Walmart spent millions to prove what no one wanted to admit

Picture the scene: a dark conference ballroom, a giant screen, and a line on a slide that made a lot of people in that room shift in their seats. Daniel Danker, an EVP at Walmart, is on stage at the Morgan Stanley Technology, Media & Telecom Conference in early 2026. He's talking about the future of shopping with AI.

For months, the story everyone wanted to believe was simple: people would talk to an AI, pick a product, and buy it right there in the chat. No website. No cart. No "add to bag." Just, "Yes, I'll take it," and the AI would handle the rest.

OpenAI built its version of that dream and called it Instant Checkout. Late in 2025, it switched it on with Walmart as the anchor partner and tens of thousands of products wired into ChatGPT. The pitch: instead of sending shoppers to Walmart.com, why not let them finish the purchase inside the chat window?

The reality was harsher.

On stage, Danker shared the numbers. When shoppers bought inside ChatGPT, those purchases converted at about one‑third the rate of people who simply clicked through and bought on Walmart.com. The same brand. The same products. The same prices. The only difference was where the checkout happened.

He called the in‑chat experience "unsatisfying." That single word did a lot of work. It meant the friction was real, and it wasn't just a minor UX bug. It was a sign that the whole idea of "buy inside the bot" was out of step with what real people were actually comfortable doing.

By late March 2026, OpenAI quietly pulled the plug on Instant Checkout and shifted to a new model: retailer‑built experiences where brands kept control of the purchase flow and the data, even if discovery still started inside ChatGPT.

Walmart didn't fail at AI commerce. It ran the most expensive "does this really work?" test anyone has done so far—and the answer was clear.

AI is not where people want to pay. It's where they want to figure out what to buy.

Buying inside a chatbot felt wrong. There's a reason for that.

You don't need a psychology degree to understand why this experiment struggled. You just have to think about how you actually use AI tools.

Most people open ChatGPT, Gemini, or Perplexity to think out loud. To compare. To explore options. To check facts. They treat these tools like a smart friend with a whiteboard, not a cashier waiting behind a counter.

So when the same box that is helping you weigh pros and cons suddenly wants your card details, something in your gut tightens. There's a quiet sense of, "Hold on, who is really in charge here?"

There are a few things happening in that moment:

• You trust AI to inform you, not to transact for you. You want help choosing, but you still want to press the final "buy" button in a place that feels like a real store, not a blank chat screen.
• When checkout moves into the chat, the brand loses its "owned experience." You don't see the full Walmart interface, the upsells, the tracking, or the post‑purchase flow. You feel that loss of structure, even if you can't name it.
• The AI becomes the face of the transaction, not the brand. That's fine for a question. It's not fine when there's money on the line.

Marketers have a name for this: the "owned experience" principle. Conversion happens where brands own the surface. That's why, after Instant Checkout stumbled, OpenAI didn't double down on being the cashier. It invited retailers to build their own branded apps inside ChatGPT—mini stores that still bring buyers back into an environment the brand controls. Target, Sephora, Nordstrom, Best Buy, Home Depot, Wayfair and others are already going down that road.

There is one big exception: TikTok Shop.

TikTok Shop turned a scrolling feed into a $30‑plus‑billion global checkout machine by making buying feel like a natural part of the entertainment. You're already watching creators you trust. The store is just an extra tap on top of the trust you already have.

TikTok is the experience. The video, the social proof, and the checkout all feel like one continuous path. A chatbot is different. ChatGPT is a tool. It's a place you go to think. No one expects a tool to suddenly become a store.

TikTok showed that integrated discovery and purchase can work when it fits the platform. Walmart's test with Instant Checkout showed that AI chat, at least right now, is not that kind of platform.

The lesson: AI is a discovery layer. Brands that try to turn it into a full transaction layer keep hitting the same wall.

The money isn't in buying inside AI. It's in being found inside AI.

If checkout isn't the opportunity, what is?

Look at how people are actually using AI today. McKinsey reports that around half of consumers now use AI‑powered search to help make buying decisions. At the same time, agencies and analytics firms are seeing AI‑referred visitors convert several times better than regular search traffic—often close to four times as well.

It makes sense. By the time someone lands on a page because an AI recommended it, three things are already true:

• They know what kind of product they want.
• They've already ruled out some bad options.
• They trust the AI enough to click the link it gives them.

That click is different from a random Google search visit. It's not a curious stranger. It's a warm hand‑off.

This is the heart of AI product discovery.

AI product discovery is the moment a person asks, "What's the best X for Y?" and an AI answers with specific products and sources. It's the new top of the funnel. Instead of typing "best camera for YouTube" into Google and skimming a wall of blue links, people type that same question into AI and expect a short, confident path forward.

Walmart's pivot shows it understands this. When it brought its Sparky assistant into ChatGPT and Gemini, it didn't try to process orders there. It tried to be the voice inside AI that helps shoppers discover what Walmart already sells. Discovery happens in the AI. Purchase happens on Walmart turf, where Walmart owns the experience.

Now imagine the same pattern, but for you as an affiliate.

In the old world, your job was to rank a page like "best vacuum cleaner 2026" on Google and fight for the click. In the new world, your job is something more specific and more powerful: to be the page the AI cites when someone asks that question.

When that happens, the traffic that comes to you is pre‑sold. The visitor didn't just stumble onto your site. They were sent. The AI has already whispered, "Start here."

The difference in intent is huge. One AI citation can be worth more than thousands of low‑intent visits.

The affiliate playbook just got rewritten (and most people don't know it yet)

Most affiliates are still playing the old game. They're chasing rankings for "what is" guides and "best of" lists that used to soak up easy traffic. The problem is that AI now answers those questions directly, without needing to send anyone to you.

Search results already show this. AI Overviews and similar features appear at the top of more and more queries, answering the basic question in a neat box before a user ever reaches the links below. The click‑through rates on those old top‑of‑funnel pages are sliding.

But here's the twist: for certain types of pages, the opposite is happening.

When AI needs to show its work—when it has to compare, explain prices, walk through trade‑offs, or cover edge cases—it looks for structured, trustworthy pages to cite. Agencies studying these patterns see a clear trend: content with FAQ sections, comparison tables, specs, and clear decision frameworks gets cited far more often than long, unstructured essays.

Think about the difference:

• A 2,000‑word article titled "What Is a Standing Desk?" that just explains definitions and benefits.
• A 2,000‑word article that compares three standing desks, shows prices "as of Q2 2026," lists pros and cons, answers "Who is this best for?" and explains how they differ.

AI doesn't need the first page. It can generate that kind of summary itself. It does need the second, because that page contains concrete facts and a useful structure to lean on.

Marketers have started calling this shift the "Citation Economy." In this new reality:

• The goal isn't just to get traffic. It's to become the source AI trusts enough to cite by name.
• A citation is more than a mention. It's a transfer of trust from the AI to you.
• Citations drive fewer visits than broad rankings, but those visits arrive with intent you could only dream about before.

There is one more important rule: AI systems don't pull citations from nowhere. Analyses of Google's AI Overviews show that almost all cited URLs come from pages that already rank in the top 10 organic results. That means AEO—Answer Engine Optimization—does not replace SEO. It stacks on top of it.

You still have to rank. But ranking alone is no longer the finish line. It's simply your ticket into the pool AI draws from.

So what does this mean for the content you create?

It means "what is" content and shallow "best of" lists are a shrinking asset. It means the opportunity is in bottom‑funnel content: comparison pages, pricing breakdowns, spec tables, FAQ pages, and decision frameworks that help a buyer in the last 15% of their journey.

That's where AI needs you.

Here's the exact format AI cites. Build more of this.

If AI is the new top of the funnel, then your pages have to be built for how AI reads, not just how humans skim.

The good news: the formats that work for AI also make life easier for your readers.

Start with FAQ sections.

Every serious comparison or review page should have a block of questions and answers that match how people actually ask things out loud: "Is X worth it?" "What's the difference between X and Y?" "Which one should I buy if I travel a lot?" On the back end, you mark these up with FAQ schema so AI systems can parse them quickly.

Next, comparison tables.

Forget screenshots of tables. AI can't read those. Use real HTML tables with clear headings: price, key specs, "best for," and even a "bottom line" row that sums each product up in one clean sentence. Those one‑sentence lines are often what AI will lift and reuse in its answer.

Then, spec tables.

For each product, list the simple facts: model name, size, weight, battery life, warranty, compatibility, and so on. This sounds boring until you realize this is exactly the kind of information people keep asking AI about. "Will this fit in my carry‑on?" "How long does the battery last?" "Is this compatible with X?" The clearer and more complete your specs, the more useful your page is to the AI.

Your headings should mirror human questions.

Instead of "Features of the DJI Mini 4 Pro," write, "What can the DJI Mini 4 Pro actually do in 2026?" When your H2s look like the questions people type, you make it obvious to the AI what each section is for.

Under all of this is a simple shift in mindset: from keywords to entities.

AI doesn't count how many times you used a phrase. It checks whether the important "entities" are present and clear. For each product, it wants to know: what is it, who is it for, what problem does it solve, how is it different from alternatives, how much does it cost, where can you buy it? Cover those fully, and the keywords mostly take care of themselves.

If you want a quick checklist, it looks like this:

1. Add FAQ schema to every serious comparison and review page.
2. Turn all comparison tables into real HTML. No images.
3. Add a "X vs Y" section to every review where it makes sense.
4. Include a "Who should buy this / who shouldn't" block with simple bullets.
5. Show prices with time stamps like "as of Q2 2026" so AI knows how fresh your data is.
6. Implement product structured data so search engines and AI can cleanly identify each item.

Your page doesn't have to be the most visited. It has to be the most useful—and the most readable—for the systems now deciding what to recommend.

You know what to build. The real problem is scale.

By now, the pattern is clear.

• The dream of frictionless AI checkout hit a wall in months.
• The real leverage sits in discovery: being the page AI points to when someone is ready to decide.
• The pages that get cited most are structured, factual, and focused on decisions, not definitions.

You don't need a Walmart budget to play this game. But you do need volume.

One properly built comparison page, with FAQ schema, HTML tables, clean specs, and helpful "who should buy this" sections, can easily take half a day to create if you do it by hand. Do that once, and it feels great. Try to do it 50 times across all your categories, and you run into a wall of your own: time.

That's the bottleneck most affiliates are feeling right now. Not "What should I build?" but "How do I produce enough of this, at a standard high enough that AI will trust it?"

In a world where AI is the new front door, your edge is not the single clever article. It's the depth and structure of your entire library.