Be the brand the assistant puts in the basket.

They are live and scaling quickly. Advertiser counts moved from 2,585 to 3,770 in one week last June and reached 7,378 distinct brands by mid-July. Formats, inventory and buying routes are still changing month to month, which is exactly why we run it as a capped, measured test rather than a rebuild of your media plan.

Enough to gather honest data over a quarter without destabilising the channels that already pay your bills, which for most stores is a small share of existing paid budget. The benchmark is your own economics rather than a platform figure: with a median conversion rate of 1.74% and a median order value of $312, we can model the traffic volume needed for a credible read before spending anything.

No. Prompts are how we research the category — what shoppers ask, which brands get named, what objections come up, what price bands are quoted — and that research shapes ad copy, product pages and often your paid social angles too. Selling a folder of prompts would be selling you the notebook rather than the work.

Shopping catches someone who already knows roughly what they want, and Meta interrupts someone who was not shopping at all. An assistant conversation sits in between: the shopper is actively deciding, and the answer arrives as a recommendation rather than a grid of options. That changes the copy and the product data far more than it changes the buying discipline, which stays identical.

Some overlap is inevitable and we would rather measure it than pretend otherwise. We isolate the channel in analytics, watch new customer rate rather than total orders, and keep an eye on whether blended margin actually improves as spend moves. If ChatGPT ads are only reshuffling orders you would have won anyway, the report will say so.

Almost everything. An assistant assembles its answer from whatever it can read about your products, so thin titles, missing attributes, out-of-date stock or reviews buried behind a script quietly disqualify you. Cleaning that up is unglamorous work that pays twice, because the same fixes lift shopping campaigns and organic product visibility whether or not the ads succeed.

Well enough to make budget decisions, and we are honest about the edges. Tagged placements and referral data attribute direct clicks cleanly. Some influence stays indirect, where a shopper reads a recommendation and later searches your brand, so we track branded search and direct sessions alongside the channel report rather than quietly claiming everything as ours.

Considered purchases where people ask for advice: anything with specifications, sizing questions, compatibility, ingredients or a genuine comparison to make. Impulse and pure fashion buys are harder, because the conversation adds less. We start with the categories that combine assistant demand with margin that survives returns, then let the data move the budget.

We plan for it rather than react. Budgets, copy and product priorities are agreed before the period starts, stock and delivery promises are checked so we never advertise something you cannot ship, and spend is throttled the moment inventory tightens. Peak is also when the maths changes fastest, so reporting moves to a weekly rhythm for the duration.

A fixed monthly fee scoped to the work rather than a percentage of media, so our incentive is efficiency instead of volume, and you pay the platforms directly. There is no long-term lock-in. If ChatGPT ads do not earn its place inside a quarter, we will recommend moving the money back and the research we produced stays with you.

Directional signals within a few weeks, a decision-grade read at about ninety days. That gap is deliberate: assistant-driven shoppers tend to research over several sessions, so an early last-click view understates the channel while an over-eager scale-up overstates it. We show the leading indicators weekly and hold the scale-or-stop call until the margin data can carry it.

OpenAI places sponsored results inside conversational answers rather than on a separate results page, so an ad appears in the context of what the user asked. There is no keyword match type to bid on in the Google sense; the platform decides relevance from the conversation and your product data, which is why catalogue quality matters more here than bid tinkering. Formats and controls are still changing, and we brief you each month on what OpenAI has shipped and what it means for your campaign.

Google gives you keyword-level control, mature reporting and predictable auction behaviour; ChatGPT ads give you presence at the moment a buyer is being advised, with far less control and much less history to lean on. Neither replaces the other. In practice the stores getting value run Google and paid social as the base, add a capped ChatGPT budget on their best categories, and judge all three on the same margin line.

Not with the audience granularity you are used to. Targeting today is closer to context than to segments: the conversation, the product category and your catalogue signals decide who sees a sponsored placement, and the free tier of users carries most of the current ad inventory. So we shape who we reach through product selection, offer and copy instead of layered audience rules, and we say clearly what the platform does not yet let anyone control.

Less than Meta or Google, which is exactly why our tracking goes in first. Platform reporting covers the basics of delivery and clicks; everything that matters commercially — orders, new customers, returns, margin — comes from your own analytics with each placement tagged. If ChatGPT reporting and your store disagree, we treat your store as the truth and show both numbers in the monthly read.

There is no evidence that paying makes an assistant recommend you in an unpaid answer, and we will not sell that idea. What genuinely helps both is the same content and product work: clear specifications, real reviews, consistent descriptions across the web, and pages that are easy to read. We run that alongside the ads because it lifts your visibility whichever way the platform evolves.

Pricing is still settling, and anyone quoting you a firm ecommerce CPC for this platform is guessing. What we can say from live accounts is that cost varies enormously by category and by how commercial the conversation is, and that early advertisers are generally paying less than mature search auctions for comparable intent. We model your break-even before launch from your own margin and order value, then hold ChatGPT ads to it like any other line.

Get your products readable. Fix titles and attributes, publish real reviews, make specifications and returns plain text, keep the feed accurate, and check that your Shopify content is not hidden behind scripts. That work costs no media budget, improves what assistants say about you today, and means the day you do sign off on a test the money is not paying to send buyers to a page that cannot close them.

Less time than most agencies ask for. One working session at the start with whoever owns merchandising and margin, then about thirty minutes a week: a quick view of the performance summary and a decision on product priorities. Ecommerce teams are stretched, and the reporting is written so a founder can read it between meetings rather than needing a dashboard tour.

Yes, and the numbers get more honest when one team holds them. Our paid social, Shopify and analytics specialists work from the same plan, so an order is attributed once instead of claimed by three vendors. Hiring us for this channel alone is equally fine — everything is documented and left in accounts you own.