How to track sales from ChatGPT and AI assistants

Shoppers increasingly ask AI assistants what to buy. ChatGPT, Perplexity, Gemini, and Claude now recommend products and link to stores. That traffic is real, but most analytics either mislabel it or miss it entirely.

There are three separate problems: seeing AI-driven visits, attributing the sales they produce, and paying for them when an agent or partner is involved.

Problem 1: AI traffic often looks like direct traffic

When someone clicks a link inside an AI answer, a few things can happen:

• The visit arrives with a referrer or a utm_source naming the assistant, and analytics can see it.
• The app opens the link without a referrer, so it shows up as direct.
• The user copies the product name and searches for it later, so it shows up as organic search or direct.

So the AI share of your traffic is almost always undercounted.

How to see more of it

1. Check referrers and UTM sources in your analytics for AI assistant domains and labels, and group them into one channel.
2. Ask at checkout. A "How did you hear about us?" field with an "AI assistant" option catches what tracking misses.
3. Watch branded search. Rising searches for your brand alongside flat ad spend often reflect AI recommendations.

Problem 2: Attributing the sale

Seeing a visit is not the same as crediting the sale. For that you need the same machinery affiliate programs use: a tracked link that records the click, and a server-side sale from your payment system matched back to it. See how affiliate tracking works.

If an assistant or agent shares your tracked link, the click is recorded and the purchase can be matched to it, even if the buyer pays days later.

Problem 3: Paying AI agents commission

A new kind of partner is emerging: AI agents built by developers and companies that recommend products to users. These agents can earn affiliate commission just like a creator or a publisher.

Two things make agent attribution trustworthy:

• A signed recommendation. When the agent recommends a product, it gets a short-lived signed token or a tracked link tied to that specific recommendation. A sale that carries the token is verified: the agent provably sent it.
• A human operator who gets paid. The person or business that runs the agent is the partner of record. They complete payout setup and receive the tax forms, exactly like any affiliate.

Sales without a token, where the agent claims credit after the fact, should be treated as claimed, not verified, and paid under stricter rules if at all.

Problem 4: Getting recommended in the first place

Assistants recommend what they can find and verify. Make that easy:

• Answer questions directly. Pages that open with a clear, specific answer are the easiest for AI to quote.
• Use comparison tables with real specs, prices, and differences.
• Add FAQs with structured data, so questions and answers are machine-readable.
• Keep product data accurate in your store and product feeds.
• Be specific and checkable. Concrete numbers and facts beat marketing language.

How Bridgly handles AI attribution

Bridgly treats AI agents as partners. An agent operator gets an agent key; each recommendation can produce a share link carrying a signed token. When a buyer clicks and pays, the sale arrives from Stripe or Shopify and is matched back. Verified sales and claimed sales are labelled separately, and commission is paid to the agent's operator through the same payout flow as any partner.

The short version

• AI assistants send buyers, but analytics usually undercounts them.
• Group AI referrers into one channel and add an "AI assistant" option at checkout.
• Tracked links plus server-side sales attribute AI-driven purchases properly.
• AI agents can be paid as affiliates, with signed tokens separating verified from claimed sales.
• To be recommended more, publish direct, specific, checkable answers.

See agent attribution in Bridgly → · Mobile attribution basics