The Section Nobody Writes

Seven answers to the question “how do I sell this,” captured across Google’s AI Overview and both tiers of ChatGPT, all cover the same five topics. Zero of seven cover who buys the product, when they buy it, or what words they use when they go looking for it.

The question every merchant asks

There is one question a person actually asks before they start a product business, and they ask it in almost exactly these words: how do I sell this.

Until recently that question went to a person, someone who had done it before, over coffee or for a consulting fee. That is no longer where it goes. It goes into an assistant, and the assistant answers immediately, at length, and for the most part correctly.

So in August 2026 the question was asked the way a merchant would ask it, about a real product category, and the answers were kept. Five answers for the first category, captured across Google’s AI Overview and both the free and paid tiers of ChatGPT, reviewed on August 9 and 10. Two more answers, captured the same way, for a second category. No leading questions, no pushing back on the answers. The point was to record what a merchant gets on the first try.

What all seven answers cover

Five topics appear in every one of the seven answers. Not most of them. All seven.

Regulatory classification and compliance. Manufacturer or supplier sourcing. Labeling requirements. Selection of an e-commerce platform. Generic marketing tactics.

That is a consistent curriculum, and it deserves fair credit before any criticism, because the coverage is not padding. The regulatory material in particular is specific and checkable: what a label must carry, which claims change a product’s legal classification, which registrations a state expects. A merchant who follows that material avoids real trouble, and getting it in seconds instead of billable hours is not a small thing.

The paid tier goes further. It produces margin tables. It sequences registrations in the order a founder should file them. It lists the questions to ask a co-manufacturer before signing. It reads like consulting because it is consulting, delivered at a speed no consultant can match.

Where every one of them stops

Now the finding, and it is a count rather than an impression.

Who buys this product. When they buy it. What words they use when they go looking for it. Zero of seven answers cover any of it.

Not thin coverage. Not one hedged paragraph. Absent. Every answer takes the merchant carefully from an idea to a manufactured, compliant, labeled, listed product sitting on a platform, and not one of them asks whether anybody wants it, who that person is, or what they would type.

Every answer stops at the factory door.

The marketing section is where this is easiest to miss, because a marketing section is present in all seven and it looks like the missing material. It is not. What it contains is channels and mechanics: run paid social, build an email list, work with creators, post consistently. Every one of those is a container. Not one of them tells the merchant what to put inside it. An email list is not a position, and a merchant handed five channels and no sentence will write the sentence themselves, out of the vocabulary they arrived with, which is the vocabulary of the person who makes the product rather than the person who buys it.

The sharpest moment

The place where this stops being an omission and becomes something stranger is in the paid answer, which is the best of the seven.

It tells the merchant to pick a positioning. One positioning, chosen up front, because a product with a diffuse story does not sell. That advice is sound. Then it offers a list of candidate positionings to choose from.

There is nothing behind the list.

No count of anything. No record of what buyers in the category actually ask for. No indication of which candidate has a market already organized around it and which is a phrase somebody thought of. The options are plausible and well written, and they come from the same place a smart person’s first guess comes from: a general sense of how categories like this usually work.

Sit with what is being asked of the merchant there. Bet the business on one phrase. Choose it by intuition. Then spend the months that follow on sourcing, compliance, labeling, inventory, and advertising, all of it downstream of that phrase, all of it expensive, and all of it wasted if the phrase was wrong.

The missing section can be written from data

The section the answers skip is not unknowable. It is countable.

For any category, the open market already records which audiences appear around the products, which occasions and situations the products serve, which seasons carry them, and which words do the describing. Those are counts in a commerce map, not guesses, and they are exactly the material a positioning decision needs before the first compliance form gets filed.

That is the section the Reach Dog selling guides publish. Each guide opens one category from the buyer’s side: the jobs, the occasions, the audiences, the seasons, and the language, with every claim tied to a stored source. The guides exist because of the count in this study. The most sophisticated answer a merchant can get today hands them a factory and wishes them luck with the sentence. The sentence is the part that decides whether the factory was a good idea.

How this was measured

Seven answers across two product categories, captured from Google’s AI Overview and the free and paid tiers of ChatGPT in August 2026 and reviewed on August 9 and 10. Questions were phrased the way a merchant would phrase them, with no follow-up prompts and no corrections. An answer was scored as covering a topic if it addressed the topic at all; on who buys, when, and in what words, none of the seven did.

Seven answers in two categories is a capture, not a census. The pattern is consistent across all seven, and the study does not claim it is universal. Assistant answers change as models and retrieval change; the captures are dated, and the finding should be re-measured rather than assumed.