The Answer Unit: One Paragraph That Tells AI What You Sell

A US appliance brand’s ecommerce lead reruns a shopper’s question one Tuesday: “Which ice maker fits under a 16-inch cabinet, makes enough ice for four, and needs no drain line?” Four brands come back; theirs is third, summed up as “also offers several ice makers.” The page opens with “Premium ice, made simple.” The width sits in an image caption.

An answer unit is one self-contained paragraph — who it serves, what can be checked, how it differs, what comes next — and without it a model fills the gap with a sentence that costs you the comparison.

The empty sentence is a symptom, not a mistake

A retrieval system does not summarize a page; it slices it, keeping spans that survive removal from their surroundings. A span survives when it carries a subject, a condition, and a value someone else can check.

“Premium ice, made simple” has none of the three. To a person it is praise; to a synthesis step it is a claim the model must vouch for, so it is dropped. With nothing quotable left, the answer falls back on the safest true line: this brand also sells this category.

Block boundaries decide the rest. Extraction works on paragraphs, so a block arguing three things gets cut, and the first block is read as the page’s self-definition.

Four slots, in the order a model resolves them

The unit is not a description or a slogan. It has four jobs, in extraction order.

Slot	What belongs there	Passes when	Failure
1 Role	Who buys it, in what setting	Category obvious without the headline	“Designed for modern living”
2 Proof	Measured values, test conditions, warranty terms	A stranger could repeat it	“Premium,” “lab-tested”
3 Contrast	Who it beats, on what dimension, who should not buy	Answers one exclusion question	“Better and more durable”
4 Next step	Price, shipping, stock, returns	Matches the other numbers	A discount code first

Move slot 4 first and the unit becomes a coupon with an expiry date. Drop slot 2 and the contrast collapses into an adjective; drop slot 3 and you are one of several interchangeable listings.

Example: one ice maker, two versions

Example — brand, specs, and prices are invented; no real product is described.

Before: “Brand A’s countertop ice maker uses premium components, smart sensors, and a sleek design for modern kitchens.” From that block alone, an assistant can say the brand sells ice makers and no more.

After, as four paragraphs:

“Brand A’s 12-inch ice maker serves four to six people in a rental kitchen with a 15-inch cabinet gap and no drain line.”

“In our test kitchen at 70°F it made 26 lb of ice in 24 hours, ran at 44 dB from one meter, and weighs 31 lb empty; the basket holds 1.8 lb.”

“At $279 it costs about $60 more than the loudest 26 lb unit we tested at 56 dB. In a bedroom that premium is the reason to buy; in a garage, the reason not to.”

“It ships from a US warehouse in two to four days with a two-year warranty; returns are free for 30 days.”

“Premium components” became a width, a cabinet gap, a temperature, a decibel reading, a weight. Slot 3 names a real disadvantage and the buyer who should walk — the sentence assistants reach for when asked about downsides.

Five steps from one paragraph to a template

Step	Action	Owner	Passes when
1 Choose pages	20 questions from 30 days of chat, site search, and ads, mapped to URLs	Content, growth	Every question has a page
2 Fix the filters	The four or five fields buyers screen on: width, capacity, noise, price, warranty	Content, category	One value per field
3 Write four sentences	Role, proof, contrast, next step; one per paragraph, no dangling “it”	Content	Hidden page, all four still stand
4 Test in isolation	Paste it alone into two assistants; ask what it is, why believe it, who it is not for	Content	All three answered, category intact
5 Template it	The slots go into PDP, category, comparison, and article templates	Content, front end	New pages ship with them

Step 4 is the only falsifiable step. Drift usually means a boundary or a dangling reference: sentences starting “It” lose their anchor without the heading. A screenshot is an anecdote — sample, as an AI visibility baseline would.

Reuse the skeleton, re-source the evidence

Four slots travel; slots 2 and 3 do not. A category page’s proof is range, its contrast is “how to choose within this category,” never “why we beat a rival”; a comparison page needs identical units; an article’s slot 4 is a link, not a price.

Localization is not translation: the same paragraph in German, with inches, 120 V, and a 30-day US return window, describes a different object. Re-source units, compliance and warranty wording, and the local word for the setting — a Berlin student flat is not an Ohio rental kitchen — or the versions will contradict each other.

Where this breaks

First, the paragraph is not reachable: copy inside a JavaScript-only module or a hero image is not a paragraph. Fix access first — see the AI crawler map.

Second, the brand is unresolved. Three spellings and no off-site confirmation mean four precise sentences only sharpen the confusion; check whether your brand entity exists first.

Third, the category resists compression: made-to-measure furniture, footwear needing a fitting, anything whose value appears weeks later. The unit becomes supporting material.

Fourth, revenue arrives inside a marketplace, where AI reads listings, reviews, and fulfillment. Fifth, the slots become a form for an agency: every cell has words, none is citable.

This piece opens the April line; settle the entry layer with the four AI search entry points, and browse the rest on the content and semantics hub.

This month, pick one action: take your best-selling SKU, run its real buyer question through two assistants, and copy the sentence each uses to describe you. If it reads like “also offers,” rewrite it in four sentences, publish them as separate paragraphs high on the page, and ask again in a week.