers. Shopify’s agentic storefronts are infrastructure for this new visibility game.

B) The New “Content Moat” is Structured + Evidence Rich

To be recommended, assistants need:

  • Consistent product attributes (dimensions, materials, compatibility, certifications)
  • Pricing + inventory (real-time where possible)
  • Policies (shipping, returns, warranty)
  • Proof (reviews, Q&A, credible third-party coverage)

Google’s AI shopping answers explicitly emphasize bringing together price/reviews/inventory. Merchant programs exist to keep this data current.

C) Checkout Can Move Upstream Into the Conversation

When Instant Checkout exists, conversion stops being “landing page optimization” and becomes:

  • Recommendation relevance
  • Trust in the assistant’s framing
  • Frictionless confirmation + payment

That’s the strategic meaning of OpenAI’s Agentic Commerce Protocol + merchant onboarding.

D) Attribution is the Next Battleground

Merchants will demand to know: Did this sale come from ChatGPT? Perplexity? Copilot?

Shopify’s pitch is basically: “AI platforms are a new channel; you need controls + attribution.”

Expect “AI referral” to become a standard analytics dimension, alongside organic, paid, social, email.

The Readiness Checklist

If you want to show up (and convert) inside LLM shopping flows, the near-term winning moves are boring but decisive:

Ship a clean product feed
Complete attributes, variants, GTIN/MPN where applicable, pricing, availability. Think of this as your “Schema.org compliance” but for AI.

Make policies machine-readable
Returns, warranty, shipping thresholds, delivery windows-easy to parse and summarize.

Invest in Q&A and review quality
Assistants summarize social proof aggressively. High-quality, authentic reviews become ranking signals.

Define “best-for” positioning per SKU
Quietest, lightest, safest for sensitive skin, best value under £X. The assistant needs hooks to match you to intent narratives.

Prepare for assistant-first UX
Fewer clicks, more confirmations. Your product page may be secondary to the assistant’s explanation. Design for trust + verification, not persuasion.

Treat trust as a KPI
Anything ad-like or misleading can get you filtered out-by both the platform and user reaction loops.

Why We’re Covering This Now

At Performics Labs, we’ve been tracking the geometry of intention-how LLMs can infer not just what users say, but what they actually want. Our deep dive on Context-Conditioned Intent Activation laid the foundation: LLMs can recognize human goals with 75-82% accuracy when given proper context.

Agentic commerce is that framework, deployed at scale.

The same principles we explored-context capture, intent recognition, pattern discovery, activation-are now powering shopping experiences for 800 million people weekly.

And the patterns described above (full intent narratives, conversational filtering, guide-first results, delegation of chores) are exactly the behaviors our intent recognition research predicted. They’re just manifesting in commerce first, because that’s where the financial incentive is strongest.

What’s Coming Next

Over the coming weeks, we’ll be publishing a deep dive series on agentic commerce, breaking down:

  1. The Intent Taxonomy of LLM Shopping – Mapping the 5 core shopping intents to technical implementation patterns
  2. Building for AI Discoverability – Product data architecture, GEO vs SEO, and structured data strategies
  3. Measuring the AI Channel – Attribution models, conversion tracking, and analytics for LLM referrals
  4. Conversational UX Patterns – Designing for dialogue-driven discovery and checkout
  5. Agent-to-Agent Commerce – What happens when AI shops for AI (yes, really)

Each deep dive will come with open-source implementations-following our “one paper with code” philosophy. Just as we built the Intent Recognition Agent with MCP tools (demo) to complement our geometry of intention research, we’ll ship working examples of:

  • Product feed optimization for LLM discovery
  • Conversational shopping assistants with memory
  • Attribution tracking for AI referrals
  • A/B testing frameworks for assistant-first experiences

Because understanding the patterns is step one. Building for them is how we learn.

The Question That Matters

Is agentic commerce overhyped?

No.

Is the timeline uncertain? Absolutely.

Will traditional e-commerce disappear? Of course not.

But here’s what’s undeniable:

  • 800M weekly active ChatGPT users
  • 25% of consumers have already purchased through AI assistance.
  • 51% of Gen Z now start product research in LLMs like ChatGPT/Gemini
  • Traffic from AI agents to e-commerce sites: +4,700% year-over-year
  • Every major platform (OpenAI, Google, Amazon, Microsoft, Meta) is shipping commerce features right now.

The channel is real. The behavior change is measurable. The infrastructure is deploying.

The only question left is: Are you building for it?

Sources

1. 800M weekly active ChatGPT users (October 2025)

2. 51% of Gen Z start product research in LLMs

3. 25% of consumers have made AI-assisted purchases

4. 4,700% year-over-year traffic increase

5. $1-5 trillion projected market by 2030

Follow our deep dive series on Agentic Commerce as we unpack the technical patterns, behavioral insights, and implementation strategies that will define this new channel.

Other Articles