Track ChatGPT Traffic to Your Store and Find AI-Referred Orders

To see how many visits and orders come from ChatGPT and Perplexity, open GA4's traffic acquisition report, switch the dimension to session source, and filter for chatgpt.com and perplexity.ai. Check the AI Assistant channel in your default channel group, then match purchases to orders in your store admin. Expect the figure to be an undercount.

The undercount is the part people miss. Some visits arrive with no referrer and are filed as Direct, some people read a recommendation and search for your name later, and some assistants open links inside apps that strip the source. The method below gives you the tracked number, a sensible way to estimate the gap and a monthly routine that takes about twenty minutes.

Follow me on Instagram@sagnikteaches

What an AI-referred visit looks like when it arrives

When someone clicks a link in a ChatGPT or Perplexity answer, the browser usually tells your site where the click came from. That label, the referrer, is what analytics tools read to decide which channel gets the credit. Two things can identify an AI visit:

Connect on LinkedInSagnik Bhattacharya

• The referrer domain. The visit shows a source such as chatgpt.com or perplexity.ai. When the referrer is on Google's list of AI assistants, GA4 now sets the medium to "ai-assistant"; older sessions, and assistants Google hasn't listed, show the medium "referral". Other assistants appear under their own domains, and the domains change as products are renamed (copilot.microsoft.com now redirects to copilot.com), so a list you write today needs reviewing.
• A UTM tag on the landing page address. UTM parameters are labels added to the end of a web address, such as ?utm_source=chatgpt.com, that tell analytics where a click came from. Reports on ChatGPT referrals commonly describe links carrying this tag. OpenAI's own crawler documentation doesn't mention it, so treat it as something to look for in your data, not something to depend on.

Either label can be missing. A visit with no referrer and no tag is filed as Direct, and nothing you do afterwards can recover it. That's why the setup has three layers: the tracked count, a check for the traffic that hides in Direct, and a human question at checkout.

Subscribe on YouTube@sagnikteaches

Step 1: Find the visits in GA4 without changing anything

Start with what is already in your data. In GA4, open Reports, then Acquisition, then Traffic acquisition. The report's default primary dimension is Session default channel group. Change it to Session source / medium and look through the rows. You are looking for entries such as chatgpt.com / ai-assistant, chatgpt.com / referral (the label on older sessions) or perplexity.ai / referral, and a row such as chatgpt.com / (not set), which matches no default rule and so sits in Unassigned rather than in any AI figure. Use the search box above the table to type "chatgpt", then "perplexity", then the names of the other assistants you know of.

Write down for the last 90 days: sessions, engaged sessions, key events (what GA4 used to call conversions) and total revenue for each AI source. Ninety days gives enough visits to mean something on a small store, and a single month can be dominated by one article that was cited by an assistant for two weeks.

The AI Assistant channel Google now supplies

Google's documentation for the default channel group lists an AI Assistant channel, covering visits from sources such as ChatGPT, Gemini, Deepseek, Copilot and Grok. When a session's referrer is on Google's list of AI assistants, GA4 sets its medium to "ai-assistant" (and its campaign to "(ai-assistant)"), and the channel's rule is simply that the medium exactly matches "ai-assistant". Google publishes its source list as a downloadable spreadsheet on the same help page, so you can see which assistants it recognises. Two details matter. First, Google's own AI Overviews and AI Mode are excluded from this channel and counted in Organic Search, so you can't separate them in the channel report. Second, the default channel group can't be edited, and your property might use a custom primary group that predates the new channel. If the channel doesn't appear in your Traffic acquisition table, check whether your property's primary channel group is the default one, or add your own rule as described below.

Step 2: Add your own AI channel for control

A custom channel group lets you define the list of AI sources yourself, so a new assistant is added the day it appears, not whenever the default catches up. In GA4 this is done under Admin, Data display, Channel groups (you need the Editor role or above), where you create a new group, add a channel called "AI assistants" and give it a condition on Source. If you have no groups yet, GA4 starts you with a copy of the default group, which is the quickest route: add your channel to the copy and every other channel keeps Google's rules. A standard property can hold two custom channel groups alongside the default, each with up to 50 channels, so don't spend both slots on experiments. Menu names move around, so if you can't find them, search the Analytics help for "custom channel group".

The condition is a regular expression, a compact pattern that matches several names. Use "matches regex" and wrap the list in .* at both ends. GA4's "matches regex" is treated as a match on the whole value (Google's own example pattern wraps every name in .* for this reason), so a bare chatgpt can fail to match chatgpt.com:

.*(chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|copilot\.com|grok\.com|deepseek\.com).*

Order matters. Google's help says traffic goes into the first channel whose definition it matches, in the group's current order. If your channel sits below Google's own AI Assistant channel (which a copy of the default group includes), that channel claims every assistant on Google's list; if it sits below Referral, Referral claims the assistants Google hasn't listed. Either way your channel undercounts. Move it to the top with the group's Reorder option (Google's own example puts its AI channel above Referral). Google's help says custom channel groups can be applied to reports retroactively, so once the group is saved you can point it at last quarter and see the AI channel's history straight away; the exception is audiences, where a change only affects who qualifies from then on.

Two cautions. A domain on your list can host something else (a lookalike site, or a product that shares the name), so check the landing pages of your first week's visits. And lists go stale: put a calendar reminder to review the sources every quarter. You can ask the AI to do this review:

Here is an export of my store's last 90 days of GA4 traffic: session source,
medium, sessions, purchases, revenue (about 300 rows). Label each source as
one of: search engine, social network, email, AI assistant or AI search,
marketplace or comparison site, partner or other site, unknown. List the AI
ones with their totals. Flag any source you are unsure about instead of
guessing, and give the one-line reason. Do not change any numbers.

Illustrative output: it finds chatgpt.com (360 sessions), perplexity.ai (85), copilot.com (31), gemini.google.com (14) and flags "you.com / referral" and "phind.com / referral" as probably AI search, which you check and add to the pattern. Your job afterwards is to verify the totals against the GA4 table: the AI may mislabel a source, and it may drop a row from a long list without saying so. Recount the rows it labelled AI.

Step 3: Read the tag, not just the source

A visit's landing page address, with everything after the question mark, tells you more than the source. In an Exploration report (the free-form tool in GA4's Explore section), add the dimension Landing page + query string, which Google defines as the page path and query string of the first page view in a session. Filter for addresses containing utm_source=. Then check the values:

• utm_source=chatgpt.com on a visit that GA4 also lists under source chatgpt.com is consistent, and you can trust the row.
• A visit tagged utm_source=chatgpt.com that sits in Unassigned or Direct rather than an AI channel tells you the tag survived while the medium or referrer didn't. Count it as ChatGPT.
• Tags you didn't create and can't explain, such as utm_source= followed by an unfamiliar name, are worth a look. They often belong to a new assistant or a browser feature.

Be careful with one trap: your own tagged links. If you or your agency use UTM tags in emails and posts, give them different source names from the AI ones, so that "chatgpt" in your reports never means a newsletter. The names you choose are free, so keep a short list of your own conventions in a document.

Step 4: Find the orders in Shopify

If your store runs on Shopify, you have a second set of figures that doesn't depend on GA4. Shopify's acquisition reports include Sessions by referrer, which shows whether visitors came directly, from search or by a referral, and Shopify's documentation describes dimensions you can group by, such as the referring channel and the referring medium. It also says UTM-related dimensions show the raw data from the session. These reports apply to the Online Store sales channel only, so sales through other channels aren't included. Shopify's pages don't state which plans include each report, so check what your plan shows. Look in Analytics, then Reports, open the Category filter and choose Acquisition.

To find AI-referred orders, use the report's sessions by referrer for the source names, then look for a sales view grouped by the same dimension if your plan offers one. Where it doesn't, take the date and landing-page list of AI visits from GA4 and match them to orders by day. For a small store this is a monthly job on a handful of orders, so you can open them. Each order page in Shopify has a Conversion summary section: open View conversion details, then View full sessions, and the visit details include the session referral and any UTM parameters. The summary is shortened or blank when cookies were blocked, on headless storefronts, for draft orders and for orders from channels other than the Online Store or Hydrogen, so a blank one is not proof the customer came direct. Using the order page as ground truth is still worth it, because it is the one place where the store records what actually happened to a real customer.

The same questions can be asked of Shopify Sidekick in plain words, though verify any figure it gives against the report itself; and this comparison of Shopify reports, GA4 and ChatGPT for ecommerce analytics explains which tool to open first for which question.

The worked example: one month for an online shop

An illustrative online shop sells grooming tools for dog groomers. In one month GA4 shows 62,000 sessions. By default channel:

The AI Assistant channel is 0.8% of sessions and 1.2% of orders (15 of 1,223), because its visitors convert at 3.0%, better than the 2.0% overall, which is plausible: someone who has asked an assistant for a recommendation and clicked through is further along than someone browsing. With an average order of $64, those 15 orders are about $960 of revenue. That is the tracked figure and it is small enough to be tempting to ignore. Three further checks stop that being the end of the story.

Estimating what is hiding in Direct

Direct is 16,200 sessions, and some of that is AI traffic whose referrer was stripped. You can't prove which, but you can narrow it. In GA4's landing-page report, filter Direct sessions to pages nobody types in by hand: a long product-comparison article, a specific product with a long name, a guide page. A person typing your address lands on the home page. Say 2,100 of the 16,200 Direct sessions landed on deep content pages and 1,300 of those were new users. That's a group of 1,300 that includes AI visits, links shared in messaging apps and bookmarks from other devices. Don't treat all 1,300 as AI, and don't multiply the tracked 500 by an invented factor. Report it as "tracked AI: 500 sessions; possible additional AI inside Direct: up to 1,300, unproven".

Asking the customer

Add one question after purchase, on the order confirmation page or in the confirmation email: "How did you first hear about us?" with options including "An AI assistant such as ChatGPT or Perplexity". Illustratively, of 200 answers in the month, 9 chose the AI option, which is 4.5% of respondents, against 1.2% of orders attributed in analytics. That gap is more informative than either figure alone: it suggests analytics is catching a quarter of the effect. It's a small survey with self-selected answers, so treat it as a pointer. If the question returns 4 answers in 200 over three months running, it's a smaller effect than you hoped; if it returns 20, you have learnt that assistants are influencing buying well beyond what is tracked.

Branded search as a third signal

Some people ask an assistant who sells good grooming tools, get your name, then search for it. They appear as Organic Search from a branded query. In Google Search Console, watch queries containing your shop's name month on month. A rise in branded searches after your content begins appearing in assistant answers is circumstantial evidence, not proof, but across several months it carries weight, especially when the survey agrees.

Why the number is always low

Hold these causes in mind each time you present the figure:

• Stripped referrers. Privacy settings, some apps and some browsers pass no referrer, so a real click arrives as Direct.
• No click at all. An assistant recommends you in the answer, the person remembers the name and arrives by search or by typing later. Nothing links the two.
• Consent and blockers. A visitor who declines analytics cookies, or runs a blocker, may not appear in GA4 at all, though your store's own order record still shows the sale.
• Tags without a medium. A visit that carries utm_source=chatgpt.com but no medium can show as chatgpt.com / (not set), which matches neither the AI Assistant rule nor Referral and lands in Unassigned. Check Unassigned's sources each month.
• Different devices. The question goes to an assistant on a phone, the purchase happens on a laptop, and the second session is Direct.
• AI search inside a search engine. Google's AI Overviews and AI Mode are counted as Organic Search, so they inflate your search number without a separate label.

Give your reader the tracked number, the range and the reasons in one sentence. A figure that is honestly described as a lower bound is more useful than a confident number that someone later disproves.

A report that showed zero AI traffic

Here is an illustrative failure that is easy to repeat. A small store copies an AI channel from an older guide: "Medium exactly matches referral" and "Source matches regex" with the list above. It worked while assistant visits arrived as referrals. Once GA4 began labelling recognised assistants with the medium "ai-assistant", ChatGPT visits stopped meeting the first condition, the custom channel fell to almost nothing, and the owner concluded that people had stopped using assistants to find the shop. The source / medium report told a different story: 410 sessions from chatgpt.com / ai-assistant that month. The check that would have caught this in a minute is to compare the AI channel's total with the sum of sessions for the AI sources in the source / medium report. If they differ, the conditions, the pattern or the rule order is wrong. Condition on source alone, as in Step 2.

Quick variants for other store types

• A bakery selling gift boxes through its own site: low volume, so skip the channel group and do a monthly check of source / medium plus the checkout question, which gives you more than analytics on thirty orders.
• A café with an online beans shop: watch the landing pages, since assistants tend to send people to guides and comparison pages rather than the home page, and make sure those pages carry a clear path to the shop.
• A driving school selling lesson packages online: the "order" is a booking. Track booking-form completions as the key event, and ask on the form where people heard about the school.
• A store with several markets and languages: segment the AI channel by landing-page language to see where assistants send people, because the regional split often differs from search.

The monthly routine, in twenty minutes

1. Open Traffic acquisition for the last month, sorted by session source / medium. Export it.
2. Check the AI channel total equals the sum of its sources; fix the rule if not.
3. Paste the new sources (those not on last month's list) into the AI labelling prompt and review.
4. Record sessions, orders, revenue and conversion rate for the AI channel on a one-line sheet, with the 90-day trend.
5. Record the number of survey answers choosing AI, and the month's branded search clicks.
6. Look at the five landing pages that received the most AI sessions and ask what they have in common. If one page draws half the visits, make it the best page on your site.

That last step is where the numbers become decisions. AI assistants tend to cite pages that answer a question plainly, with facts, prices and clear structure. If a guide is already attracting assistant traffic, update it, link it to the products it mentions and check that your robots.txt isn't blocking OAI-SearchBot, the crawler behind ChatGPT's search: OpenAI's crawler documentation says sites opted out of it are not shown in ChatGPT search answers, though they can still appear as navigational links. The wider question of getting recommended in the first place is in how to get your business recommended by ChatGPT and AI search, and the Perplexity and ChatGPT comparison explains how the two assistants differ in the answers they give.

How to know your tracking works

Run a test. Ask an assistant a question your store should answer, find your page in the response and click it, using a normal browser. Within a day, look in GA4's real-time report or the next day's source / medium table for the visit: you should see a source naming the assistant. Repeat on your phone in the assistant's app. If the app visit appears as Direct, you have just seen the undercount first-hand, and you know it exists on your own store, not just in a tutorial. Keep the test results in your notes, with the date, since assistants' apps change how they open links, and re-run it each quarter.

Questions about tracking AI traffic to a store

Does GA4 have a channel for ChatGPT traffic?

Google's documentation lists an AI Assistant channel in the default channel group, covering sources such as ChatGPT, Gemini, Deepseek, Copilot and Grok, with the medium set to ai-assistant. Google's own AI Overviews and AI Mode are counted under Organic Search instead. Older custom primary groups may not show the channel, so check yours.

Why do some ChatGPT visits show up as Direct?

When a browser, app or privacy setting strips the referrer, the visit arrives with no source and GA4 files it as Direct. Links copied from an answer and pasted elsewhere, apps that open links without a referrer and people who read a recommendation and type your name later all look the same.

Can I add a UTM tag so ChatGPT visits are labelled?

You can't control the links AI assistants build, so you can't rely on adding tags. You can tag links you publish yourself, and you can read the tags that already appear on incoming visits. ChatGPT links are often seen carrying utm_source=chatgpt.com, but OpenAI doesn't document that as a promise, so check your own reports.

How many AI-referred orders is normal for a small store?

There is no reliable benchmark to quote, and published figures vary with the products and the sample. Measure your own baseline for three months, track sessions, orders and revenue from the AI channel, and judge movement against that rather than against someone else's headline figure.

How do I find the orders and not just the visits?

Filter the traffic or acquisition report by your AI channel and look at the purchases or revenue columns, then compare the order records in your store admin for the same period. Add a short post-purchase question about where the customer heard about you, since it catches people who never clicked a tracked link.

Further reads

• How Wedding Businesses Get Found When Couples Ask AI for Suppliers — See the same visibility problem for a service business.
• Best AI for Ecommerce Operations: Orders, Stock, Returns, Suppliers — Look at the orders and stock side of an online store.
• Best AI Assistant for Shopify: Sidekick, ChatGPT or Claude by Job — Pick the right assistant for each Shopify job.
• How to Check Sources and Citations in AI Research — Check the sources an assistant cites before relying on them.
• Best AI Chatbots for Small Shopify Stores (2026) — Capture the visitors who arrive, with a store chatbot.
• How to Write Shopify Product Descriptions With AI That Convert — Improve the product pages assistants send people to.
• AI for Ecommerce Optimisation: Audit One Product Page at a Time — A single-page teardown uses analytics, customer questions and a controlled edit to find a plausible conversion obstacle.
• Shopify RFM Analysis: Segment Customers With ChatGPT From an Export — Score customers on recency, frequency and spend from a Shopify export, then write one email per segment, keeping personal data minimal.
• Shopify SEO With AI: Fix Titles and Meta Descriptions in Bulk — A CSV round trip for Shopify SEO text: export, rewrite with AI in batches, check length and duplicates, import, and judge it in Search Console.
• llms.txt for Shopify: Test Whether AI Bots Fetch It and Send Traffic — A before-and-after test of llms.txt on one Shopify store, with what Shopify serves by default and the product data that matters more.
• Perplexity Merchant Program: Should a Small Online Store Join? — What vendor guides say about the free programme, who can join, what a product feed needs, and a simple sum to decide whether the prep time is worth it.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Google Analytics Help (default channel group definitions including the AI Assistant channel, custom channel groups, regular expressions, Traffic acquisition report); Shopify's help pages (acquisition reports, order conversion summary); OpenAI's crawler documentation; the figures in the example store are illustrative.