The observability layer for AI shopper experience

AI shoppers are leaving your store empty-handed

agent-session / shopping-assistant / northwind-trail-2l

I need a rain shell for a wet hiking trip next weekend. Is the Northwind Trail 2L actually waterproof, and can I get a medium in time?

What the agent could read

  • Product name
  • Price
  • Availability
  • Waterproof rating
  • Fit and sizing
  • Variants in stock
  • Delivery estimate
  • Return window

What the shopper was told

Machine readability score0/100

0/8

buying questions answered correctly

$0

monthly leak on this one product

Illustrative walkthrough built from a sample audit, not a live customer store. Your report uses your own catalog, traffic and order values.

Assistants read your catalog, answer questions about your products, and quietly give up on your checkout. None of it reaches your analytics. Product Gale is the layer that sees every one of those interactions, prices what goes wrong, and proves the fix.

Your first AI Revenue Leak report in under a week. No code, no tag to install

Nothing in your stack can see this

Three systems should have caught it. Each one misses for a different reason, and none of them is a bug you can file.

GA4 and web analytics

What it sees

Sessions, referrers, conversions

Why it misses this

Assistant traffic arrives unattributed or as direct, and analytics has no concept of the answer a shopper was given before they decided not to arrive.

APM and infrastructure monitoring

What it sees

Requests, latency, errors

Why it misses this

A successful 200 response is still a failure if the agent could not parse your variant selector. Nothing errored. The journey just ended.

LLM observability tools

What it sees

The model calls your own systems make

Why it misses this

They instrument the calls going out of your systems. They have nothing to say about the assistants coming in, which is where the shoppers are.

No bounce. No abandoned cart. No error in the logs. A customer existed, and was routed to a competitor.

This is not a 2028 problem

AI shoppers already arrive at your store, already outperform search traffic when they buy, and already fail to read most of what you sell. The gap between those three facts is the money on the table.

+4,700%

Growth in AI referral traffic to US ecommerce, year over year

Adobe Analytics, July 2025

30%

More likely to convert than a shopper arriving from traditional search

Adobe Analytics, holiday 2025

66/100

Average score of a retail product page on machine readability

Adobe, 2026

-18%

Revenue per visit from AI traffic versus what it should be earning

Adobe, 2026

One layer, every AI interaction

AI touches your store in three places. All three fail quietly, all three belong on the same schema, and we are honest about which one we cover today.

Live today

AI shoppers coming in

ChatGPT, Perplexity, Gemini, Claude, shopping agents

They read your catalog, answer questions about your products, and give up somewhere in your buying flow. This is where the revenue is today, and it is the surface nobody instruments.

Next

The assistant you run

On-site chat, product finders, support bots

The assistant you put in front of your own shoppers quotes a price, a stock status and a returns policy on every conversation. Nobody is checking whether it gets them right.

Next

The models your systems call

Categorisation, descriptions, attribute extraction

Your own pipelines call models to write copy, sort products and fill attributes. When one starts drifting, it shows up in the catalog long before anyone traces it back.

Same event model, same dollar figure, same alerts, whichever surface the failure happened on.

A diagnosis you can act on

Most tools hand you a readability score and stop there. A score doesn't tell you which products to fix or what fixing them is worth.

01

Detect

We identify the AI assistants and agents already reaching your store, the pages they read, and the ones they give up on. No tag to install. We start from your public store and your logs if you have them.

02

Test

We put your catalog in front of the assistants your shoppers actually use, with the questions they actually ask, and record every wrong answer. Then we send an agent through your real buying flow and log where it stalls.

03

Price

Each failure gets a dollar figure built from your own numbers: traffic, average order value, conversion rate. You get a monthly leak total and a ranked list of what's costing the most.

04

Fix

We repair the pages behind the top failures so the assistants can answer correctly, then re-run the same tests to prove the answer changed. The loop repeats as your catalog and the models move.

Find the leak, then close it

Detect the AI traffic you can't see today, prove what it's costing you, and fix the pages responsible, in one loop that keeps running.

AI Traffic Detection

See which assistants and shopping agents hit your store (ChatGPT, Perplexity, Gemini, Claude and the rest), how often, and which pages they read before they leave.

Answer Accuracy Testing

We ask the assistants your shoppers use what they'd recommend, and record every time you're skipped, misquoted, listed at the wrong price, or beaten by a competitor.

Agent Journey Testing

We send a real AI shopper through your flow: search, variant, cart, checkout. Then we record the exact step it fails on, with the reason and a replay.

Revenue Leak Attribution

Every failure carries a number, not a grade. Sessions, intent, your AOV and conversion rate turn each broken answer into a monthly dollar figure.

Automatic Page Repair

Finding the leak is half the job. Product Gale rewrites what the AI couldn't read, whether that's missing specs, ambiguous variants, or unclear price and stock, then re-tests until the answer is right.

Continuous Monitoring

Answers drift as models update and your catalog changes. We re-run on a schedule and alert you when a new agent shows up or an answer regresses.

A layer, not another dashboard to check

Everything we record is available as data. Product Gale is meant to be a source your stack reads from, not one more place someone has to remember to look.

REST API and SDKs

Pull findings, leak figures and fix status into your own reporting. Python and JavaScript SDKs.

Webhooks

Scan complete, new finding, answer regression, fix published. Delivered as events, with history.

Alerts where you work

Slack, email or your own tooling, the moment an answer that used to be right stops being right.

Store connectors

Push corrected product content back to the storefront, reviewable before it publishes.

Server log ingestion

Optional, and it turns agent detection from inference into a per-agent record of who arrived.

Simple, usage-based pricing

You pay per product we put in front of AI shoppers. Start free, no hidden fees.

Free

Your first scan, on us, with $20 lifetime credit

$0/month
  • Own 1 workspace
  • Join unlimited workspaces
  • $20 lifetime AI credit
  • AI Revenue Leak report
Scan My Store
Most Popular

Growth

Pay-as-you-go for stores that keep shipping

Pay-as-you-go
  • Own multiple workspaces
  • Continuous monitoring & alerts
  • Pay only for the products you cover
  • Full revenue leak attribution
  • Priority support
  • Store & platform connectors
Start Free

Custom

For large catalogs and multi-store groups

Custom
  • Everything in Growth
  • SSO & SAML
  • Dedicated account manager
  • Custom contracts
  • SLA guarantee
Contact Sales

Questions we get every week

Short answers to what buyers and assistants ask most about AI shopper traffic.

What is Product Gale?

Product Gale is the observability layer for AI shopper experience. It detects which AI assistants and shopping agents interact with an ecommerce store, records what they were told about the products and where their journey ended, prices each failure in lost revenue, and repairs the pages responsible. A first scan needs only your domain, with no tag or code to install.

Why can't Google Analytics see AI shopper traffic?

Assistant traffic usually arrives unattributed or as direct, and analytics has no concept of the answer a shopper was given before deciding not to visit. Application monitoring does not catch it either, because a page can return a successful response and still fail an agent that could not parse the variant selector. The visit either never happens or completes without an error, so nothing in a standard stack records it.

What is an AI Revenue Leak?

AI Revenue Leak is the monthly dollar value of business lost to AI interactions that failed. Product Gale computes it per failure as AI sessions with that intent, multiplied by the probability the store would have won the query, the store's conversion rate, and its average order value. Every assumption is shown, ranges are modelled, and the conservative case is what gets reported.

What is Agent Completion Rate?

Agent Completion Rate is the share of AI agent journeys against a live storefront that reach checkout. An agent is driven through the real buying flow of search, product, variant, cart and checkout, and each journey is scored as completed or failed, with the failing step, the reason, and a replay recorded for every failure. It measures completion rather than whether a store was mentioned.

How is this different from an AI visibility or readability score?

A score out of 100 does not say which products are affected, whether anyone is asking about them, which to fix first, or what fixing them is worth. Product Gale reports failures in dollars per month with the assumptions visible, ranks the fix list by revenue at stake, repairs the pages behind the top failures, and re-runs the identical question afterwards to prove the answer changed.

Do I need to install anything to get started?

No. A first scan runs from your public storefront and needs only your domain. Connecting server logs or your store platform is optional and adds per-agent detail, and page repairs are reviewable before they publish and fully reversible.

Which AI assistants and agents does Product Gale cover?

The major consumer assistants and shopping agents that drive real ecommerce traffic, including ChatGPT, Perplexity, Gemini and Claude, plus the shopping agents built on them. Coverage is updated as new agents appear, and continuous monitoring alerts you when an unfamiliar agent starts visiting your store.

How do I know a fix actually worked?

The identical question is re-run against the same assistant after the repair is published, and the new answer is recorded next to the old one. The before and after are stored, so the claim is verifiable by you rather than asserted by us.

Find out what AI shoppers cost you last month

Give us your domain and we'll turn the layer on over your store. You get the failures, the step each one breaks on, and the monthly number attached. Nothing to install.

Ask AI to summarize