ChatGPT Ads mistakes in e-commerce and how to fix them

Short answer: E-commerce stores usually get ChatGPT Ads wrong in three ways. They run it like Google Search with exact keywords, they judge it only by last-click sales in GA4, and they send clicks to a generic homepage. ChatGPT picks ads based on the conversation, so your ad, landing page and measurement all have to fit that conversation.

I run ChatGPT Ads campaigns alongside Google Ads and Meta Ads for e-commerce clients. Most of the early disappointment I see comes from habits brought over from other platforms, not from the channel itself. If you need the basics first, read how ChatGPT ads work and which businesses they suit. This article is about what goes wrong after launch.

Why ChatGPT Ads are not Google Search with a different logo

In Google Search you bid on keywords, and exact match gives you control over which query shows your ad. ChatGPT Ads work differently. According to the OpenAI Help Center (Ads in ChatGPT: The Basics), ads appear below the ChatGPT answer. They are selected using four signals: the context and intent of the conversation, your landing page, your ad text and your “context hints”.

Context hints are topics or keywords you set at ad group level. They are not exact match, and delivery is not guaranteed. So the habit of building hundreds of tightly matched keywords and expecting precise control does not carry over.

The auction is a second-price auction weighted by relevance. In practice, how well your ad fits the conversation affects both whether it shows and what you pay. Relevance is something you can work on, not a nice extra.

What to do instead

1. Build ad groups around buyer situations, not keywords. “Choosing a first espresso machine” and “replacing a broken one” are two different conversations.
2. Write a handful of context hints per ad group, each one a topic that describes that situation.
3. Write the headline and body text so they answer the question the person was asking. The ad text is itself a selection signal.
4. Keep the landing page on exactly that topic (more on this below).
5. If you are testing for sales, use the Clicks objective (CPC) with a maximum bid at ad group level, not the Reach objective (CPM).

Why last-click sales in GA4 make ChatGPT Ads look worse than they are

A common story: the owner opens GA4 two weeks after launch, checks last-click revenue, sees very little and switches the campaign off. The real problem is how it was measured. Someone who clicks an ad under a ChatGPT answer is often still comparing options.

They may come back days later through brand search, a direct visit or your newsletter. Last-click attribution gives the whole sale to that final visit and nothing to ChatGPT.

That is no reason to trust any single number. It is a reason to look at three views:

• Platform view: Ads Manager reports impressions, clicks, spend, CTR (click-through rate), average CPC or CPM and conversions. Use it to optimise inside the platform.
• GA4 view: UTM parameters are kept, so tag every ad, for example utm_source=chatgpt and utm_medium=cpc. Then look at sessions, engagement and conversion paths, not just last-click revenue.
• Business view: MER (marketing efficiency ratio) is total revenue divided by total ad spend across all channels. If MER holds steady or improves after you add ChatGPT spend, the channel is probably earning its place.

Before launch, agree which view decides whether the test continues and how long it runs. Otherwise the decision gets made on whichever report someone happens to open first.

Why a generic homepage wastes ChatGPT Ads clicks

The landing page helps decide whether your ad is shown, and it is also where the conversation carries on. Someone who just asked ChatGPT which running shoes suit flat feet does not want a homepage slider with your autumn sale. They want the answer to their question, in your shop.

Good landing pages for ChatGPT Ads in e-commerce are usually one of these:

• A filtered category page that matches the situation (“running shoes for flat feet”, not “all shoes”).
• A buying guide or comparison page with the products built in and a clear path to the cart.
• A single product page, but only when the conversation is about one specific product type and you are clearly the right answer.

Check the page on mobile. Make sure prices show VAT the way your market expects, and that the consent banner does not hide the content. If GA4 shows ChatGPT sessions ending within seconds, suspect the page before the ad.

Symptom, cause, fix: a quick diagnostic table

Symptom	Likely cause	Fix
Few or no impressions	Context hints too narrow, ad text and page do not match the topic, or the bid is too low	Broaden hints into topics, align the text and page, review the maximum bid
Clicks but very short sessions	Homepage or broad category used as the landing page	Send each ad group to a page that answers its conversation
Ads Manager shows conversions, GA4 shows almost none	Last-click attribution, or UTM tags missing or overwritten	Confirm UTM tags reach GA4, check conversion paths, judge on MER
CPC far above what your margin allows	Low order value product, recommended bid copied without doing the maths	Calculate break-even CPC, test only higher-value categories
Ads showing next to loosely related conversations	Broad hints or several topics mixed in one ad group	Split ad groups by buyer situation
Good CTR, no sales	The ad promises something the page does not deliver	Make the headline, body and page say the same thing

What I see in the accounts I run

The pattern I see most often is not a bad campaign. It is a reasonable campaign judged by the wrong report. When I put ChatGPT UTM traffic next to the conversion paths in GA4, ChatGPT tends to show up early in the path, with brand search or a direct visit closing the sale. If you switch it off after two weeks of last-click data, you never find that out.

The second thing: the ads that work best for me read like the next sentence of the answer, not like a banner. The person has just read a plain-language explanation and your ad sits directly below it. A headline that names their situation fits that moment better than a brand slogan.

The ad unit is small: advertiser name, favicon, headline, body text, image and landing page. Every one of those needs to point at the same buyer situation.

When ChatGPT Ads are not worth it for your store

Do the maths before you launch. OpenAI recommends a starting CPC of $3-5. Compare that with your break-even CPC, the most you can pay for a click without losing money on the first order:

Break-even CPC = average order value × conversion rate × gross margin

An illustrative example: say your average order is €60, your site converts 2% of visitors and your gross margin is 40%. Your break-even CPC is 60 × 0.02 × 0.40 = €0.48. At a starting bid several times higher, that store loses money on first orders, and only repeat purchases could change the picture.

Hold off, or test something else first, if:

• You sell low-value products that people rarely buy twice.
• Your site already converts poorly from Google and Meta. Then the problem is the shop, not the channel.
• GA4 tracking and consent are not set up properly. Fix those first or you will not be able to judge anything.
• Your product is aimed at teenagers. Ads are not shown to people under 18, and they are also not shown to Plus, Pro and Business users.

The stores that fit best sell higher-value, considered purchases: products people ask for advice about before they buy.

What to do first

Pick one category where people really do ask for advice before buying and where your break-even CPC is above the recommended starting bid. Build one ad group per buyer situation, give each a matching landing page and clean UTM tags, then agree how you will judge the test before you spend the first euro.

If you want this set up inside your own accounts, see my PPC and performance marketing services or what working with a PPC specialist involves. Email me and I’ll reply within 24 hours.