← Back to Blog2026-04-18

AI Shopping Attribution: How to Track Sales From ChatGPT, Perplexity, and Google AI in 2026

AI shopping traffic is growing fast. Tracking it is not. If you have looked at your analytics recently and seen a spike in "direct" or "(none) / (none)" sessions you cannot explain, you are not alone — that traffic is increasingly coming from ChatGPT, Perplexity, and Google AI Overviews, all of which obscure their referrer in ways that frustrate traditional attribution.

This guide covers what is trackable today, what is not, and how to build a defensible AI-attribution model so you can measure the channel's real impact on revenue.

The Attribution Problem

Traditional ecommerce attribution depends on the referrer header — the URL the user came from. Google search shows google.com as referrer. Facebook shows facebook.com. UTM parameters add campaign details on top.

AI engines break this model in several ways:

ChatGPT cites sources but routes clicks through its own UI, so the referrer is often chat.openai.com or, depending on configuration, a redirect that strips the source attribution.
Perplexity surfaces sources prominently and passes a perplexity.ai referrer, but only when the user clicks through. Many users get their answer without clicking.
Google AI Overviews show citations but the click-through referrer is google.com — indistinguishable from regular Google search.
Microsoft Copilot behaves like Bing in most respects.
Anthropic Claude (via web tools) typically does not pass clean referrer data.

The deeper problem: many AI shopping interactions never produce a click at all. A shopper asks ChatGPT "what are the best budget noise-cancelling headphones?", reads the answer, and either remembers your brand for later or doesn't. There is no click event to attribute. Traditional click-based attribution misses this entirely.

What You Can Track Today

Despite the limitations, you can extract meaningful signal from referrer data and indirect indicators. Here is what to monitor.

1. Direct ChatGPT and Perplexity referrers

When a user clicks through from a citation, the referrer header includes the source domain. Set up custom segments in Google Analytics 4 (or your analytics tool of choice) to track:

chat.openai.com — direct clicks from ChatGPT citations
chatgpt.com — newer ChatGPT domain (post-2025)
perplexity.ai — Perplexity citation clicks
www.perplexity.ai — variant
copilot.microsoft.com — Microsoft Copilot
claude.ai — Claude direct citation clicks

This captures the direct-click portion of AI traffic. It will be small (most AI users do not click) but it is real, attributable, and growing.

2. Branded search as AI proxy

The largest signal of AI exposure is not direct clicks — it is branded search lift. When ChatGPT recommends your brand, users do not always click the citation. Many open a new tab and search Google for your brand directly.

If your branded search volume (your store name, your brand name, your product names) is rising faster than your overall organic search, that delta is largely attributable to AI exposure. Set up a Google Search Console weekly report that tracks:

Total impressions for branded queries (your name + variants)
Total impressions for non-branded queries
Branded-to-non-branded ratio over time

Rising ratio = AI is mentioning you more, even if you cannot pinpoint which AI engine or which query.

3. UTM tagging on llms.txt links

This is a tactic almost no one uses. Add UTM parameters to the URLs in your llms.txt file:

## Products

- [Trail X — Carbon Plate Trail Shoe](https://yourstore.com/products/trail-x?utm_source=llms_txt&utm_medium=ai&utm_campaign=catalog)

When an AI engine cites a URL from your llms.txt and a user clicks, the UTM parameters travel with the click. You will see utm_source=llms_txt in your analytics, definitively attributable to AI surfacing your llms.txt-listed pages.

Do not over-tag. Use UTMs only on llms.txt links to keep the signal clean. Your sitemap.xml URLs and on-site internal links should not have UTM parameters.

4. Conversion path patterns

AI-influenced sessions often have telltale patterns:

Long landing-page-to-conversion paths — a user lands on a deep product page (not the homepage), browses a few related products, and converts. This is the fingerprint of a researched purchase, often AI-influenced.
Specific product searches with no Google referrer — direct traffic to a specific product page (not via your homepage) often comes from AI citations.
Long sessions with high time-on-site — AI shoppers tend to do more research before buying.

Build a custom segment in GA4 that combines these patterns: direct or empty referrer + landing page = product page + session duration over 60 seconds. This segment is your "likely AI-influenced" cohort.

Setting Up GA4 for AI Attribution

Specific GA4 configuration steps:

Custom dimension: AI source

Create a custom dimension at the session level called "ai_source". Populate it via Google Tag Manager based on the page referrer:

function() {
  var ref = document.referrer.toLowerCase();
  if (ref.indexOf('chat.openai.com') > -1 || ref.indexOf('chatgpt.com') > -1) return 'chatgpt';
  if (ref.indexOf('perplexity.ai') > -1) return 'perplexity';
  if (ref.indexOf('copilot.microsoft.com') > -1) return 'copilot';
  if (ref.indexOf('claude.ai') > -1) return 'claude';
  if (ref.indexOf('gemini.google.com') > -1) return 'gemini';
  return null;
}

Now you can filter and report on AI-sourced traffic specifically.

Custom event: ai_referral

Fire a custom event on every page view where ai_source is non-null. This lets you track AI-referred sessions separately from total sessions and measure their conversion rate.

Branded vs non-branded organic split

Connect Google Search Console to GA4. In Search Console, build a regex query filter for your branded terms vs non-branded. Compare the two over time. Rising branded ratio = rising AI exposure.

Multi-Touch Attribution: The Honest Gap

The hard truth: even with everything above set up correctly, you will not be able to fully attribute AI's impact on sales. Reasons:

Zero-click conversions — many AI-influenced purchases never produce a measurable click event from an AI engine.
Cross-device lookups — a shopper asks ChatGPT on their phone, then buys on their laptop hours later. The link is broken.
Memory-based purchases — AI mentions your brand, the user remembers it, and they search Google directly weeks later. Attributed to "branded search" but caused by AI.

Industry analysts estimate that direct AI referrer tracking captures only 10-30% of true AI-influenced sessions. The rest leaks into "direct," "branded search," and "(unknown)" buckets.

This is the same problem that plagued early SEO attribution (organic search dropped many users into "(direct)" buckets when referrer headers were stripped) and early social attribution (people would see a Facebook ad and Google the brand later, attributing to organic). The pattern repeats with AI.

The Pragmatic Measurement Framework

Given the limitations, the most useful approach is a layered measurement model:

Measure direct AI clicks via referrer-based segments. Track week-over-week growth.
Measure branded search lift via Search Console. Compare quarter-over-quarter changes.
Measure llms.txt-tagged conversions via UTM parameters. This is your cleanest direct-attribution signal.
Measure conversion-path quality for direct + product-page-landing sessions. These are the AI-likely cohort.
Run periodic citation checks — manually query ChatGPT, Perplexity, and Google AI for your category. Track which sites get cited and how often you appear.

None of these signals alone is conclusive. Together, they give you a defensible picture of AI's impact on your business — enough to justify continued investment in AI-readability infrastructure and content.

What Not to Do

Three common mistakes:

Do not optimize for AI traffic in isolation

Most AI shoppers also use Google. Most also use social. AI is a layer in the discovery stack, not a replacement. Optimize for AI as a complement to traditional SEO and social, not as a substitute.

Do not chase vanity metrics

"Mentions in ChatGPT" is not the same as "revenue from ChatGPT." Track citations, but tie them to revenue indicators (branded search, direct conversions, product-page-landing patterns). Citation count without revenue context is theater.

Do not over-tag your URLs

Tagging every internal link with AI-attribution UTMs creates analytics chaos. Reserve UTMs for the cleanest external entry points: llms.txt links, AI-engine partnerships (when those exist), and dedicated AI-channel campaigns.

The Bottom Line

AI shopping attribution in 2026 is where Google search attribution was in 2008: imperfect, fragmented, and still worth doing. The merchants who set up tracking now will have the data to optimize when the channel matures. The merchants who wait will have years of missing data when AI shopping becomes a meaningful share of revenue.

Three things to do this week:

Add custom segments in GA4 for ChatGPT, Perplexity, Copilot, Claude, and Gemini referrers.
Set up Search Console branded vs non-branded query tracking.
If you have an llms.txt file, add UTM parameters to the product URLs in it.

If your platform does not give you llms.txt or static HTML out of the box, the attribution problem is the smaller of your two challenges. Fix the discoverability first.

See how BusinessCart.ai builds AI-readability into every storefront →