A shopper asks ChatGPT for a birthday gift under $80 for a sister who likes small-batch skincare. By the time she clicks a store’s ad, she has answered questions about sensitive skin and shipping, narrowed her options, and formed a shortlist. Send her to your homepage and you make her do that work again.

That is the order an online store needs to understand. On Google Shopping, a product listing can answer much of a buying query before the shopper reaches your site. Inside ChatGPT, the conversation does that work first; the ad is a labeled card below the answer, visually separated from what the model wrote. The ad does not become the recommendation, and it is not your storefront. The landing page has to finish a sale that started somewhere else.

Step 1: The shopper arrives with a shortlist

Our gift shopper did not type “buy rosehip gift set $68.” She described a person and a budget. Then she clarified that her sister has sensitive skin and that the gift needs to ship by Friday. By the time a product card appears, those details have ruled out much of the catalogue.

That is the first fact everything else follows from: the shopper has stated her buying conditions before she clicks an advertiser’s card. She may still compare products or change her mind, but she is not arriving as a blank slate. She wants to check whether the store can meet the conditions she has already explained.

Google can do some of this qualifying through the query and the product listing. The difference here is that the shopper has worked through the conditions in a conversation before reaching your page. If the page makes her rediscover the right category, apply a price filter, check ingredients and hunt for a shipping estimate, it discards the useful part of that conversation.

I used to tell clients the product page was for closing. That description is too simple for this path. The chat opens and narrows the search; the page has to show, quickly, that the product fits. Build for the shopper who already knows what she needs to verify.

Step 2: The ad can match the conversation, not join it

Once the shopper has narrowed her options, an advertiser can compete for a card beside that moment. OpenAI describes a relevance-weighted second-price auction that considers conversation context and intent alongside the landing page, title, copy and advertiser-supplied context hints. The buying options are Views on CPM, Clicks on CPC with a suggested $3 to $5 max bid, and Conversions-optimized buying billed as oCPC or oCPM.

The distinction from a keyword campaign matters. You write plain-language context hints at ad-group level, up to 2,000 per group with a $25-per-day minimum per campaign. Those hints are not exact-match instructions and do not guarantee an appearance in a particular chat. For the gift shopper, a useful hint would describe the kind of need the store can satisfy. It would not force the model to recommend a specific set.

Nor does paid placement turn the ad into part of ChatGPT’s answer. Ads run on separate systems, and advertisers cannot shape, rank or alter responses. That boundary is easy to lose in a pitch deck: “reach the right conversation” starts sounding like “become the recommendation.” They are not the same thing.

You can pay for a chance to show a well-matched card underneath the answer. You cannot bid your way into the answer itself. Write the ad for the need the shopper expressed, then make sure the click delivers on it.

Step 3: The landing page inherits the conversation

The card carries an advertiser name and favicon, headline, description, image and landing page. Our gift shopper taps it with three facts still in her head: under $80, suitable for sensitive skin, ships by Friday. The landing page does not receive that conversation as a substitute for good merchandising. It receives a shopper who expects those facts to be easy to confirm.

Start with the destination. If the chat has narrowed the choice to one SKU, send the click to that product page. If it has narrowed the choice to a type but left several plausible sets, use a tight collection page. Three fragrance-free sets under $80 give this shopper a useful decision; an entire skincare catalogue gives her another search task.

The homepage is the weakest destination for this click. It might be a perfectly good introduction to the brand. It still does not tell this shopper whether the gift meets her price cap, skin constraint and deadline. Sending her there asks her to rebuild the shortlist ChatGPT just helped her make.

This is why the landing page does work that a Google results page and a product listing used to split. It must help the shopper recognize the right item and make the remaining purchase decision. Put the relevant constraint in the headline. Make the price and shipping information visible without a hunt. Give her a clear add-to-bag action. If the page is a collection, make the remaining choices small enough to compare.

None of that requires pretending the chat has already sold the product. The shopper still needs proof from the store. The practical test is simpler: can she check her three conditions without starting over?

Step 4: The order and its tracking move to your site

The click does not finish the transaction in the answer. OpenAI introduced Instant Checkout in September 2025, starting with Etsy and built with Stripe on the Agentic Commerce Protocol, then stepped back from that path by March 2026 after only about 30 Shopify merchants went live against a million promised. The path to plan around is a referral: discover in ChatGPT, buy on the merchant site.

For Shopify stores, products can surface with live price and stock; a click opens the store’s normal Shopify checkout in an in-app browser or a new tab, and the store remains the merchant of record. Our gift shopper still chooses the set, pays and places the order through the store. Plan for the order to land in Shopify or WooCommerce, not in the chat.

That also tells you where the page’s job ends. It has to turn the conditions established in the conversation into a confident purchase on your site. The checkout then has to work as it would for another referred shopper. A convincing ad cannot repair a product page that hides the shipping deadline, and a convincing product page cannot record a purchase if the tracking handoff fails.

Where conversion data gets lost

The referral has another consequence: the visit needs to be identified after it leaves ChatGPT. ChatGPT auto-adds utm_source=chatgpt.com, but during an interface rollout those tags briefly appeared on about 5% of answers before recovering to near 99%. Visits without the expected tag can fall into Direct. Paid clicks also carry an oppref reference; relying only on default tagging makes it harder to separate paid from organic visits and send clean conversion data back for oCPC.

The fix is unglamorous. Put your own static UTMs on each ChatGPT landing URL. Keep the pixel and Conversions API firing with the correct purchase value. Build a custom GA4 channel grouping that separates ChatGPT paid from organic. Then check the journey from click to recorded order, rather than assuming a purchase in the store means the ad account can see it.

That work belongs with whoever manages the test, not with a store owner trying to reconcile a report after the fact. Without it, a working sale can appear in the wrong traffic bucket, while the paid campaign lacks the signal needed to judge what happened. Tag the click before arguing about the bid.

Step 5: Compare the jobs, not just the channels

On Google Shopping, a query can carry a great deal of intent. Someone searches for a model number and size; Google matches the product feed, and conversion history helps determine who shows. Our gift shopper did not start with a model number. She described a recipient, a budget and a deadline, then used the conversation to narrow the field.

That difference changes what your catalogue and page need to prove. A Shopping listing may show the product, price and merchant while the shopper is still scanning results. The ChatGPT ad appears after the shopper has discussed what would fit. Its landing page must make the match visible when she arrives.

Product data matters in both paths. Organic product picks are ranked independently of ads or partnerships. When several merchants sell the same item, availability, price, quality and whether a seller is the maker or primary seller can affect the order. Bad price or stock data can leave a product skipped or misrepresented. For our gift set, the page cannot promise Friday delivery while the underlying product information says otherwise. Keep the catalogue honest before trying to improve the ad.

Do not fund a ChatGPT test by cutting Shopping or Performance Max simply to make room for it. Use incremental budget for 30 to 60 days and read blended revenue against a simple geo holdout. Spend reporting can lag seven to eight hours, so do not mistake an incomplete day for a verdict. The longer test window serves a different purpose: it gives you time to judge the additional revenue rather than killing the experiment on week-one last-click ROAS.

What this means for who you hire

If the sale happens on your product page and the purchase signal comes from your tracking, managing these ads is not just a buying job. I would ask a manager to account for five pieces of work:

  1. Keep the feed accurate on price, stock and identifiers so the product information can be trusted.
  2. Write and prune plain-language context hints so the ad has a chance to meet relevant conversations.
  3. Match each ad group to a product or tight collection page that reflects the shopper’s constraints above the fold.
  4. Tag every landing URL and keep purchase value firing cleanly through the pixel and Conversions API.
  5. Read the test against blended revenue and a holdout instead of strangling it on week-two last-click ROAS.

That is also an interview. Ask what they would change on the landing page before spending a dollar. Ask how they would organize hints for a 40-SKU catalogue versus a 4,000-SKU one. Ask what happens to their reporting when default tags drop. The answers should describe work in the store and in the measurement setup, not just settings inside an ad account.

If the answer is a course certificate and a speech about prompts, you hired a course, not a manager. What the deck calls “ChatGPT expertise,” I call familiar ecommerce jobs moved closer to the conversation that starts the sale. Hire for the handoffs, not the terminology.

Who should wait before testing

Not every store should test yet, and saying so is part of the job. If your feed cannot hold accurate price and stock today, wait. A context hint will not fix a product that appears unavailable or points to the wrong variant.

If your checkout conversion rate is under 1% on warm traffic, wait. The page and checkout already have a problem to solve, and a mid-conversation click is not a reason to expect them to solve it on their own. If you sell one $24 accessory with no bundle and no repeat purchase, the math of a $3 to $5 click is hard to make work at small budgets. The reporting lag does not change that math; it just makes a quick read less trustworthy.

Fix the catalogue, fix tracking and prove that the product page converts from Shopping first. Then test ChatGPT with incremental money and a holdout, not by starving what already sells. A new source of clicks is not a repair plan for a broken purchase path.

Try the model on a different aisle

Take trail shoes. A shopper wants waterproof shoes in men’s size 11, wide fit, under $150, before a Friday trip. ChatGPT asks about width and trail use, rules out two popular models that run narrow, and leaves one category standing. A card underneath shows a store’s wide-fit waterproof model at $139 with Friday delivery.

Work forward from the conversation. The shopper should click through to that variant page, not the footwear homepage. Size 11 should be preselected; the width note, price and delivery date should be visible above the fold; reviews filtered to wide feet should help answer the remaining fit question. Checkout still happens in Shopify. The store’s UTMs mark the visit as ChatGPT paid, and the pixel records purchase value.

Now work backward from a failure. If the feed says size 11 is out of stock, the product may never surface. If the ad earns a click but the page opens on a general size selector, you paid for a qualified conversation and asked the shopper to repeat it. If the order lands but the visit loses its paid attribution, you cannot read the test cleanly. Each failure sits at a different handoff.

I mapped where the trail goes cold after the click in more detail. The pattern is the same: the auction can bring a shopper to the door, but the page and purchase signal determine whether you can turn that visit into a sale you can measure. That is why groas treats Search Ads across Google, Bing and ChatGPT as one service for a flat $1,499 a month, with around-the-clock execution, a named strategist and actions logged with their reasoning. For a store, that fits the work better than a monthly report and a hint list.

Build the page to finish the conversation ChatGPT started. Keep the feed honest, tag the click yourself and judge the test on blended revenue. Do that for the gift set or the trail shoe, and within 60 days you can tell whether your catalogue has a second shelf worth paying for.