How to Track AI Search Referrals in Plausible and GA4 (ChatGPT, Perplexity, Claude)

To track AI search referrals, look for chatgpt.com and perplexity.ai in the Sources or Referral report of your analytics tool, and filter for utm_source=chatgpt.com on tagged links. ChatGPT shows up in analytics as the referrer domain chatgpt.com and appends utm_source=chatgpt.com to clickable links (since mid-2025). Perplexity passes perplexity.ai as the referrer on every citation click. Both appear under the Referral or Sources section of Plausible and GA4 — but only for the roughly 20–30% of AI visits that include a referrer header. The rest show up as Direct.

TL;DR: Most AI traffic is invisible in default analytics, but 20–30% is trackable right now using referrer values and UTM parameters that ChatGPT and Perplexity already pass. This post walks through the exact Plausible and GA4 setup, explains why the rest shows up as "Direct," and shows how TheSaaSDir — a curated directory of SaaS and AI products with dofollow backlinks — used its own Plausible data (11 ChatGPT visits, 2 Perplexity visits) to measure AI referral quality and decide where to invest next.

Here is the scenario this post is for: you have done some GEO work, someone tells you ChatGPT mentioned your tool in a recent answer, and you go to your analytics expecting to see the proof. You find nothing obvious. Maybe a small bump in Direct traffic. Maybe nothing at all. This post covers the exact setup to track AI search referrals in Plausible and GA4, with the referrer values and UTM patterns that ChatGPT and Perplexity actually pass. The fix for that confusion is straightforward — but you need to set up analytics correctly first, and you need to know what you can and cannot actually track. If you have not yet checked whether AI tools are recommending your product at all, start with how to monitor if your SaaS is being recommended by AI — this post picks up where that one ends.

The Attribution Problem You Need to Understand First

AI referral traffic attribution is harder than standard referral tracking for three structural reasons. Ignoring them is why most founders feel like their analytics is broken.

• Referrer-Policy stripping. chatgpt.com sets a strict referrer policy (strict-origin-when-cross-origin), which strips the full referrer path when linking out to your domain. The referrer header that arrives at your server contains the origin (chatgpt.com), not the full conversation URL — so you can see the source but not which ChatGPT thread sent the visitor.
• Mobile app and copy-paste traffic. ChatGPT's mobile app does not pass any referrer at all. Users who copy a URL from a ChatGPT response and paste it into a new browser tab also arrive with no referrer. Both show up as Direct in every analytics tool.
• The 70/30 split. Approximately 70.6% of AI-referred traffic arrives with no referrer header (per Clickport's 2026 analysis). The setup below captures the ~30% that does pass referrer data. Do not expect your analytics to show "all AI traffic" — expect it to show a traceable floor that underrepresents actual AI impact.

Set this expectation correctly up front. A founder who sets up the GA4 custom channel expecting to see every ChatGPT click will think the setup failed. It did not fail — that is how the web works right now.

What the Dark Traffic Tells You Anyway

Even unattributed AI traffic leaves a signal. A sustained spike in Direct traffic after GEO work — a directory submission, a new comparison post, a Reddit thread that took off — is a strong signal of AI referral activity. The triangulation method is simple: baseline your Direct traffic volume in the four weeks before any GEO push, then monitor for sustained uplift in the four weeks after. Not perfect attribution, but directionally useful and free.

The Referrer and UTM Reference Table (Read Before Setting Up Either Tool)

You need to know exactly what string to filter on. Here is the reference table — the practical lookup no competitor publishes cleanly:

AI Source	Referrer Domain	UTM Parameter (if any)	Reliability
ChatGPT Web	chatgpt.com	utm_source=chatgpt.com (since mid-2025)	High — passes referrer + UTM on desktop web
ChatGPT Mobile App	(none — shows as Direct)	none	None
ChatGPT (older / API)	chat.openai.com	none	Low — legacy domain, less common now
Perplexity	perplexity.ai	none	High — passes referrer on all citation clicks
Claude	claude.ai	none	Inconsistent — some sessions pass referrer, many do not
Google AI Overviews / Gemini	gemini.google.com	none	Low — mostly direct
Microsoft Copilot	copilot.microsoft.com	none	Medium
Phind	phind.com	none	Medium
You.com	you.com	none	Low

Two key callouts. Perplexity is the only major AI engine that both (a) reliably passes referrer headers and (b) never uses UTM parameters — so referrer-based tracking is the only way to catch it, but it works on essentially every citation click. ChatGPT is the only major AI engine that appends UTM parameters by default, which is why ChatGPT desktop web is the most reliably trackable form of AI referral traffic in 2026. Everything else is inconsistent.

How to Track AI Search Referrals in Plausible

Plausible's Sources view records referrers automatically — most founders running Plausible already have AI referral data they have not looked at yet. Here is the full setup.

Step 1 — Check Your Sources Tab First (You May Already Have Data)

Many Plausible users have AI referral data sitting in their dashboard, unfiltered. Exact path: Dashboard → Sources tab → scroll the list or use the search box for chatgpt.com and perplexity.ai. If any AI-referred visits exist, they appear automatically with no configuration required.

Worked example: TheSaaSDir's own Sources tab showed chatgpt.com with 11 visits and perplexity.ai with 2 visits during a recent audit period. Both appeared without any special setup — the referrer headers were already being captured. That is the floor. The real number of AI-driven visits is higher (the dark-traffic majority is invisible), but 13 confirmed sessions is enough to verify the citation pool exists and the signal is real.

If nothing shows up for you, it means one of three things: (a) AI engines are not citing you yet, (b) all your AI traffic is in the dark-traffic bucket, or (c) your GEO visibility is still limited. That is useful information either way — it means the next investment is getting into the AI citation pool. If you have not yet confirmed whether any AI tool is actually recommending your product, start with how to monitor if your SaaS is being recommended by AI before running a GEO campaign. The step-by-step GEO guide is the playbook once you have confirmed the citations exist.

Step 2 — Create a Saved Segment for AI Sources

Plausible lets you filter by multiple sources at once and save the filter as a reusable segment. Exact path:

1. Dashboard → Filter button → Source → contains → enter chatgpt.com.
2. Add filter → Source → contains → perplexity.ai.
3. Repeat for claude.ai, gemini.google.com, copilot.microsoft.com.
4. Save as Segment → name it "AI Traffic."

Once saved, activate the segment from the dashboard any time to see AI-only metrics: sessions, top landing pages, bounce rate, average session duration. This is the recurring view you check weekly.

Step 3 — Set Up Goals to Measure AI-Referred Conversions

Seeing sessions from ChatGPT is interesting. Seeing that those sessions convert to signups at 3x your organic rate is what tells you whether to keep investing in GEO. Exact path for goal setup: Account → Site Settings → Goals → Add Goal → Pageview Goal (for a thank-you page or signup confirmation URL) OR Custom Event (requires adding a plausible() JS call on the conversion action).

For SaaS founders, the recommended goals to set up first:

• /signup/complete or equivalent confirmation URL
• Pricing page visit
• Demo booking confirmation
• Trial activation event

Once goals are live, filter the dashboard by your saved AI Traffic segment to see goal completions from that channel. The conversion-rate comparison vs. organic and direct is the number that tells you whether AI buyers are higher or lower intent than your other channels. For most SaaS products in 2026, AI-referred visitors convert significantly higher: Averi.ai's 2026 data shows AI-referred conversion rates of 14.2% vs. 2.8% for Google organic — a 5x difference that is often what justifies continued GEO investment.

Step 4 — Add UTM Parameters to Links You Control

Plausible reads UTM parameters natively. For any link to your site that you control — newsletter mentions, llms.txt citations, directory listing URLs, your own social posts — append a UTM string like:

?utm_source=ai-directory&utm_medium=referral&utm_campaign=geo-outreach

(The utm_campaign value is yours to customize — use whatever names your tracked campaigns clearly.) This creates a parallel trackable channel for traffic you actively seed, separate from organic AI referrals. When you submit your product to a directory, append a UTM to the listing's outbound link wherever the directory allows custom URLs. It is the easiest way to separate "AI cited me organically" from "I seeded this citation myself."

How to Track AI Search Referrals in GA4

GA4 needs more configuration than Plausible, but the setup is one-time and the resulting custom channel works permanently.

Step 1 — Check What's Already in Your Referral Report

GA4 path: Reports → Acquisition → Traffic Acquisition → change primary dimension to "Session source / medium" → search for chatgpt and perplexity. UTM-tagged ChatGPT traffic shows up as chatgpt.com / referral or with explicit UTM values. Perplexity appears as perplexity.ai / referral.

Most founders will find some AI traffic here already — it is just not labeled as an AI channel, so it hides inside the generic Referral bucket alongside every other referring domain. The custom channel group below pulls it out.

Step 2 — Create a Custom Channel Group for AI Traffic

This is the key setup step. GA4's default channel group has no AI Assistants channel as of May 2026 — you have to build it manually.

Exact path:

1. Admin → Data Display → Channel Groups → New channel group.
2. Add channel → Name: AI Traffic.
3. Condition: Session source matches regex → paste this exact regex:

chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|phind\.com|you\.com

1. Add a second condition with OR logic to capture UTM-tagged ChatGPT traffic even when the referrer header is stripped: Session campaign matches regex chatgpt OR utm_source matches chatgpt\.com.
2. Critical step: drag the "AI Traffic" channel above the "Referral" channel in the channel list. GA4 assigns traffic to the first channel whose condition matches — if Referral sits above AI Traffic, every AI referral gets swallowed into generic Referral and your new channel stays empty.
3. Click Apply and Save Group.

The ordering step is what catches most founders. The regex works. The channel exists. But Referral matches first, so AI Traffic never gets the credit. Fix the order.

Step 3 — Build an AI Traffic Exploration Report

For ongoing monitoring without reconfiguring the report every time:

1. Explore → Blank exploration.
2. Dimensions: Session source, Landing page, Country.
3. Metrics: Sessions, Engaged sessions, Conversions, Revenue (if e-commerce).
4. Filter: Channel = AI Traffic.
5. Save.

This becomes your recurring AI traffic dashboard. Open it once a week and you have the answer to "what is AI doing for me right now?" in 30 seconds.

Step 4 — Set Up a GA4 Custom Alert for AI Traffic Spikes

GA4 Insights (the bell icon) → Create → Custom → Condition: Sessions from AI Traffic channel increases more than 50% week-over-week → Notify by email.

This passive alert tells you when AI referral activity spikes — often the earliest signal that a new citation appeared or that ChatGPT started recommending you in a query category it previously ignored. Faster than checking the dashboard manually.

Reading Your Data — What Do These Numbers Actually Mean?

Setting up the tracking is step one. Knowing what the data is telling you and what to do next is the part no competitor writes. Three scenarios cover what most founders see in their first month of AI traffic data.

Scenario A: Low Volume, High Engagement

Sessions are in the single or low double digits, but session duration is long, multiple pages per session, and conversions look strong relative to organic. Interpretation: AI is recommending you but rarely. The traffic quality validates that GEO investment is worth it — these are high-intent visitors. Priority: expand your AI citation footprint. More directories, more roundup appearances, more G2 reviews. The citation share of voice strategy is the playbook for moving from "occasionally cited" to "default recommendation."

Scenario B: Zero AI Referral Traffic Despite Known Citations

You have manually confirmed that ChatGPT or Perplexity is naming your product, but your Sources tab and GA4 AI Traffic channel are flat. Interpretation: your citations exist but your traffic is mostly landing in the dark-traffic bucket — mobile app users, copy-paste visitors, sessions where the referrer is being stripped. Run the Direct triangulation (compare Direct volume before and after GEO efforts). Also verify: is your product page accessible without JavaScript? Perplexity passes referrers on all citation clicks — if you have confirmed Perplexity citations but zero perplexity.ai sessions, your page likely has a redirect that strips referrers, or your renderable HTML is broken. The 5-layer AI visibility diagnosis covers the technical layer.

Scenario C: Decent Volume, Poor Conversion

Sessions are arriving, but they bounce faster or convert at a fraction of your organic rate. Interpretation: vocabulary mismatch. AI is recommending you for a use case that does not match what your landing page actually offers. Check which pages AI-referred visitors are landing on and what the bounce rate looks like vs. your organic average. If bounce rate is significantly higher, the description AI is pulling from third-party pages does not match what visitors find on arrival. Fix the source pages first, then the landing page copy.

If your AI traffic data shows near-zero sessions, or if your dark-traffic analysis suggests AI activity that is not converting, the most efficient first move is expanding your AI-crawlable citation footprint. Submit your product to TheSaaSDir for a free listing — TheSaaSDir is a curated directory of SaaS and AI products with dofollow backlinks, explicitly crawlable by GPTBot, PerplexityBot, and ClaudeBot, and every listing includes SoftwareApplication schema markup. It is a Layer 4 citation signal you can ship in 20 minutes.

TheSaaSDir's Own AI Referral Data — A Founder-to-Founder Look

Here is what the Plausible setup above produces on a real site — TheSaaSDir's own dashboard during a recent audit period:

• chatgpt.com: 11 visits
• perplexity.ai: 2 visits
• Total tracked AI sessions: 13

The 11 ChatGPT visits landed primarily on category pages and individual product listings — pages that match the kind of recommendation query a buyer would actually run ("best [category] tools" type prompts where AI cites a directory as a source). Session duration on those AI-referred visits was meaningfully higher than the site's overall average, which lines up with the broader pattern: AI-referred visitors arrive with intent.

The 2 Perplexity visits both landed on category-level pages. Lower volume, but consistent with Perplexity's smaller overall traffic share across the SaaS category compared to ChatGPT.

What this data tells us about TheSaaSDir's current state: the site is in the citation pool for both engines. The signal exists. The absolute number is small — 13 sessions in a single audit window — but the point is not that AI is sending huge traffic yet. The point is that the signal is trackable, the citation pool inclusion is confirmed, and the volume will grow as AI citation activity grows industry-wide.

What we are doing next based on this data: more directory cross-listings, refreshed comparison content, monthly monitoring on the same query set, and quarterly updates to the highest-traffic category pages. Honest founder framing: this is early-stage data, treated as a baseline to grow from rather than evidence of dominance. Most SaaS sites running a similar audit will find similar numbers right now. The compounding starts when you treat 13 sessions as a starting point worth measuring.

Frequently Asked Questions

How do I know if ChatGPT is sending traffic to my site?

Check your analytics Sources tab (Plausible) or Traffic Acquisition report (GA4) for chatgpt.com as a referrer, or filter for sessions where utm_source=chatgpt.com. ChatGPT web has appended utm_source=chatgpt.com to outbound links since mid-2025, making desktop sessions consistently trackable. Mobile app traffic and copy-paste visits do not pass referrer data and show up as Direct — so your analytics will undercount total ChatGPT traffic, often by a factor of three or more. The number you see is a reliable floor, not the ceiling.

Does ChatGPT show up in Google Analytics?

Yes — GA4 records ChatGPT traffic under Session source chatgpt.com when the referrer header is present, and separately captures utm_source=chatgpt.com when the UTM parameter is appended. Without a custom channel group, this traffic is mixed into the generic Referral bucket and is easy to miss. Set up the AI Traffic custom channel group with the regex in this post to isolate it. Expect your GA4 total to be 20–30% of actual ChatGPT-referred traffic — the rest arrives without referrer data because of mobile apps, copy-paste behavior, and referrer-policy stripping.

Why does AI traffic show up as direct in GA4?

AI traffic shows up as direct in GA4 for three reasons. First, ChatGPT's mobile app does not pass referrer headers at all. Second, users who copy a URL from an AI response and paste it into a new browser tab generate no referrer signal. Third, chatgpt.com uses a strict referrer policy that strips everything after the domain origin when linking externally — the referrer arrives as chatgpt.com but with no path, and some analytics configurations misclassify origin-only referrers as direct. The cumulative effect: approximately 70.6% of total AI-referred traffic is unattributable through standard referrer-based tracking, no matter what tool you use.

How do I track Perplexity referrals in analytics?

Perplexity is the easiest AI engine to track because it consistently passes the referrer header perplexity.ai on every citation click and does not strip it. In Plausible, it appears in the Sources tab automatically with no configuration. In GA4, filter Traffic Acquisition for perplexity.ai / referral. Unlike ChatGPT, Perplexity does not append UTM parameters — referrer-based tracking is the only mechanism, but it works reliably on both desktop and mobile web. If you have confirmed Perplexity citations but see zero perplexity.ai sessions in your analytics, the problem is on your site (redirect stripping referrers, broken renderable HTML), not on Perplexity's side.

What is utm_source=chatgpt.com?

utm_source=chatgpt.com is a UTM parameter that ChatGPT's web interface appends automatically to any URL it cites when a user clicks a link in a ChatGPT response on desktop — behavior that started in mid-2025. When a visitor arrives at your site from a ChatGPT citation, their URL contains ?utm_source=chatgpt.com, which Plausible, GA4, and any standard analytics tool records as a trackable campaign source. This makes ChatGPT desktop web clicks the most reliably trackable form of AI referral traffic — more reliable than referrer-based tracking, because UTM parameters survive even when referrer headers do not.

Should I use a paid AI traffic tracking tool instead of setting this up myself?

No — the native Plausible and GA4 setup in this post captures the same trackable 20–30% slice that any paid tool can capture. The dark-traffic majority is invisible to every tool, paid or free, because the data simply does not exist in the HTTP request. Paid tools like Otterly, Profound, and Peec AI add value for ongoing mention monitoring across hundreds of queries (a different problem — covered in the AI mention monitoring guide), not for traffic attribution. Set up the native analytics tracking first. Add a paid tool later only if your manual monitoring workload exceeds an hour a month.

Start Tracking This Week

The signal exists in your analytics right now. You just have not isolated it yet. Twenty minutes of setup in Plausible or GA4 gives you a clean view of the trackable 20–30% of AI referral traffic, a baseline for the dark-traffic triangulation, and a recurring dashboard for measuring whether your GEO work is moving the needle. Most founders skip this step entirely and end up guessing about whether AI is sending them buyers. Do not be one of them.

If your data shows near-zero AI traffic and you want to fix that first, the fastest move is getting listed in a curated directory that is explicitly crawlable by GPTBot, PerplexityBot, and ClaudeBot. Submit your product to TheSaaSDir — free, editorially reviewed, dofollow backlinks, SoftwareApplication schema on every listing. One more vote in the citation pool you can add today.