Agentic AEO is live on the Shopify App Store

Answer-engine optimization

A list has room for you. An answer might not.

An answer names two or three stores. AEO — also called GEO — is the work of being one of them, and of being described correctly.

Run a free auditHow the score works

Why ranking is not enough

A store can rank well and still not be named.

01

Readiness moves first

A ranked list has room for you at six. An answer has three names in it, and the shopper is never shown who came fourth. Publish a fix, re-scan, and the catalog half of the score moves that day.

02

Visibility moves later, or not at all

Whether an engine’s answer changes is a separate question, measured separately, on its schedule rather than yours.

03

Neither is a promise of placement

Burnish reports what the engines said. It does not negotiate with them.

Mechanism

An assistant can only say what it can read and check.

None of what follows requires knowing how any particular model was built. It is what is observable from outside it: ask the assistants the questions your buyers ask, keep the answers, and the shape of the thing is legible. One thing this page will not do is tell you what share of shoppers now buy this way — Burnish does not hold that number, and most of the ones in circulation are an estimate wearing a citation. What Burnish does hold is what five engines answered about a specific store on a specific day, which is the evidence that survives contact with a sceptic.

One

The question arrives in sentences, not keywords

Nobody types “merino base layer” into an assistant. They type “a warm base layer for a cold office that won’t itch, under $80, here this week”. That is four constraints and a deadline. The assistant has to satisfy them, not match them.

The request 4 constraints + 1 deadline

Keyword merino base layer 3 words

a warm base layer for a cold office that won’t itch, under $80, here this week

  • warmth
  • setting
  • comfort
  • price
  • deadline

Satisfied, not matched.

Two

It assembles an answer; it does not publish a ranking

What comes back is a short piece of prose with a handful of stores named in it. There is no page two, no position six, nothing below the fold to scroll to. You are in the answer or you are not in it.

Search returns

  1. 01
  2. 02
  3. 03
  4. 04
  5. 05
  6. 06
  7. 07
  8. 08
  9. 09
  10. 10

Sixth place still gets read.

An assistant returns

For wide feet and a cold commute, three worth looking at: Threshold Footwear, Cadence Athletic and Meridian Supply.

Each carries a wide fitting in a warm merino.

There is no sixth place.

Three

It repeats only what it can support

The details an assistant states about you came from somewhere it could read. Where your own pages are thin, or disagree with each other, there is less for it to state. Burnish’s twelve-question reputation battery measures exactly this: it asks each engine to say something specific about your store — legitimacy, returns, shipping, warranty, provenance — and records whether it can.

Representative demo · Threshold Footwear
Representative demo · Threshold Footwear

Four

The answer is not stable, and that is the system working

Ask twice, get two phrasings. Ask two engines, get two shortlists. This is a property of the instrument, not a fault in it, which is why Burnish samples on a fixed rotation, scores over a rolling 21-day window, and will not draw a confidence band on fewer than ten samples.

Representative demo · Threshold Footwear
Representative demo · Threshold Footwear

The distinction

Two systems, one catalogue. They ask different questions of the same eight fields.

Search engine

Indexes pages · ranks them

What comes back
Ten links, ranked. Sixth place still gets read.
Who is reading you
A crawler building an index of pages.
How you check it
Rank tracking.
Does Burnish measure it
No. No rank tracking, no Search Console, no claim about your positions.

Answer engine

Reads pages · names a few

What comes back
One answer naming a few stores. There is no sixth place.
Who is reading you
An assistant assembling an answer out of what it can read and verify.
How you check it
Asking the assistants and keeping the answers — dated, per engine.
Does Burnish measure it
Yes. Five engines on a fixed rotation, scored over a rolling 21 days.

Burnish scores this side

What both read

Titles, meta descriptions, structured data, alt text and FAQs — the eight fields Burnish rewrites. Burnish scores only the answer side, and it is not an SEO tool. But the fields it changes are the fields a crawler reads too, so the work lands in one place and is read from two.

What the answer draws on

Eight fields a store actually controls.

They are not a theory of how the models work — nobody outside the labs has that. They are the eight places your facts live, in the form a machine reads them. The weights, the ceilings and the arithmetic are on the methodology page.

Representative demo · Threshold Footwear
Representative demo · Threshold Footwear

Description quality

The paragraph an engine paraphrases. Specifics — material, dimensions, fit, what it is for — give it something to say. Adjectives give it nothing to repeat.

FAQ coverage

Buyer questions answered on the page, in the buyer’s words. It is the part of your store shaped most like the question actually being asked.

Structured data

schema.org markup states price, availability, brand and reviews as data rather than as layout. Read, not inferred from where it sits on the page.

Title optimization

A title that names the thing, the variant and the brand can be checked against a constrained request. “Item 4 — Final” cannot.

Alt-text coverage

What the picture shows, in words. Without it, the part of your catalog that carries the most information is the part a reader can’t read.

Meta description

The summary that travels with the link when something else quotes your page — which is most of how your page gets seen second-hand.

Meta title

Your page’s name outside your own site. Left on the Shopify default, it is usually the product title twice.

Manifesto

One set of facts about the store — what it sells, who it is for, what it stands behind — so the same answer is available wherever it is read from.

The five engines.

Every engine Burnish samples, and which plans reach it.

ChatGPT All plans Perplexity All plans Gemini All plans Claude Authority · Omnipresence Grok Authority · Omnipresence

Start here

Ask them about your store.

The free audit runs the same instrument the product runs, and shows the answers it got back — per engine, with the date attached.

Run a free audit See the engines

Agentic AEO runs on Burnish.

Free AI auditFree AI audit