Shopify AI shopping agent attribution: how to see exactly what ChatGPT, Perplexity, and Google AI Mode are sending you

The attribution gap: why standard analytics misses AI traffic

When a shopper asks ChatGPT "find me a running shoe under $120 with wide sizing" and clicks through to your Shopify store, ChatGPT passes a referrer string — just like any other website link. But the string it passes (chatgpt.com) lands in your GA4 Referral channel alongside links from blog posts, affiliate sites, and coupon aggregators. There's no automatic separation.

The problem compounds across three dimensions:

• Multiple agents, different referrer strings. ChatGPT sends traffic from chatgpt.com. Perplexity sends it from perplexity.ai. Google AI Mode sends it from google.com — the same domain as organic search. Claude.ai sends it from claude.ai. Bing Copilot sends it from bing.com. Without a custom channel definition, these land in three different buckets: Referral, Organic Search, and nothing (for Bing, counted as Paid Search in some configurations).
• Crawler traffic muddies the signal. GPTBot, PerplexityBot, and Google-Extended index your products. Their requests appear in server logs with distinct user-agent strings — but they don't appear in GA4 at all, because browsers don't fire GA4 events. The crawl count and the customer count are two completely separate signals that need to be read from two completely separate sources.
• Conversion rate matters more than traffic volume. Adobe's 2025 holiday data shows AI-referred retail traffic converts +31% better than non-AI traffic. That premium means a small AI shopping segment can drive outsized revenue — but you can't act on that finding if the traffic is buried in a generic Referral report.

AI agent referrer signatures (complete reference table)

Each AI shopping agent sends a different combination of referrer URL, user-agent string (for crawlers), and UTM parameters. This table is the raw material for your GA4 channel group definition and your server-log grep commands.

Crawler traffic vs. customer referrals — the crucial distinction

This is the most common confusion in AI agent analytics. There are two completely separate signals:

1. Crawler requests — GPTBot, PerplexityBot, and Google-Extended visit your product pages to index them for their respective shopping catalogs. These appear in your server logs (or Caddy access logs, or Cloudflare traffic data) with bot user-agent strings. They never fire GA4 events. They tell you: "AI agents are able to read my catalog."
2. Customer referrals — A real shopper clicks a product link in a ChatGPT or Perplexity Shopping response and lands on your store. This person fires a GA4 session_start event with chatgpt.com as the referrer. It tells you: "AI agents are sending me buyers."

You need both signals, but they answer different questions:

GA4 custom channel group for AI shopping agents

GA4's default channel groups treat every inbound referrer the same way. To break out AI shopping traffic, you need a custom channel group. Here's the complete setup, step by step.

Step 1: Create the custom channel group

In GA4: Admin → Data display → Channel groups → Create new channel group.

Name it "AI Shopping Agents" or similar. The order of channel rules matters — GA4 applies them top to bottom and assigns the first match. Put your AI rules before the generic Referral rule so chatgpt.com doesn't get absorbed into Referral.

Step 2: Define the channel rules

Add one channel per agent. Each channel rule uses Source as the condition:

Channel: ChatGPT Shopping
  Condition: Source exactly matches "chatgpt.com"

Channel: Perplexity Shopping
  Condition: Source exactly matches "perplexity.ai"

Channel: Claude.ai
  Condition: Source exactly matches "claude.ai"

Channel: Bing Copilot
  Condition: Source exactly matches "copilot.microsoft.com"

Channel: You.com Shopping
  Condition: Source exactly matches "you.com"

Channel: Shopify Global Catalog
  Condition: Source exactly matches "shop.app"
  OR Source exactly matches "shopify.com" AND Medium exactly matches "referral"

Channel: Google AI Overview (srsltid)
  Condition: Source exactly matches "google.com"
  AND Page location contains "srsltid"

Step 3: Roll the channel group back 90 days

GA4 custom channel groups apply retroactively once created. Navigate to Exploration → Blank exploration → set Date range to last 90 days → add "Session channel group" as a dimension and select your custom group. This immediately shows you how much AI traffic you've had for the past three months, even though the channel group didn't exist then.

Step 4: Add revenue and conversion events

The channel group alone shows sessions. To see revenue, add the purchase event metric and the ecommerce_purchase conversion event to your Exploration. Segment by your custom channel group to get per-channel revenue, AOV, and conversion rate in one view.

/* Looker Studio / GA4 custom report formula */
AI Shopping Revenue =
  SUM(purchase revenue)
  WHERE session_channel_group IN (
    "ChatGPT Shopping",
    "Perplexity Shopping",
    "Claude.ai",
    "Bing Copilot",
    "You.com Shopping",
    "Shopify Global Catalog",
    "Google AI Overview (srsltid)"
  )

AI Shopping Conversion Rate =
  COUNT(DISTINCT sessions with purchase)
  / COUNT(DISTINCT all sessions)
  WHERE session_channel_group IN (above list)

Shopify Analytics limitations for AI traffic

Shopify's built-in analytics (the "Marketing" and "Sessions" reports in the Shopify admin) do not support custom channel definitions. They show a flat list of referrers — chatgpt.com appears as a referral source alongside every other link on the internet. There's no way to tag it as an AI shopping source within Shopify Analytics.

For AI shopping attribution, Shopify Analytics is useful for exactly one thing: checking total orders by referrer source, which gives you a quick sanity check on whether GA4's AI channel numbers are in the right ballpark. Navigate to Analytics → Marketing → Sessions by referrer and filter by date range to see raw session and order counts from chatgpt.com and perplexity.ai.

Server log analysis: parsing AI referral patterns

Server logs (Caddy, Nginx, or Cloudflare Workers logs) are the only source for crawler activity data. They capture every request — including bot crawls that never fire a GA4 event. Here are the grep and awk commands for the most common log formats.

Caddy access log (NDJSON format)

Caddy writes access logs as newline-delimited JSON. Use jq to parse AI agent user-agents:

# Count GPTBot requests by day (Caddy NDJSON logs)
cat /var/log/caddy/access.log | \
  jq -r 'select(.request.headers["User-Agent"] // "" | test("GPTBot|OAI-SearchBot")) | .ts | strftime("%Y-%m-%d")' | \
  sort | uniq -c | sort -rn

# Extract URLs indexed by PerplexityBot
cat /var/log/caddy/access.log | \
  jq -r 'select(.request.headers["User-Agent"] // "" | test("PerplexityBot")) | .request.uri' | \
  sort | uniq -c | sort -rn | head -50

# Full AI crawler summary: agent, URL, response code, date
cat /var/log/caddy/access.log | \
  jq -r 'select(.request.headers["User-Agent"] // "" | test("GPTBot|PerplexityBot|Google-Extended|ClaudeBot|bingbot|YouBot|Amazonbot")) |
    [
      (.request.headers["User-Agent"] | capture("(?P<bot>GPTBot|PerplexityBot|Google-Extended|ClaudeBot|bingbot|YouBot|Amazonbot)").bot // "unknown"),
      .request.uri,
      (.status | tostring),
      (.ts | strftime("%Y-%m-%d"))
    ] | @tsv' | \
  sort | uniq -c | sort -rn

Nginx combined log format

# Count AI crawler requests by user-agent (Nginx)
grep -E "GPTBot|PerplexityBot|Google-Extended|ClaudeBot|OAI-SearchBot|YouBot|Amazonbot" \
  /var/log/nginx/access.log | \
  awk '{print $12}' | sort | uniq -c | sort -rn

# List all product pages crawled by GPTBot in last 7 days
grep "GPTBot" /var/log/nginx/access.log | \
  awk '{print $7}' | \
  grep -E "^/products/" | \
  sort | uniq -c | sort -rn | head -100

Cloudflare Workers / Analytics Engine

If you're using Cloudflare in front of your Shopify store, AI crawler traffic is visible in Cloudflare Analytics under Security → Bots → Known bots. Filter for "Verified bots" and look for GPTBot, PerplexityBot, and ClaudeBot in the breakdown. Cloudflare also logs the URLs accessed, which is useful for confirming which product pages are being indexed most frequently.

Building your AI agent revenue dashboard (5 KPIs)

Once your GA4 custom channel group is set up and your server logs are parsed, you have the raw data for a weekly AI agent revenue dashboard. These five KPIs are the minimum viable reporting surface for a Shopify merchant who wants to know whether AI shopping is working.

KPI 1: AI Shopping Revenue (weekly)

The total revenue attributed to sessions from your custom AI Shopping Agents channel group. This is your north-star number. It should trend up week-over-week as you improve structured data. A flat or declining line after a structured data fix means the fix didn't work or the channel is too small to show the effect yet — check KPI 3.

How to pull it: GA4 → Explore → Free-form → Metric: Purchase revenue → Dimension: Session channel group → Filter: channel group IN (your AI channels) → Date comparison: this week vs. last week.

KPI 2: AI Shopping Conversion Rate

Sessions from AI channels that result in a purchase, divided by total AI channel sessions. Baseline this against your overall site conversion rate. A premium of +0.5pp or more indicates AI-referred shoppers have higher purchase intent — this is the signal that justifies investing more in structured data quality over other channels.

A conversion rate below your site average on AI channels is unusual and worth investigating — it can indicate a product-query mismatch (the AI agent is sending shoppers to the wrong product) or a landing page issue on the pages AI agents link to most.

KPI 3: Indexed Products (weekly crawler count)

The total number of distinct product URLs crawled by AI bots in the past 7 days, summed across GPTBot, PerplexityBot, Google-Extended, and ClaudeBot. Pull this from your server logs using the grep commands above. This is your leading indicator — it moves before revenue does, because you have to be indexed before you can be referred. A rising indexed-products count with flat revenue usually means you're winning the crawl battle but losing on structured data quality or product-query match.

KPI 4: Top AI-Referred Product Pages

The product pages that receive the most sessions from AI channels, ranked by revenue (not just sessions). Pull this in GA4: Explore → Free-form → Dimension: Landing page + Session channel group → Metric: Sessions + Purchase revenue → Filter by AI channel group → Sort by revenue.

This table answers a critical question: are AI agents sending traffic to your highest-margin products, or to catalog outliers? Use it to prioritize structured data improvements. If a $15 product is getting 80% of your AI referral traffic while your $150 products get none, the problem is usually that the high-margin products have weaker GTINs, ratings, or descriptions — not that AI agents inherently prefer cheap products.

KPI 5: AI Shopping Revenue Share of Total Revenue

AI channel revenue divided by total revenue, expressed as a percentage. This is the business-level justification number. As the AI channel share crosses 5%, 10%, and 15%, it warrants proportionally more catalog hygiene investment. Adobe projects AI-referred retail traffic will represent 15–20% of total ecommerce referrals by end of 2026 for stores that have optimized their catalog. Stores that haven't will converge toward 1–2% because AI agents deprioritize low-signal products.

Closing the loop: connecting attribution data back to structured data fixes

Attribution is only useful if it closes a feedback loop. The loop looks like this:

The payoff of this loop is a compound effect: each structured data fix raises your CatalogScan score, which raises your AI agent ranking, which raises your AI referral share, which gives you more data to prioritize the next fix. Stores that run this loop monthly outperform stores that make one batch of structured data changes and stop — because the AI shopping agents themselves are evolving, and the ranking factors shift over time.

Verification checklist

• GA4 custom channel group created with rules for chatgpt.com, perplexity.ai, claude.ai, copilot.microsoft.com, you.com, shop.app, and the srsltid Google AI Overview rule
• GA4 custom channel rules are ordered above the default Referral channel (GA4 applies first match)
• GA4 Exploration built showing revenue, sessions, and CVR by AI channel group over last 90 days
• Shopify Analytics referrer report checked for presence of AI referrers — any of chatgpt.com, perplexity.ai, shop.app in top-20 referrers confirms meaningful volume
• Server log grep commands configured for GPTBot, PerplexityBot, Google-Extended, ClaudeBot, and Amazonbot
• Weekly indexed-products count baseline established (last 4 weeks average from server logs)
• Top AI-referred product pages report built in GA4 Explore — product URL × AI channel group × revenue
• AI Revenue Share baseline established for monthly trend tracking
• Cloudflare bot settings verified: GPTBot, PerplexityBot, ClaudeBot allowed in Bot Fight Mode exceptions — see Cloudflare guide
• Structured data fix cycle scheduled: CatalogScan re-run monthly, server log crawl audit weekly, GA4 revenue dashboard weekly