Advanced Metric · Prompt Visibility Testing™

Measure what AI actually recommends.

Q: Which AI systems should I test with?

Test with at least ChatGPT, Perplexity, and Google AI Mode, since the same prompt can produce a different recommendation set on each one.

Websites tell you what an AI system can understand. Prompt testing tells you what it actually does when a buyer asks for a recommendation.

Category Advanced MetricMeasures Actual recommendation behaviorFeeds Recommendation Share

The core idea

Don't infer AI readiness. Test the recommendation.

YOUR STORE

↓

AI COMMERCE SIGNALS

↓

BUYER INTENT

↓

AI RESPONSE

↓

APPEARS

POSITION

COMPETITORS

↓

RECOMMENDATION SHARE™

Don't measure AI readiness by looking at your website. Measure it by testing the recommendation.

What a test looks like

Real prompt. Real recommendation trace.

This is a single row from a Prompt Visibility Test. Every prompt submitted produces a measurable, structured record, not a screenshot.

Prompt

"best organic skincare for sensitive skin under $60"

Brand appeared✓

Recommendation Position#1

Competitors ahead0

PlatformChatGPT

Prompt typeBuyer

Recommendation confidenceHigh

Aggregate result

Recommendation Share · 9 / 50 prompts

What we measure

Every AI recommendation leaves a measurable trace

Recommendation Share™

How often your brand is recommended.

Recommendation Position™

Where your brand appears when recommended.

#1#2#3mentioned

Recommendation Density™

How broadly your brand appears across relevant buyer intents.

Sensitive skin

Organic skincare

Anti-aging

Budget skincare

Recommendation Velocity™

How your Recommendation Share changes over time.

May

Jun

Jul

Aug

Read the full definitions: Recommendation Share, Recommendation Position, Recommendation Density, Recommendation Velocity.

Building your prompt set

We don't test random prompts

A useful visibility test must represent the real ways buyers ask AI for help. Each prompt type measures something different.

TYPE 01

Buyer Prompts

"Best protein powder for women runners"

Measures: Purchase visibility

TYPE 02

Comparison Prompts

"Best alternatives to Gymshark for women"

Measures: Competitive visibility

TYPE 03

Problem Prompts

"Skincare for acne-prone sensitive skin"

Measures: Intent alignment

TYPE 04

Discovery Prompts

"Eco-friendly activewear brands like Patagonia"

Measures: Category association

TYPE 05

AI Shopping Prompts

"Recommend a complete skincare routine under $100"

Measures: Delegated purchasing

Where to run tests

Multi-platform testing

We don't claim to know the exact internal mechanics of any AI system. What we observe, consistently, is that the same prompt produces different results on different platforms.

ChatGPT

Different retrieval and recommendation behavior. Results can lag behind recent changes to your store.

Perplexity

Strong retrieval and citation behavior. Tends to reflect current page content more directly.

Google AI

Search and generative answer behavior blended together. Structured data plays a visible role.

The same prompt can produce a different recommendation set on every AI system.

Why methodology matters

One prompt is not a measurement

AI responses are probabilistic. A single answer can be misleading. Reliable Prompt Visibility Testing requires a controlled prompt set, repeated observations, consistent scoring, and enough observations to identify patterns.

1 prompt

interesting result

50 prompts

measurement

100+ prompts

stronger dataset

Repeated monthly

trend

The methodology

The Atom Foundry Prompt Testing Protocol™

Define the merchant and category.

Define buyer intents.

Generate a controlled prompt set.

Run prompts across AI systems.

Capture every recommendation.

Normalize brand names.

Calculate Recommendation Share and Position.

Compare competitors.

Repeat over time.

The output is not a screenshot. It is a dataset.

Reading your results

What the data tells you

High visibility + low position

AI knows you, but prefers competitors.

Likely problem: Authority / trust / product relevance

Low visibility + strong site

Your store is technically ready, but AI rarely selects it.

Likely problem: External entity / recommendation evidence

High branded visibility + low buyer visibility

AI knows your name but doesn't associate you with enough buying intents.

Likely problem: Semantic positioning

Strong ChatGPT + weak retrieval platforms

Your brand has recognition, but current web evidence may be weaker.

Likely problem: Discoverability / citations / content

The bigger picture

From prompt testing to Recommendation Intelligence

PROMPTS

↓

RESPONSES

↓

RECOMMENDATIONS

↓

BRANDS

↓

POSITIONS

↓

INTENTS

↓

COMPETITORS

↓

RECOMMENDATION GRAPH

Every test adds another observation to the AI Commerce Graph™. Not a screenshot. Not a report. A dataset.

Where this fits

The product hierarchy

AI Commerce Score

= Are you structurally ready?

↓

Prompt Visibility Testing

= Does AI actually recommend you?

↓

Recommendation Share

= How much of the recommendation market do you own?

↓

Recommendation Intelligence

= Why does AI recommend you or your competitors?

↓

AI Commerce Graph

= The dataset of how AI routes commerce.

FAQ

Prompt Visibility Testing questions

What is Prompt Visibility Testing?

Prompt Visibility Testing is the process of submitting real buyer prompts to AI shopping systems and measuring whether a brand appears in the responses, at what position, and with what recommendation confidence. It is the most direct measurement of AI Commerce visibility because it tests actual recommendation behavior rather than inferring it from structural signals.

How many prompts do I need to test?

A minimum viable test is 50 prompts covering all five prompt types: buyer prompts, comparison prompts, problem prompts, discovery prompts, and AI shopping prompts. A comprehensive test uses 100 or more prompts. The more prompts you test, the more reliable your Recommendation Share calculation.

Which AI systems should I test with?

Test with at least ChatGPT, Perplexity, and Google AI Mode. These systems behave differently enough that the same prompt can produce a different recommendation set on each one. Testing all three gives a more complete picture of AI Commerce visibility than testing just one.

How often should I run Prompt Visibility Tests?

Monthly is the right cadence for most stores. Running the same prompt set every month gives you Recommendation Velocity, the rate of change in your Recommendation Share over time. Positive velocity confirms your AI Commerce improvements are working. Negative velocity means a competitor is gaining ground.

What is Recommendation Density?

Recommendation Density measures how broadly your brand appears across relevant buyer intents, not just how often overall. A brand can have high Recommendation Share for one intent and near-zero density everywhere else, which prompt testing is what actually reveals.

What is the Atom Foundry Prompt Testing Protocol?

It is the nine-step controlled methodology Atom Foundry uses to run every Prompt Visibility Test: defining the merchant and category, defining buyer intents, generating a controlled prompt set, running it across AI systems, capturing every recommendation, normalizing brand names, calculating Recommendation Share and Position, comparing competitors, and repeating over time.

Find out what AI recommends instead of you.

Run the free AI Commerce Score first. Upgrade to a Prompt Visibility Test to see exactly where your brand appears, where it disappears, and which competitors take your place.

Run a Prompt Visibility Test →Measure Recommendation Share →