Price Tiers and Bundles: How AI Understands Your Positioning

Early June, a growth lead at a US DTC pet-tech brand runs a buyer’s question through ChatGPT: “Best automatic litter box for one cat in a small apartment, under $400?” Three brands come back, hers among them, one line attached: “entry model around $199.” The tier she needs quoted is $399. Support has fielded the same question all month: what does the extra $200 buy?

Price is a positioning signal before it is a number: a model files you under the price it can lift off the page, and an incomparable bundle line parks you in your cheapest tier — where you then stay.

A model reads one price, not a pricing plan

Extraction does not weigh three tiers and conclude “mid-market.” It takes the shortest, most repeated, most structured number it can quote: the floor, recurring across grids, variants, and ranges.

The entry SKU becomes the price identity. The familiar version: the hero tier hides behind “Learn more,” the grid shows the floor, the headline says “premium quality.” Nobody lied; the page offered two statements, and the model took the extractable one.

$399 alone says nothing. $399 against “complete kit, three years of filters, 600–900 sq ft” is a different object from $399 “unit only.” Without contents and fit boundaries, the model compares numbers, and you lose on price.

The third break is quieter. Tier pricing is built for people — a middle tier carrying volume, a top tier flattering it — and a model inherits none of it. Pro does not outrank Standard unless the page says so. Nothing enters the 3–5 name candidate set before comparison is possible.

Tier names mean something to buyers, nothing to a machine

Plus, Pro, and Max are trade shorthand. A model gets no signal from a name sounding expensive; it needs what each tier adds, drops, and serves. Tiers that differ only in name collapse into one answer: “several versions.”

Field	Answers	Fails as
What’s in the box	Unit count, consumables, accessories	“premium accessory bundle”
Fit boundary	Room size, household size, pets, usage rate	“works in any setting”
Unit cost	Cost per wash, per cycle, per year of filters	Total price, no running cost

“From $199” is a claim too. If that tier is out of stock, not sold separately, or clearance, the “from” pins you to the low band while the margin sits at $399. Keep it buyable, or lead with the hero tier. Name by buyer, not rank: “small-apartment” and “two-cat household” carry their conditions; “Advanced” carries none.

Turn positioning into fields a machine can compare

Positioning is four or five extractable fields; where missing, the model substitutes price.

Step	Action	Owner	Passes when
1 Anchor	One hero tier, one price, one “for whom” line above the fold	Category, content	Tier, price, and setting on the first screen
2 Split	Contents, fit boundary, unit cost per tier	Category, merchandising	Cover the names; the ranking still holds
3 Name	The condition goes in the name	Content	The name answers “for whom”
4 Publish	Offer price, currency, availability match master data; AggregateOffer carries lowPrice and highPrice	Engineering, data	Page, feed, and JSON-LD hold one value
5 Sample	15–20 budget prompts; log the price and tier quoted back	Growth	Repeat samples agree

Step 1 is the hard one: it asks you to drop the three-across grid. “Fits a 600–900 sq ft apartment with one cat” beats “best seller,” which needs a verifiable basis; invented percentages die at review.

Step 4 is usually half-done: lowPrice without highPrice tells the machine only your floor. Field detail is in Offer, price, and availability; the pass condition is three outputs holding one value.

Step 5 is the only falsifiable one. Prompts carry a budget and a setting — “under $400, one cat, apartment” — and get sampled like an AI visibility baseline; one screenshot is an anecdote.

Example: three tiers, two versions

Example — brand, product, specs, and prices invented.

Before: Starter, Advanced, and Pro at $299, $449, and $649; one line beneath: “Find the right fit.” Asked for a robot vacuum under $500 for a one-bedroom with a dog, an assistant answers “starts around $299” and stops. The price was read; the positioning was not.

After, one paragraph and a table:

“The One-bedroom runs $449 and covers 600–900 sq ft on hard floor with one shedding dog.”

Tier	Coverage	What’s in the box	Consumables per year
Studio, $299	Up to 500 sq ft, hard floor	Robot, charging dock, spare filter	About $35
One-bedroom, $449	600–900 sq ft, one pet	Robot, self-empty dock, two filters, edge brush	About $45
Two-floor, $649	900–1,400 sq ft, two levels	Robot, self-empty dock, second dock, three filters	About $60

The same question now returns a $400–$550 range and repeats “includes a self-empty dock.” Not a recommendation — evidence the comparison now has a basis.

Where this breaks

First, revenue inside marketplaces. The platform renders price and promotions, so a clean on-site band connects to nothing; fix the listing and feed.

Second, promo-driven categories. A price that moves weekly writes positioning as contradictory sentences; publish a range or “typical price,” not a one-day promotion.

Third, quote-based B2B and made-to-order. No public price; positioning lives in configuration, lead time, and case evidence, and a forced “from” price buys mismatched inquiries.

Fourth, white-label goods where price is the only difference. The tier story carries no information; publish unit cost, fulfillment, and returns.

Fifth, invented strike-through prices. US and EU enforcement treat a reference price that never applied as deceptive, and the model learns an anchor it cannot reconcile; one discovery discounts every later price.

The price band is a strategy problem: it decides which table you are compared in. Copy belongs to content, fields to engineering, and the three move together: this line sits under strategy and market; the hero paragraph works best as a four-slot answer unit.

This month, take the SKU that carries your margin and write its three tiers as four columns — what’s included, who it fits, annual consumable cost, price. Name the hero tier above the fold, align lowPrice and highPrice with master data, then re-ask 15 budget-bearing questions in two weeks.