ChatGPT vs Perplexity: AI Shopping Assistant for Retailers

A few months ago we sat with a mid-sized apparel merchant who swore their catalog was invisible to AI. They had spent weeks optimizing product descriptions, hired a copywriter, rebuilt their feed. Then we asked ChatGPT a plain question: “find me a lightweight waterproof jacket under $150 that ships to Canada.” Their product never surfaced. A competitor with a thinner catalog but a clean, agent-readable data layer showed up three times. That gap, between a store that humans can browse and a store that an AI shopping assistant for retailers can actually read and act on, is the entire story of commerce in 2026. The two channels every merchant asks us about first are ChatGPT’s shopping experience and Perplexity’s shopping integration, and they are not interchangeable.

This article is our head-to-head breakdown. We build UCP infrastructure and implement agentic commerce for real merchants every week, so we are going to be opinionated about where each platform wins, where each one leaves money on the table, and how to decide which one deserves your attention first. Both matter. But treating them as a single “AI shopping” bucket is how merchants waste a quarter optimizing for the wrong behavior.

TL;DR

ChatGPT strengths: Massive weekly reach, conversational product discovery, and deep integration momentum with checkout rails, but it rewards structured, agent-readable catalog data over clever marketing copy.
Perplexity strengths: Answer-engine positioning with cited sources, strong comparison and research queries, and a buyer who arrives late in the funnel with high intent, but narrower reach and a different surfacing logic.
The real decision: The winner is not a platform, it is your data layer. An AI shopping assistant for retailers only works if your product feed, pricing, and availability are exposed through a protocol both engines can trust, which is exactly the problem the Universal Commerce Protocol solves.

Why This Comparison Matters Now

Let us be direct about the stakes. Search behavior is splitting. A meaningful and growing share of product research now starts inside a conversational AI interface instead of a search bar or a marketplace. When a shopper asks an assistant to “compare these three espresso machines and tell me which has the best value,” they are outsourcing the entire consideration phase to a model. The retailer who wins that query does not win because of SEO tricks. They win because the assistant could read their data, trust it, and act on it.

We have watched this shift accelerate through what we call the third wave of ecommerce AI, the move from predictive recommendation engines to genuinely agentic systems that take action on the shopper’s behalf. If you want the longer arc, we wrote it up in our piece on the third wave from predictive to agentic AI in ecommerce. The short version: the assistant is becoming the shopper’s proxy, and your store has to be legible to that proxy.

ChatGPT and Perplexity are the two highest-leverage surfaces for this right now. They have different audiences, different surfacing logic, and different checkout ambitions. Choosing where to invest first requires understanding both, not guessing.

How big is the AI shopping shift really?

Big enough that we no longer treat it as experimental, small enough that most of your competitors are still ignoring the data-layer work. According to UCP Checker, which independently monitors more than 20,000 storefronts, roughly 82% pass full UCP validation (16,376 verified). We cite that carefully: that number is the share of the stores UCP Checker tracks, a set that skews heavily toward Shopify, not a claim about all ecommerce. And a conformant UCP manifest is not the same thing as an agent being able to complete a real checkout. It tells you the plumbing exists. It does not guarantee water comes out of the tap. That caveat is the whole reason this comparison is worth 5,000 words instead of a tweet.

ChatGPT vs Perplexity at a Glance

Here is the comparison we draw on a whiteboard for nearly every merchant who asks about an AI shopping assistant for retailers.

Criterion	ChatGPT Shopping	Perplexity Shopping
Primary user intent	Discovery and conversational browsing	Research, comparison, late-funnel decisions
Reach	Very large weekly active base	Smaller, growing, higher intent per session
Surfacing logic	Model reasoning plus structured product data and partner feeds	Citation-driven, source-ranked answer synthesis
Checkout ambition	Aggressive, integrated checkout push	Present but more referral-oriented
Data sensitivity	High: rewards clean, structured, real-time feeds	High: rewards authoritative, well-cited product pages and feeds
Best for	Broad catalogs, impulse and discovery categories	Considered purchases, specs-heavy categories
Merchant control	Via feeds, protocols, and partner programs	Via feeds, crawlable content, and source authority
Setup complexity	Moderate to high, protocol-dependent	Moderate, content and feed dependent

Notice what is identical across both columns: data sensitivity is high in both cases. That is not a coincidence. Neither engine can recommend what it cannot parse. We will come back to this repeatedly, because it is the single most important thing we tell clients.

ChatGPT as an AI Shopping Assistant for Retailers

ChatGPT is the reach play. Its weekly active user base is enormous, and a large portion of those conversations now drift into product territory even when they do not start there. Someone asking for a dinner party menu ends up asking where to buy a specific serving platter. That ambient, discovery-first behavior is ChatGPT’s defining commerce characteristic.

Conversational discovery: ChatGPT excels when the shopper does not yet know exactly what they want. It handles vague, exploratory, multi-constraint requests gracefully. “I need a gift for my dad who likes woodworking but already has every tool, under $80” is a query that traditional search handles poorly and ChatGPT handles well. For retailers in gift, apparel, home, and lifestyle categories, this is a genuinely new top-of-funnel.

Checkout momentum: ChatGPT has pushed hardest on integrated checkout and agentic purchase flows. The ambition is clear: compress discovery, decision, and transaction into one conversation. We have found this is where the biggest upside and the biggest risk both live. When it works, the shopper never leaves the assistant. When your data is wrong, the assistant confidently recommends an out-of-stock SKU and the failure is invisible to you until the chargebacks or the abandoned sessions pile up.

Data dependency: Here is the uncomfortable truth we deliver to merchants expecting a marketing win. ChatGPT does not reward your clever product copy the way Google once did. It rewards structured, machine-readable, current data. Price, availability, variants, shipping eligibility, return policy, all of it needs to be exposed in a form the model and its commerce partners can consume without hallucinating. This is exactly why we built so much of our practice around product feed optimization for AI shopping agents. The feed is the product now.

Where ChatGPT falls short for retailers

It falls short when your catalog is a mess, which is more often than merchants admit. Stale inventory, inconsistent variant naming, prices that lag your real-time system by hours, these all degrade quietly. ChatGPT will still answer; it just will not answer with your products, or worse, it answers with wrong information about your products and erodes trust. There is also limited transparency into exactly why a given product surfaced, which makes diagnosis harder than classic SEO where you could at least inspect rankings.

ChatGPT checklist for retailers:

Real-time feed sync: Keep price and availability accurate to the minute, not the day, because agentic checkout acts on live data.
Structured variants: Normalize sizes, colors, and configurations so the model never guesses which SKU it is recommending.
Shipping and returns clarity: Expose eligibility and policy as structured fields, not buried prose, so multi-constraint queries resolve correctly.
Protocol exposure: Publish an agent-readable commerce layer rather than relying on the model to scrape your HTML.
Failure monitoring: Track agent-driven sessions separately so a silent surfacing drop does not go unnoticed for weeks.

Perplexity as an AI Shopping Assistant for Retailers

Perplexity plays a different game. It positions itself as an answer engine, and its commerce behavior inherits that DNA. Shoppers come to Perplexity to research, compare, and verify. The sessions are often shorter, more deliberate, and closer to the purchase decision.

Citation-driven trust: Perplexity shows its sources. That is both a feature for users and a signal for retailers. If your product pages and feeds are authoritative, well-structured, and citable, Perplexity is more likely to pull you into its synthesized answer with attribution. We have seen merchants in specs-heavy categories, electronics, appliances, outdoor gear, do disproportionately well here because their content answers precise comparison questions.

Late-funnel intent: A shopper asking Perplexity “which of these two robot vacuums handles pet hair better and has a lower long-term filter cost” is not browsing. They are deciding. The intent quality per session is high, even if the raw volume is lower than ChatGPT. For retailers, a smaller number of higher-intent impressions can convert better than a flood of casual discovery traffic.

Content and feed dependency: Perplexity rewards two things working together. First, crawlable, authoritative product content that answers real comparison questions. Second, a structured feed so the engine can anchor its answer to accurate price and availability. If your content is strong but your feed is stale, Perplexity may cite you while showing a wrong price. If your feed is clean but your content is thin, a competitor with richer comparison content gets cited instead.

Is Perplexity worth it for smaller catalogs?

Often, yes, and this surprises people. Because Perplexity synthesizes answers from authoritative sources rather than needing an enormous catalog to rank, a focused retailer with genuinely excellent product information can punch above its weight. A boutique mattress brand with detailed firmness, material, and durability data can win a comparison query against a giant generalist whose pages are shallow. Depth of structured product truth beats breadth here more often than on ChatGPT.

Perplexity checklist for retailers:

Comparison-ready content: Write product pages that directly answer “A vs B” and “best for X” questions in structured, citable form.
Source authority: Earn and maintain the kind of page quality Perplexity trusts enough to cite with attribution.
Feed accuracy alongside content: Pair strong content with a live price and availability feed so citations never carry stale data.
Specification density: Expose complete, structured specs for considered-purchase categories where buyers compare on details.
Attribution tracking: Monitor which queries cite you so you can double down on the comparison intents you already win.

The Framework We Use: The AGENT Readiness Method

Whichever platform you prioritize, the underlying work is the same, and we run every client through a five-step method we call AGENT. It is deliberately platform-neutral because the whole point is that your investment should pay off on ChatGPT, Perplexity, and whatever surfaces next.

Audit your data truth. What this achieves: It establishes whether your current price, inventory, and variant data are accurate in real time or merely accurate on paper. We start every engagement here because most surfacing failures trace back to data the merchant assumed was correct. Pull a random sample of 50 SKUs and compare the feed against your live backend; if more than a handful disagree, stop optimizing content and fix the pipe first.

Generate a protocol layer. What this achieves: It gives AI engines a single, standardized, machine-readable description of your commerce capabilities instead of forcing them to scrape and guess. This is where the Universal Commerce Protocol does the heavy lifting; rather than maintaining a brittle custom integration per channel, you publish one conformant layer. We explain the full rationale in our merchant guide to the Universal Commerce Protocol.

Enrich for intent. What this achieves: It maps your catalog to the actual questions shoppers ask the assistants, discovery questions for ChatGPT, comparison and specification questions for Perplexity. We tag products with the constraints buyers care about, shipping eligibility, use-case fit, comparative advantages, so the model has something precise to match against.

Normalize in real time. What this achieves: It keeps every channel synchronized to your live system so an agentic checkout never acts on data that is hours old. We have written extensively about why real-time data sync matters for AI agents; the one-line version is that agents transact, and transacting on stale data is how you sell inventory you do not have.

Test like an agent. What this achieves: It validates that an assistant can actually complete the journey, not just read your manifest. This is the caveat we keep returning to: passing validation is necessary but not sufficient. We script real agent queries and real checkout attempts, because a conformant manifest with a broken checkout is worse than no presence at all. Our UCP Hub demo and testing guide walks through exactly how we do this.

The platform you bet on matters far less than whether an AI agent can read your data and finish the sale, and in our experience the merchants who obsess over copy while ignoring their feed lose to competitors with thinner catalogs and cleaner pipes every single time.

Stop Guessing Which AI Channel Wins: Let Us Build the Layer That Serves Both

Here is the strategic reality we land on with every merchant: you should not have to choose ChatGPT or Perplexity at the data level, because the expensive mistake is building a separate integration for each. The Universal Commerce Protocol lets you publish one agent-readable commerce layer that ChatGPT, Perplexity, and the next entrant can all consume, which is why we tell clients to stop thinking in channels and start thinking in protocols. If you want to see whether your store is actually legible to AI agents today, and get a concrete plan to make it convert on both platforms, talk to our team at UCPhub. We will run your catalog through the AGENT method and show you exactly where the gaps are.

Which Should You Choose: A Decision Framework

We refuse to give a one-size answer, because the right choice genuinely depends on your category, catalog, and goals. Here is the decision logic we use, mapped to specific situations.

Choose ChatGPT first if you sell discovery-driven categories. What this covers: apparel, gifts, home goods, lifestyle, beauty, anything where shoppers browse and get inspired rather than spec-compare. ChatGPT’s enormous reach and conversational discovery pattern fit impulse and inspiration shopping. If your margins support broad top-of-funnel exposure and your catalog is large, start here.

Choose Perplexity first if you sell considered, specs-heavy purchases. What this covers: electronics, appliances, tools, outdoor and technical gear, anything bought after comparison. Perplexity’s citation-driven, research-oriented behavior rewards deep product truth, and the late-funnel intent converts well even at lower volume. A focused catalog with excellent data often wins here against larger, shallower competitors.

Choose both, in sequence, if you have the resources. What this covers: most serious retailers in 2026. The smart move is not picking one forever; it is sequencing. Build the shared protocol layer once via UCP, then light up ChatGPT for discovery reach and Perplexity for high-intent conversion. Because the data layer is shared, the second channel costs a fraction of the first.

Do neither until your data is clean, if your feed is a mess. What this covers: the uncomfortable but common case. If your audit shows stale inventory and inconsistent variants, optimizing for either platform is pouring water into a cracked bucket. Fix the pipe first. We have seen merchants burn a quarter chasing surfacing wins when the real problem was a feed that lagged their backend by six hours.

The channel debate also connects to a bigger architectural question we keep hammering: point solutions do not scale. If you build a bespoke ChatGPT integration and a separate bespoke Perplexity integration, you now maintain two brittle things that break independently. We laid out why this fails in UCP vs custom AI integrations, and agencies managing many clients feel this pain most acutely, which is why they consolidate, as we described in how agencies use UCP to cut maintenance 80%.

Decision checklist:

Category match: Map your products to discovery intent (lean ChatGPT) or comparison intent (lean Perplexity) before committing budget.
Catalog depth versus breadth: Favor Perplexity when you have deep product truth, favor ChatGPT when you have broad discovery-friendly range.
Data readiness gate: Do not launch on either platform until a 50-SKU audit shows real-time accuracy.
Shared-layer first: Build the UCP protocol layer once so your second channel is cheap, not a second project.
Sequence, do not split: Launch one channel, measure for 30 days, then expand rather than running two half-built integrations.

Measuring Success: 30/60/90 Day KPIs

We do not let clients fly blind, and “we feel more visible in AI” is not a metric. Here is the scorecard we hold ourselves to on an AI shopping assistant for retailers rollout.

First 30 days, prove legibility:

Validation pass rate: Reach a conformant protocol layer and confirm it with independent checking, while remembering validation is the floor, not the ceiling.
Agent surfacing rate: Measure how often scripted test queries surface your products at all, targeting a clear baseline you can improve against.
Data accuracy delta: Drive the gap between your feed and your live backend toward zero on a weekly 50-SKU audit.
Completed test checkouts: Confirm an agent can finish a real purchase, not just read your manifest, because this is the caveat that bites merchants.

Through 60 days, prove intent capture:

Query coverage: Expand the share of your target discovery and comparison queries where you surface, tracked per platform.
Citation rate on Perplexity: Grow how often Perplexity cites you with attribution on comparison intents you care about.
Agent session quality: Watch conversion rate on AI-driven sessions versus your site baseline, since high-intent agent traffic should convert at least as well.
Silent failure alerts: Have monitoring that flags a surfacing drop within hours, not the three days it took the merchant in our opening story.

Through 90 days, prove revenue:

Agentic revenue share: Track the percentage of revenue attributable to AI channels and its week-over-week trend.
Cost per agent acquisition: Compare acquisition cost on AI channels against your existing paid channels to judge reallocation.
Second-channel marginal cost: Confirm that adding the second platform cost materially less than the first, validating the shared-layer thesis.
Return and chargeback rate on agent orders: Keep this at or below your overall baseline, because a spike signals stale data reaching checkout.

For the deeper revenue mechanics behind these numbers, our AI shopping assistant integration playbook breaks down how merchants actually grow agentic commerce revenue, not just presence.

The Protocol Question Behind Both Platforms

We keep returning to one point because it is the point: ChatGPT and Perplexity are channels, and channels change. What does not change is the need for a standardized way to tell any AI agent what you sell, for how much, in what quantity, with what policies. That is the role of a commerce protocol.

This matters even more as AI agents themselves become the primary shoppers rather than human-assisted tools. We explored that near-future in what happens when AI agents become the primary shoppers, and the conclusion holds: in a world where agents transact autonomously, the retailers who thrive are the ones machines can transact with reliably. There is also a live standards conversation, with different protocols competing and overlapping, which we mapped in our guide to agentic AI protocols including MCP, A2A, and UCP and in our comparison of ACP vs UCP. For Shopify merchants specifically, the agentic capabilities are becoming native, which we covered in Shopify AI automation and the agentic plan.

The practitioner takeaway is unglamorous but true. Pick your first channel based on your category, yes. But build the layer underneath it so your bet is portable.

Protocol-first checklist:

One layer, many channels: Publish a single agent-readable commerce layer rather than per-platform integrations.
Policy as data: Expose shipping, returns, and warranty as structured fields every agent can parse.
Portability test: Confirm your layer serves a second AI channel with minimal new work before you call it done.
Standards awareness: Track protocol developments so you are not locked into a channel-specific dead end.
Checkout verification: Validate real transactions continuously, because a passing manifest with a broken checkout still loses sales.

Final Verdict

If we had to hand a merchant one sentence, it is this: ChatGPT wins reach and discovery, Perplexity wins high-intent comparison, and neither wins anything if an agent cannot read your data and complete a checkout. For broad, inspiration-driven catalogs we start clients on ChatGPT. For considered, specification-heavy purchases we start on Perplexity. For anyone serious about 2026, we build the Universal Commerce Protocol layer first so the choice of channel becomes a tactical decision rather than an expensive architectural one. The platforms will keep changing. A clean, agent-readable, real-time commerce layer is the durable asset, and it is the one we would protect first.

If you are just getting started, prioritize the data audit before anything else; do not touch channel optimization until a 50-SKU sample proves your feed matches your live backend, because every surfacing problem we diagnose eventually traces back to that gap. If you are auditing something that already exists, start at the checkout end instead: run real scripted agent purchases on both ChatGPT and Perplexity and find out whether an agent can actually finish the sale, since a conformant manifest with a broken transaction is the failure mode that costs the most and hides the longest.

Next Steps:

Run a 50-SKU accuracy audit this week comparing your feed against your live backend and record the delta.
Script five real agent queries per platform, covering your top discovery and comparison intents, and note where you do and do not surface.
Book a review of your commerce layer so you launch on a shared protocol rather than two brittle per-channel builds.

Frequently Asked Questions

What AI shopping assistants do retailers use?

Most retailers we work with concentrate on two surfaces first, ChatGPT and Perplexity, because that is where the highest-leverage shopper behavior is concentrated today. ChatGPT draws the broadest audience and captures discovery-stage demand, the shopper who does not yet know exactly what they want and asks an open-ended question. Perplexity draws a smaller but more deliberate audience doing comparison and research, closer to the purchase decision.

Beyond those two, the broader landscape includes native capabilities inside commerce platforms themselves, Shopify’s agentic features being the most relevant for the merchants we serve, plus emerging answer and shopping surfaces from other model providers. We tell clients not to chase every surface individually. The mistake is building a separate integration per assistant. The durable approach is to expose one standardized, agent-readable commerce layer through a protocol like the Universal Commerce Protocol, so any assistant that can read that layer can recommend and transact with you.

In practice, the specific assistants matter less than whether your data is legible to all of them. We have watched merchants obsess over which platform to join while their underlying feed lagged their real inventory by hours, which undermined every channel at once. Fix the data layer first, then the question of which assistant becomes a tactical sequencing decision rather than an architectural one.

How can retailers implement AI shopping tools?

We run implementations through the same five-step AGENT method described earlier in this article, and the first step is always a data audit, not a channel launch. Pull a representative sample of your catalog, around 50 SKUs, and compare the price, availability, and variant data in your feed against your live backend. If they disagree, that is the first thing to fix, because agentic checkout acts on live data and stale data produces silent, expensive failures.

Next, generate a protocol layer so AI engines get a single machine-readable description of your commerce capabilities rather than scraping and guessing from your HTML. This is where a standard like UCP earns its keep: you publish one conformant layer instead of maintaining brittle, per-platform custom integrations that break independently. Then enrich that layer for the actual questions shoppers ask, discovery constraints for ChatGPT, comparison and specification details for Perplexity.

Finally, and this is the step merchants skip at their peril, test like an agent. Script real queries and real checkout attempts on each platform. Passing validation proves the plumbing exists; it does not prove water comes out of the tap. We script completed test purchases because a conformant manifest with a broken checkout loses more sales than having no AI presence at all, and it hides the failure from your normal analytics.

What are the best AI shopping assistants for e-commerce?

There is no single best assistant, there is a best fit for your category, and that is the honest answer we give every merchant. For discovery-driven categories, apparel, gifts, home, beauty, lifestyle, ChatGPT tends to be the strongest first channel because of its reach and its talent for handling vague, multi-constraint, exploratory queries. The shopper asking for a gift idea under a budget is a ChatGPT shopper.

For considered, specification-heavy purchases, electronics, appliances, tools, technical and outdoor gear, Perplexity often performs better because its citation-driven answer synthesis rewards deep, authoritative product truth and attracts late-funnel comparison intent. A focused retailer with genuinely excellent structured data can win comparison queries against larger competitors with shallower pages, which is one of the more encouraging patterns we see for smaller brands.

The best overall strategy for a serious retailer in 2026 is not to pick one forever. It is to build a shared protocol layer once, then launch the channel that matches your category, measure for 30 days, and expand to the second channel at a fraction of the original cost. The best assistant is whichever one converts for your products, and you only learn that by making your data legible to both and measuring real outcomes.

Does optimizing product copy still help with AI shopping assistants?

Copy still matters, but its role has changed, and merchants expecting a pure marketing win are usually disappointed. AI shopping assistants do not reward persuasive prose the way classic search once rewarded keyword-rich pages. They reward structured, accurate, machine-readable product truth. Price, availability, variants, shipping eligibility, return policy, and specifications need to exist as clean structured data, not buried inside flowery descriptions.

That said, well-written comparison content genuinely helps on Perplexity, because its answer engine cites authoritative sources that directly answer comparison and “best for” questions. So the useful copy in 2026 is not marketing copy, it is answer copy: clear, specific, honest content that resolves the exact questions buyers ask. The highest return comes from pairing that answer-oriented content with a clean, real-time feed. Strong content with a stale feed gets you cited with a wrong price, which is worse than not being cited at all.

How is ChatGPT shopping different from Perplexity shopping for merchants?

The core difference is intent and surfacing logic. ChatGPT leans into conversational discovery with enormous reach, surfacing products through model reasoning combined with structured data and partner feeds, and it pushes hard on integrated checkout to keep the whole journey inside the conversation. Perplexity leans into research and comparison with a citation-driven model, surfacing retailers by synthesizing and attributing authoritative sources, with a buyer who arrives later in the funnel and higher in intent.

For merchants, that translates to different wins. ChatGPT gives you broad top-of-funnel discovery exposure; Perplexity gives you fewer but higher-converting comparison impressions. ChatGPT’s aggressive checkout integration means your data accuracy matters at transaction time, so real-time feed sync is non-negotiable. Perplexity’s citation model means your content authority and specification depth matter more for whether you get pulled into the answer at all. The shared requirement, and the reason we treat them as one infrastructure project, is that both are intensely sensitive to the accuracy and structure of your underlying data.

Do I need the Universal Commerce Protocol to sell through AI assistants?

You do not strictly need a formal protocol to appear in an AI shopping assistant, assistants can and do surface products by scraping and inferring. But relying on scraping is fragile and leaves you with no control, no guarantee of accuracy, and a high risk of the assistant confidently stating wrong information about your products. A protocol layer like UCP gives the assistant a single authoritative, machine-readable description of what you sell, for how much, in what quantity, with what policies, which dramatically improves both legibility and transaction reliability.

The stronger argument is portability and maintenance. If you build bespoke integrations per channel, you maintain multiple brittle things that break independently, and every new AI surface is a new project. A shared protocol layer means you build once and serve ChatGPT, Perplexity, and the next entrant with minimal additional work. Agencies managing many merchants feel this most acutely, which is why protocol consolidation cuts their maintenance burden substantially. We think of the protocol as the durable asset and the individual channels as tactical choices on top of it.

How do I know if an AI agent can actually complete a checkout on my store?

You test it, directly and repeatedly, because this is the gap that costs merchants the most. A conformant manifest tells you an agent can read your store. It does not tell you an agent can finish buying from it. According to UCP Checker, which tracks more than 20,000 storefronts and skews heavily toward Shopify, roughly 82% pass full UCP validation, but passing validation and completing a real checkout are two different things, and we have seen the gap between them cause silent revenue loss.

The practical method is to script real agent purchase attempts, not just read-only queries, on each platform you target, and run them continuously rather than once at launch. Watch for the full journey: product surfacing, accurate price and availability at the moment of purchase, successful order creation, and correct policy handling. Pair that with monitoring that alerts you to a surfacing or checkout drop within hours, so a failure does not run for days before anyone notices. In our experience, the merchants who treat checkout verification as an ongoing discipline rather than a one-time check are the ones whose agentic revenue actually compounds.