Can you attribute leads and sales to AI search?

You can attribute leads and sales to AI search only where the visit identifies itself. That happens when an AI platform passes a referrer or appends a tag to the link. Everything else lands in direct, organic, or nowhere.

Three gaps cause this, as of October 2026:

• ChatGPT tags some links, not all. OpenAI states that ChatGPT automatically includes utm_source=chatgpt.com in referral URLs. The evidence shows free-tier and mobile-app clicks frequently arrive untagged anyway.
• Google AI Overviews send no referral. A click from an AI Overview cannot be separated from ordinary Google traffic in your analytics.
• Google's own AI report has no clicks. Search Console's generative AI report shows impressions for AI Overviews and AI Mode, with no clicks, no CTR and no position.

So the honest position is this. You can count a slice of AI-driven conversions with confidence. You cannot count the whole, and nobody outside the platforms can either. Report the slice as a floor, not a share.

Attribution needs an unbroken chain: an answer, a click, a recorded source, a session, a conversion. AI search breaks that chain at different points on each platform.

ChatGPT

The UTM tag is the best signal any AI platform provides. It is still incomplete. Untagged ChatGPT visits often show up as direct, which is why ChatGPT traffic shows up as direct in so many reports. Logged-in context adds noise too: OpenAI says ChatGPT may use saved memories when it rewrites a search query, so two users asking the same thing can get different sources.

Google AI Overviews and AI Mode

Google states that AI Overviews and AI Mode may use different models and techniques and return different links. Neither emits a distinct referral. The Search Console generative AI report gives impressions by page, country, date and device. It has no API, and the Search Analytics API does not cover this surface. A Google-side bug depressed impressions in that report from roughly 12 to 13 August 2026, so treat any trend that spans those dates with care.

Microsoft Copilot and Bing

Bing Webmaster Tools has an AI Performance report with citation counts and the grounding queries the AI actually ran. Microsoft calls Citation Share an observational metric, not a ranking or traffic share. It has no API as of this writing.

The citation problem underneath

Even a perfect click count misses influence. Research on how language models cite (arXiv:2606.28358) found citation output can be changed without changing the sources used. A buyer can read an answer built partly from your page, never click, and search your brand a week later. That conversion shows as branded search or direct.

You can still build a useful picture from four layers. Each one answers a different question.

1. Tagged and referred sessions. Filter sessions by utm_source=chatgpt.com and by known AI referrer domains. Follow those sessions to form fills, trials and purchases. This is your measured floor. Our guide to measuring AI referral traffic covers the setup, and GA4 added a native AI Assistant channel in May 2026.
2. Platform-side exposure. Use Search Console impressions for Google's AI surfaces and Bing's citation counts for Copilot. These measure exposure, not outcomes.
3. Self-reported source. Add a free-text "how did you hear about us" field to demo and contact forms. It catches buyers who read an AI answer and never clicked. It depends on memory, so label it as such.
4. Answer checks. Check whether AI platforms name you for the questions your buyers ask. This tells you whether you are in the answer at all, which explains why the other three layers may be empty.

None of these joins cleanly to the others. Keep them side by side, not summed.

Labels stop a partial number from being read as a complete one. Use one of three on every AI line in a report.

Two rules keep the labels honest. First, never add a measured and a self-reported number into one "AI-sourced revenue" total, because the same buyer can appear in both. Second, never present a share of trackable referrals as a share of all AI traffic. ChatGPT tags its links and Google AI Overviews do not, so ChatGPT will always look larger in referral data than it is. The page on whether ChatGPT really sends most AI referral traffic explains that distortion.

• Prompt trackers are samples. In Otterly's 2026 experiment, Microsoft's own report counted 42,817 citations for one test site, against 3,934 logged by Otterly's prompt-based tracker. The time windows may not match, but a first-party count sees more than any prompt sample.
• Citations without clicks. A page can be cited often and still send little traffic. See why AI cites a site but sends almost no traffic.
• Retrieval varies. ChatGPT search is used at the model's discretion, and it rewrites your query into its own searches. An answer with no search behind it cites nothing and refers nobody.
• Engines differ. In the same Otterly test, Copilot produced most of the citations while other engines barely cited the sites. A blended figure would have hidden that.

1. Set up AI referrer and UTM filters, then connect them to your conversion events.
2. Add a self-reported source field to your highest-intent forms.
3. Export Search Console's generative AI impressions and Bing's AI Performance data monthly, since neither has an API.
4. Label every AI number as measured, inferred or self-reported before it reaches a slide. Our playbook on reporting AI visibility to your CEO shows a format.
5. Check whether you appear in answers at all. The free checker on this site shows whether AI platforms name you for up to three buyer questions, which tells you if an empty referral report means low exposure or a tracking gap.

What you control is the instrumentation and the labels. What you don't control is whether a platform passes a referrer, so say that plainly in the report.