"Save 15%" Is on the Page. The Cancellation Policy Isn't.

In this article: the part of Interpretability Engine Optimization (IEO) that handles what you never said. Its job is to tell the difference between a "no" and a silence, keep what you did say intact, and make sure a machine treats the silence as a question for you, not as an answer for the shopper.

Here is what happens without it.

A coffee roaster sells a monthly box: Subscribe and save 15%. A shopper asks an AI assistant: "Can I cancel after the first box?"

The page explains the discount and the monthly delivery. It says nothing about a minimum term and nothing about how or when to cancel. The assistant answers anyway: "Yes, you can cancel anytime."

The roaster's actual policy, which lives in a help-desk article nobody linked, is three deliveries minimum, then cancel from the account page before the next charge. The shopper subscribes on the assistant's word, tries to cancel after box one, and gets a second charge. Now she's writing a one-star review about a policy you never published and a promise you never made. The assistant turned a condition that wasn't there into a selling point.

Why silence gets filled

Your page said two things clearly: 15% off, monthly. On the third thing, cancellation, it said nothing. To a machine working through a shopper's question, the two stated facts and the one missing fact don't look different. All three are "what we know about this subscription", and the gap gets handled the way gaps often are: with what's typical. Subscriptions are usually cancellable anytime, so this one must be. That's the machine's reasoning, and it isn't a fact about your store.

Assistants vary in what they do with a gap. Some guess, some ask, some abstain. OpenAI tells shoppers that its own shopping research may still get price or availability wrong and that the merchant's site is the place to confirm details. A subscription's cancellation terms sit further from the page than a price does, and they're just as easy to fill with an assumption.

A missing purchase condition can become a promise you never made.

The technical name: absence

In IEO this has a name: absence. The information does not exist in the source. No stated term, no stated condition, no provenance. The Methodology is strict about the remedy: an honest empty, with guidance on what it would take to resolve. Never inference.

Two things follow from that, and they're the whole lesson.

First, keep what's known. The discount and the frequency are stated. They stay, exactly as stated. Absence never means wiping the record; it means marking the one piece that isn't there.

Second, the gap goes back to the merchant, not to the shopper. The correct machine-readable value for cancellation on this subscription isn't "anytime" and isn't "never". It's unresolved: not stated, with a question attached: what's the minimum term, and how does a customer cancel? That question has exactly one person who can answer it, and it isn't the assistant.

Three moves on the coffee box:

1. Separate the stated from the missing. Stated: 15% off, monthly. Missing: minimum term, cancellation method, cancellation deadline.
2. Mark the missing as missing, with the question written out.
3. Get the answer from the owner and publish it, on the page and as data, so the next shopper's question has a source.

Where it lands: the merchant fills the gap

Here is the subscription, before and after. Notice who does the filling.

Nothing in the third column was invented, and nothing in the fourth column came from the machine. The merchant answered two questions, the page and the data got the answers, and the assistant can now say: "After three deliveries. Cancel before the next charge from your account page." Less friendly than "anytime", and true.

What to do this week

1. Open your subscription or promotion page and read it as a shopper's questions. Minimum term? How to cancel? When does the next charge happen? Can I pause? Can I skip? Does the discount apply to the first box only?
2. For each question, mark it: stated, or not stated. Be strict. "Flexible" isn't an answer. "Cancel anytime" is an answer only if it's actually your policy.
3. Answer the "not stated" ones with the policy you actually enforce, with whoever owns it.
4. Publish each answer in both places: on the page where a person reads it, and as data where a machine does.
5. Do the same for every promotion. Promo codes, first-order discounts, bundles. Each one has conditions, and each unstated condition is a question a machine may get.

Practical example

A sneaker store has a pair marked down from $120 to $90. The site banner says "10% off your first order."

A new shopper asks an assistant: "Can I use the 10% on these sneakers?"

Before you read on, decide what the machine can state and what it can't.

It can state: sale price $90. First-order discount: 10%. Both are on the site.

It can't state: whether the two combine. Nothing on the site says whether the 10% applies to sale items. That's the absent condition, and the honest output is: Combinable with sale price: not stated. The machine should not compute $81 and present it as the price. It should say the discount exists and that whether it applies to this pair isn't stated.

The merchant answers: "First-order discount valid on full-price items only." Now the data says so, the banner gets a footnote, and the assistant's answer is "No, the 10% applies to full-price items; these are already reduced to $90." Shorter sale, no argument at checkout.

That's the absence failure mode from section 5.3 of the IEO Methodology: an honest empty, with guidance on what it would take to resolve, and never inference. Keep what's stated, mark what isn't, and hand the question to the only person who can answer it.

Check what's explicit, what's missing and what contradicts itself on one of your products → Run the free audit

To go deeper → The IEO Methodology, section 5.3 (doi.org/10.5281/zenodo.22416784) · nsolvia.com/ieo

Sources: OpenAI, "Introducing shopping research in ChatGPT," Nov 2025 (shopping research may make mistakes on price or availability; confirm on the merchant's site).

← Your Brand Has Three Names Online. A Machine Sees Three Companies. · ⌂ IEO · Next: The Method Behind Every Before-and-After You Just Saw →

Series: Interpretability Engine Optimization · Knowledge Domain: Interpretability (IEO)

Interpretability Engine Optimization (IEO)
The UNDERSTAND Layer
nsolvia.com/ieo