// use case · E-commerce & DTC
Before ordering, a shopper tells ChatGPT what they want to do with the product, how much they want to spend and which brand they already know. Your product slips into that conversation, and the account is judged on margin per order, never on the click.
looking for a mattress for a couple, one has back pain and the other gets too hot at night
Two different constraints: back support and temperature regulation. A firm hybrid mattress with pocket springs often answers both, as long as the comfort layer isn't a dense foam. Check the home trial period and the return terms before looking at the price.
Hybrid, 100-night trial
Firm support, breathable foam, free returns.
your-brand.com
// why this use case
None of the native report's six columns (impressions, clicks, spend, CTR, CPC, CPM) says whether an order was profitable. Yet every bidding decision an online retailer makes comes down to net margin, once product cost, logistics and returns have been taken out.
For a brand that lives on reorders, a first order with no margin is still a good customer purchase, provided the second one arrives. That is why we split new and returning customers from day one: CAC is computed on the former, LTV on the latter.
With entry CPCs observed between $3 and $5, a basket under €100 leaves little room to pay for the click and then absorb a return. Above it, the maths holds. It is the first question we put to a brand, before we even talk about ads.
The auction is second-price and weighted by relevance: an ad that fits the conversation better can win with a lower bid. Creative therefore becomes a CAC variable, and gets argued about as one rather than on taste.
// intents
Typical prompt
"A backpack for cycling to work in the rain, with a sleeve for a 15-inch laptop"
What the ad has to do
Match one precise product to the described use: the ad names the use, and the landing page shows the product, not the collection.
Typical prompt
"A bean-to-cup espresso machine under €600 that lasts"
What the ad has to do
Show a price and a reason. When the product sits over budget, we don't write a hint for that conversation: it becomes a separate ad group, or nothing.
Typical prompt
"A cheaper alternative to the big bed-in-a-box brand, with a real home trial"
What the ad has to do
Carry one verifiable difference (trial, returns, manufacturing), a single argument per ad, proven on the product page.
Typical prompt
"A gift for my dad who is taking up running, around €150"
What the ad has to do
Reassure on the choice (exchanges, wrapping, delivery time). The hint describes the person giving, not the one receiving.
// context hints
An e-commerce hint tells the shopper's story before the catalogue: the one who sleeps badly, the one torn between two technologies, the one who wants to try before paying. Here is how a direct-to-consumer mattress brand might lay out its account:
Generate your context hintsIllustrative excerpt. The real structure is built from your search terms, sales calls and support tickets.
AG · Use case
AG · Alternative and comparison
AG · Budget and timing
// scenarios
Three examples that tie a conversation to an ad, a page and a metric. They are not delivery promises.
// measurement
The native report goes no further than the click. Yet a store is steered at checkout. The OpenAI Pixel and the Conversions API therefore pass back items_added, checkout_started and order_created with the order value, and that is where the account gets read, against the Hiints Index.
Margin-based ROAS
Revenue divided by spend is not enough: we compute on margin after product cost, logistics and returns, ad group by ad group.
New-customer CAC
Acquisition cost of first-time buyers only, the number you set against Meta and Google over the same period.
Share of new customers
If the channel mostly recruits your existing customers, ROAS flatters and CAC lies. The split is built into the tracking itself.
checkout_started → order_created
The drop between the two says whether the fault lies with the ad (wrong expectation) or the page (fees, delivery time, payment).
Second order
For refills and subscriptions, LTV is read at the second purchase, tracked by first-purchase cohort.
CPC and CTR vs the Index's stores
OpenAI gives no reference point for CPC or CTR. Against the e-commerce accounts Hiints operates, your entry CPC can finally be read in context.
// honest qualification
Yes, clearly
Your average order value is above €100, and you know your margin per product and your return rate. You sell on your own site, with a page per product or per category, and the Pixel can pass back order value.
Yes, with one caveat
Your average order value is under €100 but refills or subscriptions carry the LTV: workable if we measure the second order, not just the first. Same caveat if part of your sales goes through marketplaces, since we will only see what closes on your site.
Not yet
You don't yet know your unit margin, your stock can't keep up with demand, or your monthly media falls short of €3,000. The channel would only speed up an existing problem, and it is better to say so than to let you spend.
Free setup, done in under 48h. Then 15% of ad spend under €5,000 of monthly media, 10% above. See pricing →
// questions
No. Raw ROAS deducts neither product cost, nor logistics, nor returns, and it flatters a channel that mostly recruits your existing customers. So we compute on margin per ad group and on the CAC of new customers only. The e-commerce sector page describes the conversations by product category; this one covers the economics of the account, and the two are meant to be read together.
Because entry CPCs are observed between $3 and $5, and a return costs money. Under that threshold, margin per order struggles to absorb the click; above it, or with proven reorders, the maths holds.
Because the conversation was about a precise product or use. The landing page carries it on: product page or category page, with price, stock and delivery time visible. The homepage, by contrast, starts from scratch.
Campaigns fed by a product feed form a distinct format. We test it when the catalogue lends itself to it (many references, price and stock up to date), alongside hand-written ad groups and never in their place.
On new-customer CAC and margin per order, using the same period for every channel. A flattering CTR says nothing about what an order brings in.
It all depends on what you earn on an order and how many orders it takes to read a CAC. Setup is free, the fee follows your media (15% under €5,000 a month, 10% above), and under €3,000 of monthly media we advise against testing.
// keep reading
How Much Does ChatGPT Ads Cost in 2026: CPC, CPM and Minimum Budget
Observed CPC of $3 to $5, CPM of $25 to $60, $25 per day minimum per campaign, no overall minimum since May 2026. What sets each cost line.
Read →ComparisonChatGPT Ads vs Google Ads: eight criteria to read them together
ChatGPT Ads vs Google Ads on eight criteria: targeting, signal, timing, format, auction, landing page, measurement, risk. Where each one wins, and why.
Read →TrackingMeasuring ChatGPT Ads Conversions: Pixel, API, UTM and CRM
The native report has six columns and no conversions. OpenAI Pixel, Conversions API, a UTM convention, events by business model and CRM reconciliation.
Read →// shall we talk?
Thirty minutes, live. We open your category in the Hiints Index, set your margin per order next to the entry CPC, and tell you whether the equation holds.
Book a call30 minutes · no commitment · zero pitch
