Paid search needed a third off to close. AI referrals needed a quarter.

Most channel reporting stops at revenue. Revenue is the number before the discount comes out. If you want to know which traffic is actually worth buying, the more useful question is what that traffic costs you at the checkout, and that number moves more by source than almost anything else you measure.

Here is one store's answer. A direct to consumer brand, one month of orders, counted at the order level. The brand is not named. The share of orders that needed a discount to close, and how deep the discount ran:

• Paid search: 93% of orders used a discount, averaging about a third off
• Google organic: 81% of orders, at about 30%
• AI assistants: 69% of orders, at about 25%
• Direct and everything else: 27% of orders, at about 15%
• Whole store: 45% of orders, at about 21%

Read the first line and the third line together. On paid search the store pays for the click, then pays again at the checkout, on nine out of ten of those orders. The traffic that arrived from an AI assistant cost nothing on either side and closed at ten points less off.

Over the period the store gave back about 22% of gross sales in discounts. Where that lands is a channel decision, and almost nobody makes it deliberately.

They cannot make it deliberately, because they cannot see it. A discount rate by source means matching every order back to whatever sent the buyer, and that join sits between two systems that do not talk to each other. We run it for every store we work with. It is usually the first number that changes how a team thinks about its media plan.

Why the same buyer needs a smaller discount

Because it is not the same buyer arriving.

Someone who found you through a paid ad is meeting you for the first time, mid scroll, with no reason to believe you yet. The discount is doing the persuading. Someone who arrives from an AI assistant has already had the comparison done for them. The assistant weighed the options, read the reviews, and handed over a reason before the click ever happened.

A buyer who was persuaded before the click needs a smaller discount after it.

The price tag has less work to do, because something else already did it.

That turns your discount problem into a visibility problem

If the buyer who arrives persuaded is your best margin buyer, the question stops being how to discount better. It becomes two questions instead. How do you get named by the assistant, and what does it say about you when it does.

That is the work eLLMo does. We build the structured data and semantic alignment layer underneath a catalog, so the engines doing the recommending can read your products, your reviews, your pricing and your availability, and can quote them to a shopper without having to guess.

Getting named is the first half. The brief is the second, and it matters more. An assistant does not pass on your headline. It passes on a short set of claims it thinks it can defend, built out of whatever it was able to read about you. So the brief your buyer arrives holding is not luck. It is an input, and you can write it.

That is why the layer carries more than a specification list. It carries the comparisons a shopper actually makes, including the ones where you are not the right answer. A recommendation that says who a product is not for is a recommendation a shopper believes, and it sends you the buyer who fits instead of the buyer who sends it back. That buyer is the one who arrives without needing a third off.

Those are two levers on one outcome. More of the traffic that arrives already persuaded, and a better reason in its hand when it gets there. Both of them show up in the same place, which is the discount you did not have to give.

The rest of the funnel agrees

Over the same period, buyers referred by AI assistants added something to the cart in about 23% of sessions, against about 15% for everyone else. That is a gap of 1.5 times. They reached the checkout more often, and once there, slightly more of them finished.

The question worth taking back to your own dashboard

What is your discount rate by source?

Most teams cannot answer it. They can give you cost per acquisition by channel and revenue by channel, and then the discount sits in one store wide line that averages a buyer who needed 35% off with a buyer who needed nothing at all. Those two are not the same customer and should not be bought at the same price.

eLLMo answers both halves of that question. We show you which traffic is costing you margin at the checkout, and we build the layer that brings you more of the traffic that is not, briefed the way you would brief it yourself. If your discount line is growing faster than your revenue, start with the split by source, then go and win the channel that does not need the discount.

*Related Links: AI traffic went from converting 38% worse to 42% better in twelve months, Discovery has left your website. The data on where it went..*