ChatGPT ads for auto parts sellers, built on fitment truth.

Yes, on a cost-per-click basis through OpenAI's ads manager, with conversion bidding and the web pixel as the measurement layer, and product carousels available for catalogue ranges. Digiday reports the ads business now runs in 31 markets with cost-per-click accounting for most of the spend. What does not exist yet is lookalike or interest targeting, so relevance is earned through product data, campaign context and conversion signals. We keep the platform overview current as new controls ship.

Fix the ranges you intend to advertise, then launch on those — waiting for a perfect catalogue means never starting. The cost of skipping it is measurable: 19.4% of auto parts orders come back, and 86% of automotive returns trace to inaccurate fitment information, so ads pointed at ambiguous products buy revenue you refund later plus the shipping both ways. We audit coverage by product family, rank the families by margin and data quality, and open campaigns in that order.

Shopping matches a query to a product; this matches a conversation to a recommendation. The user has usually told the assistant the year, make, model, engine and symptom before any product is mentioned, which is far more context than a search term carries, and it means the quality of your product attributes decides whether you are considered. The practical differences are that you cannot buy audiences here, creative matters less than data, and the same catalogue work lifts both channels. We run them together and compare them on contribution rather than on clicks.

Enough to read margin on two or three product families over eight to twelve weeks, which for most retailers is a modest share of an existing shopping budget rather than new money. The right size depends on average order value and return rate far more than on the platform: a business selling brake kits at healthy margin can read results quickly, while a low-margin commodity range needs volume before anything is conclusive. We agree the margin threshold that would make it a keeper before launch, and stop early when the answer is obvious.

By refusing to write a compatibility claim the product data cannot support. Ads name the vehicle range the feed can prove, landing pages confirm fitment before price, and anything ambiguous is excluded from the campaign rather than hedged with 'fits most models'. We also keep an approved language list so nobody improvises a claim in a rush, and we watch return reasons by campaign so a data problem surfaces in weeks rather than at the quarterly review.

Diagnosis and shopping questions, mostly in the same conversation. People describe a noise, a warning light or a failed inspection, ask what part is likely at fault, then ask what it should cost and which brand is worth buying for their vehicle. Trade buyers ask narrower things: cross-references from an OE number, availability, and whether an aftermarket equivalent is acceptable. The best campaigns come from the questions your own counter staff answer every day — they already know the prompts, they have simply never treated them as advertising ideas.

Indirectly, and the overlap is the reason we do the catalogue work first. Being cited in an assistant's answer depends on clear, machine-readable product information, honest specifications and content that answers the question a person actually asked — the same assets that make paid placements perform. Paid and organic are still separate systems and we never pretend that buying ads purchases citations. If AI visibility is the priority, our GEO and AEO team handles it as its own programme.

Revenue net of returns, contribution margin by product family, new versus repeat customers, and cost per order compared with shopping and search. Everything is reconciled against your own order data rather than the platform's self-reported figures, because the two rarely agree and the difference matters when returns run high in this category. The monthly review is one page in plain language, with a change log underneath and a clear recommendation when a range should stop being advertised.

Yes, and the numbers get more honest when we do. Most retailers we meet have shopping with one supplier, marketplaces with another and an agency for content, each reporting its own best month. Our paid search, ecommerce and analytics specialists work from one plan so an order is attributed once. It is equally fine to hire us for this channel alone — we document the setup and leave the accounts open, because a campaign only one supplier can operate is not an asset.

A fixed monthly fee, quoted separately from media spend and scoped to the product families you want advertised. After the audit you get a plan tied to a margin target and a written view of which ranges we would not advertise yet, rather than a bundle sold on day one. Where the useful scope is a feed clean-up and a measurement build instead of a retainer, we scope it that way and say so on the first call. You own the ad account, the pixel and the product feed throughout.