How do you attribute leads and revenue to AI search?

Why doesn't AI search show up cleanly in analytics?

An AI answer can influence a purchase in three ways, and click-based analytics only sees the first one reliably.

• Click with a referrer. The visitor clicks a cited link in ChatGPT, Perplexity, Claude, Gemini or Copilot in the browser, which sends the origin, for example chatgpt.com. Measurable.
• Click without a usable source. The link opens from an app, is copied to another device, or comes from Google's AI Overviews or AI Mode, which Google reports as organic search. The visit is credited to direct or organic.
• No click at all. The person remembers the name and later types your URL or searches for your brand. Google Analytics files typed URLs under (direct) / (none) (Google Analytics Help); a later brand search counts as Organic Search.

Microsoft's Bing team describes the same shift: the conversion can happen “later or on another device”, and many of these signals “are not captured in traditional analytics as it exists today” (Bing Webmaster Blog, 2025). In this post, dark AI traffic means influence from AI answers that reaches you without an AI source attached.

Which AI platforms pass a referrer, and which don't?

No AI vendor publishes a complete referrer specification for every app and surface. The table lists what public documentation confirms as of September 2026; everything else you should verify in your own analytics.

Two consequences follow. Even a clean referral tells you little: you learn that a visit came from perplexity.ai, never which prompt produced it. And Google's AI features are invisible as a traffic source: you can see how often they show your pages, but their clicks arrive as google / organic. Links opened from apps, copied between devices or retyped usually land in direct; no vendor documents this per app, so check your own reports.

Where do UTM parameters help?

Only on links you control. Nobody can tag the links an AI model decides to cite; ChatGPT's own parameter is the documented exception.

• Tag the links you place yourself, such as links returned by your ChatGPT app or Claude connector, with one utm_source per platform.
• Know the precedence. In GA4 the referrer only sets the source “when no other campaign or traffic source fields have been set” (Google Analytics Help).
• Returns stay credited, within limits. Because “a direct-traffic visit that follows a referred visit will never override an existing referrer”, a same-browser return keeps its AI source. A return via a Google brand search, or on another device, does not.

How do you set up GA4 to report AI referrals?

Start with the default AI Assistant channel. Since May 13, 2026, GA4 assigns medium ai-assistant, campaign (ai-assistant) and channel AI Assistant when the referrer matches a recognized assistant. Google does not publish the full list, or say whether older data was reclassified.

A custom channel group gives you control over the list and history before May 2026. Steps, with English UI labels as of September 2026 (Google Analytics Help):

1. In Admin, under Data display, click Channel groups, then Create new channel group, which starts as a copy of the default group.
2. Click Add new channel, name it (for example “AI assistants”) and use + Add condition group to set Source with matches regex.
3. Click Reorder and move the channel above Referral, because GA4 puts traffic “in the first channel whose definition it matches”. Then Save group.
4. Open Acquisition → Traffic acquisition and select the new group as the primary dimension.

Standard properties allow two custom groups (five on 360) with up to 50 channels each, and custom groups apply to reports retroactively. GA4 regex is a full match by default (Google Analytics Help). Google's example pattern is broad and also matches unrelated sources that merely contain “gpt” or “gemini”. A narrower starting point, to test against your real sources:

^(.+\.)?(chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|chat\.deepseek\.com|grok\.com|chat\.mistral\.ai|meta\.ai)$

Google's AI features stay out of any such rule: their clicks carry Google's referrer, so GA4 can't split them from classic organic results. On the two surfaces, see AI Mode vs. AI Overviews.

Why does “How did you hear about us?” catch dark AI traffic?

A self-reported answer doesn't depend on referrers, cookies or devices. Someone who read a ChatGPT recommendation three weeks ago and then typed your URL on a work laptop appears as direct in GA4, but can still pick “ChatGPT” in your demo form.

The measured share is small: Microsoft Clarity found AI referrals at under 1% of traffic on the publisher sites it studied (2025). Referral counts can't show how much influence sits behind that number; asking the buyer can.

How should you ask the question?

• Ask at the conversion moment: in the signup or demo form, at checkout or on the thank-you page, so the answer sits on a record you can match.
• Make AI search a named option, with a follow-up “Which AI assistant?”. Without it, AI answers disappear into “Google” or “Other”.
• Keep a free-text “Other” and normalize it: “found you on Perplexity” is AI search.
• Mind the option order. Survey research has documented primacy effects in visually presented lists, where earlier options are picked more often (Krosnick & Alwin, 1987). Rotate the order, or at least don't put AI search first by default.

What do self-reports get wrong?

Answers are memories, not logs. People can name the most memorable touchpoint, merge several into one, or say “Google” when they searched after an AI conversation. Treat the answer as the buyer's view, report the response rate next to it, and check it against referral data.

How do you match survey answers to deals and orders?

An answer only becomes attribution once it is attached to money. Match each answer to a conversion with the strongest key available:

1. Order or transaction ID, when the question sits in checkout or on the thank-you page.
2. Email address, ideally hashed, when the answer comes from a form and the purchase or deal follows later.
3. A first-party browser ID, when the same browser answers and converts.

Then fix the rules: a lookback window for how old an answer may be, strict 1:1 matching so one answer can't claim two orders, and no matching on renewals, which would count the original deal twice.

In B2B, store the answer as a field on the contact, carry it to the deal, and report closed-won amount by answer.

How do you triangulate referrals, surveys and branded search?

No single signal is right on its own. Use each one for what it can actually measure:

For branded search, use the branded queries filter in the Search Console Performance report, introduced in November 2025 and available to all eligible sites since March 11, 2026. Google classifies brand queries with “an internal, AI-assisted system”, including typos and unique products, warns that “some queries may occasionally be misidentified”, and offers the filter only for top-level properties with enough volume (Google Search Central Blog).

For visibility, Search Console's Generative AI report shows AI Overviews and AI Mode impressions, and Bing's AI Performance report shows Copilot citations. Neither reports visits: use them to explain trends, not to count revenue.

To size dark AI traffic, apply the survey's AI share to all orders and the average order value, then subtract the AI revenue analytics already attributes. This assumes respondents represent all buyers, so report it as an estimate, never as booked revenue.

What does this look like in practice?

Example (hypothetical; all numbers are illustrative): a B2B software company wants to know what AI search contributed last quarter.

1. Referral baseline. GA4 shows 14 demo requests from the AI Assistant channel, out of 200 in total.
2. Survey. The demo form asks “How did you hear about us?”. 150 of 200 people answer (75%); 33 choose AI search (22%): 20 ChatGPT, 8 Perplexity, 5 Gemini.
3. Match to deals. The answer is stored on the contact and copied to the deal. Of 40 closed-won deals worth €240,000, 9 carry an AI search answer, worth €54,000.
4. Size the gap. If non-respondents resemble respondents (an assumption), about 44 of 200 requests were AI-influenced (22% × 200), against 14 that analytics saw. Roughly 30 arrived as direct, organic or referral.
5. Sanity-check. Branded clicks in Search Console rose while non-branded stayed flat, and tracked prompts show the brand cited more often in ChatGPT. Both fit the survey; neither proves it.
6. Decide. The team reports 14 requests as the analytics floor, €54,000 as self-reported AI revenue and about 44 requests as an estimate.

What does this mean for you?

• In GA4's Traffic acquisition report, check the AI Assistant channel and look for chatgpt.com or perplexity.ai rows still in Referral or Unassigned.
• Add AI search as an answer option to your existing form or checkout, with a follow-up for the assistant.
• Attach the answer to the deal or order and report revenue by answer, not just leads.
• Track branded clicks in Search Console next to the survey's AI share.
• Report three numbers side by side: analytics floor, self-reported AI revenue, estimated gap.
• Re-check the setup quarterly; GA4 and Search Console changed several times in the past twelve months.

How do you measure this with AutoSEO?

AutoSEO's AI search attribution implements the survey-and-match method described above:

• Survey snippet. It detects an existing “How did you hear about us?” field in your forms (English or German) and records the answer on submit, or shows a short popup after form submits and purchases. Each visitor is asked once (localStorage, no cookies); emails are SHA-256 hashed in the browser where possible.
• AI search as a channel. Answers land in AI Search, Google / Bing, Social Media, Online Ads, Referral, Content or Other, with an optional “Which AI assistant?” step. Free text is normalized (“Google Gemini” is AI, “Google Ads” is ads), and you can reorder, relabel and hide options.
• Conversions and deals. Stripe (signed webhook; renewals are never matched), a Shopify Custom Pixel, WooCommerce and Shopware webhooks, and GA or Meta pixel events the snippet captures passively. HubSpot, Salesforce, Pipedrive, Attio and Close send lead source and deal value via workflow webhooks with field mapping; see attribution integrations and HubSpot.
• Matching. By order ID, then hashed email, then same-browser visitor ID, strictly 1:1, with a 90-day lookback and answers accepted up to 48 hours after the order.
• Reports. Deal value from AI search per channel and assistant, plus Hidden AI revenue (survey AI share × orders × average order value, minus AI revenue already in analytics), Survey response rate and Visibility ↔ AI leads, a correlation with your AI visibility tracking.
• Referral side. AI traffic analytics syncs AI-referred sessions from Google Analytics, Matomo or Piwik PRO by platform and landing page. Like any referral tool, it cannot separate AI Overviews or AI Mode clicks from organic search.
• For agents. The MCP server exposes get_attribution_summary and list_attributions; the REST API serves the same data.

AutoSEO is open source: self-host it for free or use the managed AutoSEO Cloud.