Google Ads Direct Offer AI Mode: How to Prepare Now

Google Ads Direct Offer AI Mode placements put a discount in front of a shopper at the moment they're deciding. For e-commerce advertisers, that makes the Direct Offer asset a pricing decision first and an ad format second. Our position is to prepare now, but don't switch it on account-wide. Set a margin floor per product group, clean up your feed and promotion data so every surface shows the same price, fix the page the click lands on, and test on a margin-safe subset of SKUs before you scale. If you can't say what a 10% offer costs you on your top 50 products, you aren't ready to run one inside an AI answer.

The rest of this piece explains why we land there and gives you a sequence to work through before the asset reaches your account.

What the Direct Offer asset changes in AI Mode

The name tells you most of what matters. An offer, attached directly to your ad, shown inside AI Mode. AI Mode is Google's conversational search experience. The shopper isn't scanning ten blue links and a shopping carousel. They're asking a question, reading a synthesized answer, and following up. An ad that carries an offer shows up inside that exchange.

We won't restate Google's specs here. Eligibility rules, supported offer types, placement behavior, and reporting fields are the kind of detail that shifts during a rollout. Treat Google's current Ads Help documentation as the source of truth on mechanics. What we can tell you is what this changes for the advertiser, because that part doesn't depend on the fine print.

Three things shift.

The offer becomes part of the answer. In a classic search results page, a promotion competes for attention with everything around it. In a conversational answer, the shopper has already narrowed their intent through their own questions. An offer that appears at that point lands at the decision stage, which is exactly where discounting does the most damage to margin if you didn't need it to win the sale.

Your price data gets compared in context. AI Mode pulls from product data, merchant information, and web content to build an answer. If your ad says one thing, your Merchant Center promotion says another, and your product page says a third, the shopper sees the mismatch in a single view. So does Google's system.

Attribution gets murkier. Conversational sessions don't map cleanly to the last-click model that a lot of e-commerce reporting still runs on. A shopper might see your offer in AI Mode, leave, and come back through branded search or a direct visit two days later. If your measurement can't connect those, you'll undervalue or overvalue the asset depending on how your conversion settings are configured.

We covered the broader budget and landing page side of AI Mode advertising in our guide to Google AI Mode ads and PPC strategy. This piece narrows in on the offer layer.

Why we treat it as a margin decision first

Every discount you run splits into two groups of buyers. The first group wouldn't have bought without the offer. The second group would have bought anyway and now pays less. The offer earns money on the first group and loses money on the second. Whether it's profitable comes down to the ratio between them.

Here's where AI Mode makes that ratio harder to win. A shopper deep in a conversational session about, say, insulated bottles for hiking has already done the research. They're closer to buying than someone typing a two-word query. That pushes more of your offer spend into the "would have bought anyway" group. You're paying a discount to people who were already at the register.

That doesn't mean Direct Offers lose money. It means the math is specific to your catalog, your margins, and how price-sensitive your buyers are in each category. A store selling commodity accessories with five competitors on the same answer faces different math than a store selling a branded product nobody else carries. The commodity seller probably needs the offer to win. The branded seller probably doesn't.

So before anyone opens the campaign settings, the question for your team is: which products can absorb a discount, how deep, and what do we expect to gain?

Step 1: Build an offer ladder with a margin floor

An offer ladder is a short table that tells your team, and your agency, exactly what's allowed. Build it at the product group level, not the SKU level, unless your catalog is small enough to manage individually.

For each product group, record:

1. Contribution margin after fulfillment. Not gross margin. Include shipping cost, payment processing, packaging, and average return rate. A product with a 45% gross margin and a 20% return rate isn't a 45% margin product.
2. Margin floor. The lowest contribution margin you'll accept on a discounted sale. This is a business decision. Write it down so nobody makes it in the moment.
3. Maximum offer depth. The discount that takes you exactly to the floor. Anything deeper is off the table.
4. Competitive pressure. Whether you face direct price competition on identical products in this group. Groups with no direct competition should start with no offer or the shallowest one.
5. Inventory position. Overstocked groups are the natural first test. You were going to discount them anyway.

Once this table exists, the Direct Offer asset stops being a judgment call. You know which groups qualify, how deep to go, and which ones to exclude entirely. It also gives you a clean way to evaluate results later, because you set the expectation before the data came in.

The common advice we push back on here is "start with your bestsellers because they have the most volume." Bestsellers usually sell because people already want them. They're the most likely group to hand discounts to buyers who didn't need one. Start with products where you have margin room and a reason to move units.

Step 2: Make your feed and promotions agree with each other

AI Mode answers draw on structured product data. If that data is inconsistent, your offer either won't serve or will serve alongside conflicting information that erodes trust.

Audit these before launch:

• Price consistency. The price in your product feed, the price on the product page, and the price in your structured data markup should match exactly. Check for variants. A feed that sends the base price while the page defaults to a larger size at a higher price creates a mismatch.
• Sale price fields. If you already run sale pricing through your feed, decide how a Direct Offer interacts with it. Stacking an offer on top of an existing sale price can blow through your margin floor without anyone noticing.
• Merchant Center promotions. Existing promotions with loose end dates or broad product eligibility are a common source of overlap. Inventory them and close anything that's stale.
• Availability. An offer on an out-of-stock product wastes the click and frustrates the shopper. Make sure your inventory sync runs often enough that availability in the feed matches the store.
• Product titles and attributes. Conversational queries are specific. A shopper asking for a "32 oz stainless bottle with a straw lid" needs your feed to carry size, material, and lid type as attributes, not buried in a description. Thin attributes mean your product doesn't match the question, and the offer never gets a chance to show.

This overlaps heavily with the feed work required for Google's agentic checkout features. If you've started on that, you're partway there. We laid out the feed and checkout requirements in our piece on Google Universal Cart for retailers, and the same feed hygiene applies here.

Step 3: Fix the page the offer lands on

A shopper who clicks an offer in AI Mode arrives expecting to see that offer, immediately, on a page that loads fast and checks out cleanly. Check for three failure points before you send that traffic.

The offer doesn't appear on the landing page. The ad promised 15% off. The product page shows full price with a code field buried in the cart. The shopper assumes the offer was bait and leaves. Make sure the discount is visible on the product page itself, applied automatically where your platform allows it, and confirmed in the cart before payment.

The page is slow. We've inherited a Magento store whose checkout took 11 seconds to load because nobody had touched the database indexing in three years. Paying for a high-intent click from an AI answer and then making the buyer wait that long at checkout throws away the click and the discount together. Check your product page and checkout load times on mobile, on a real connection, before you send paid traffic.

The store is carrying weight it doesn't need. We regularly find 40 to 70 extensions on Magento stores that need maybe 15. Each one adds scripts, database queries, and potential conflicts with promotion logic. A discount extension fighting with a cart extension is a classic source of offers that display on the page but don't apply at checkout. If your store has accumulated extensions over years of projects, audit them before you test a new offer type.

And if your store still runs on Magento 1, the conversation about Direct Offers comes second. Magento 1 reached end of life in June 2020 and gets no security patches. We still find stores running it, completely unpatched. Treat any Magento 1 store as already compromised. The move to Magento 2 comes before any new paid traffic.

Step 4: Measure incremental margin

Platform-reported return on ad spend will make Direct Offers look good almost regardless of whether they are. The platform credits the conversion to the ad that showed the offer. It doesn't know whether that buyer would have purchased at full price through another path.

Set up measurement that answers the right question:

• Track offer redemption separately. Use a distinct promotion identifier or code for Direct Offer traffic so you can see exactly how many orders carried the discount and what it cost in margin.
• Report contribution margin per order. Revenue doesn't tell you whether the offer worked. Margin does. Pull order-level margin, net of discount, into your reporting.
• Hold out a control group. Run the offer on some product groups and not on comparable ones. Compare conversion rate and margin between them over the same period. This is the closest you'll get to measuring incrementality without a formal lift study.
• Watch branded and direct traffic. Conversational sessions often end with the shopper returning later through a different channel. If branded search or direct traffic rises in the product groups running offers, some of that likely traces back to AI Mode exposure. GA4 already struggles to classify AI-driven visits. We covered the attribution side in our audit guide for ChatGPT traffic showing as direct in GA4, and the same diagnostic approach applies to AI Mode.

If you can't build a holdout, at minimum compare margin per order on offer traffic against margin per order on the same products from non-offer traffic. If offer traffic converts at a similar rate with lower margin, the discount is paying people who would have bought anyway.

Step 5: Keep your organic presence in the answer

Paid placements inside AI Mode sit alongside organic content in the same answer. A shopper who sees your offer and also sees your product cited in the organic portion of the response gets two signals that you're a credible option. A shopper who sees only the ad gets one.

Our own data on AI Overviews points the same way. In a July to September 2026 analysis of client sites we manage, covering 2,350 tracked keyword snapshots across 314 keywords where an AI Overview appeared, the AI Overview cited the site 73% of the time when it held organic position 1. That dropped to 49% at positions 2 to 3, 42% at positions 4 to 10, 9% at positions 11 to 20, and 3% when the site ranked beyond position 20 or didn't rank. That's a correlation, not proof of cause, and AI Overviews aren't identical to AI Mode. We break down the full pattern in our analysis of AI Overview citations and organic rank.

The practical implication is that buying your way into AI Mode doesn't replace ranking. If your category pages and product pages don't rank, you lose the organic citation and you're relying on the paid offer to do all the work. That's the scenario where you'll feel pressure to discount deeper.

So run the SEO work in parallel. Strong product and category pages, complete structured data, and comparison content for the queries where shoppers weigh options all keep you present in the organic layer of the answer.

How we'd roll it out: NOW, NEW, NEXT

We frame client roadmaps in three stages. Here's how that maps to Direct Offers.

NOW: groundwork that pays off regardless of the asset. Build the offer ladder with margin floors. Audit feed, structured data, and promotion consistency. Close stale Merchant Center promotions. Measure product page and checkout load times and fix the worst offenders. Set up promotion-level tracking and margin reporting. Every one of these improves your current Shopping and Performance Max campaigns too, so none of this work is wasted if you decide not to run Direct Offers at all.

NEW: a controlled test. Once the asset is available in your account, pick two or three product groups from your ladder. Prioritize overstocked inventory and categories with direct price competition. Run the shallowest offer depth that's still meaningful to a shopper. Hold out comparable groups as a control. Run long enough to get a clear read on conversion rate and margin per order, which depends on your order volume. Low-volume stores need longer test windows.

NEXT: scale what earned its place. Expand to product groups where the test showed incremental margin, not just incremental revenue. Adjust offer depth per group based on what you learned. Feed the results back into your offer ladder so it reflects real data instead of the assumptions you started with. Revisit quarterly, because competitive pressure and inventory positions change.

If you'd rather not run this sequence in-house, this is exactly the scope our google ads management team handles, with the feed, landing page, and Magento work done by the same people who run the campaigns.

Where we could be wrong

Our position assumes that shoppers in AI Mode are further along in their decision than shoppers in a standard search, and that offers at that stage pay a larger share of discounts to buyers who'd have converted anyway. If Google's placement logic shows Direct Offers earlier in the conversation, at the research stage rather than the decision stage, the offer would reach more undecided shoppers and the case for broader use gets stronger.

The test structure above is built to catch that. If your holdout groups show offer traffic converting at a meaningfully higher rate than control without a matching drop in margin per order, the asset is reaching people who weren't going to buy. In that case, expand faster than we'd suggest by default. Let your own data overrule our starting assumption.

The work that comes before the asset

Direct Offers in AI Mode reward stores that already know their margins, keep clean product data, and run fast checkouts. They punish stores that discount by instinct and send paid traffic to slow pages with conflicting prices. Nothing in this sequence requires waiting for Google. Build the offer ladder, fix the feed, speed up the checkout, and set up margin-level measurement now.

We run strategy, SEO, paid media, and Magento development under one roof, so the team adjusting your offer depth is the same team fixing the checkout it lands on. No handoff between a PPC shop and a dev shop. Clients stay with us for years because the work moves the numbers they care about.

Request a proposal and we'll map your catalog against an offer ladder, audit your feed and checkout, and tell you which product groups are ready to test.

---