TL;DR: Growth and SEO leads at B2B software companies can watch their AI Search mentions climb, yet few can tie those mentions to qualified demos. AI Search attribution closes much of that gap by separating four signals: answers that mention you, agents that crawl you, people who click through, and the actions those people take on your site. Standard web analytics file many AI referrals under Direct, and no tool, Limy included, can split Google's AI Overviews clicks out of regular organic search. A dedicated AI traffic pixel, CDN-level agent data, conversion events with real values, and one survey field give your team numbers you can defend in the next budget review.

Your CEO forwarded a screenshot of ChatGPT recommending your product, and now the board wants to know what AI Search is worth. You open your analytics and find a few dozen chatgpt.com sessions next to a swollen direct bucket, with no line from either one to the pipeline.

The gap is fixable, even if part of it stays dark. In this guide we walk through how to measure AI Search traffic, which conversions to report, how one customer tracked the full chain, and where attribution honestly runs out.

In this guide

• What is AI Search attribution

• How to tell mentions, agent visits, referrals, and conversions apart

• How to measure AI Search traffic in five steps

• Which AI Search conversions and metrics to report

• What the Owl Labs example shows about measurement

• What AI Search attribution can and cannot prove

• See your AI Search traffic and pipeline in one place

What is AI Search attribution

AI Search attribution connects what AI engines say about your brand to the visits and pipeline that follow. For a demand gen manager at a B2B software company, it answers one budget question: did ChatGPT, Perplexity, Claude, Gemini, or Google's AI features send buyers who later booked a demo.

The difficulty comes from how AI answers work. A buyer asks a long question in a private chat, reads a summary, and may never click a link.

When they do click, some AI apps strip the referrer, so your analytics files the visit under Direct. And classic attribution assumes a trail of clicks and UTM tags, while AI Search leaves a thinner trail that partly lives in server logs and in a buyer's memory of where they heard about you.

So AI Search attribution works more like detective work than a single report. You collect the signals you can observe directly, label the ones you infer, and show leadership both.

Teams that keep the two apart earn trust in the room, because they never claim a number the data can't support. The first step is knowing which signal you're looking at.

How to tell mentions, agent visits, referrals, and conversions apart

Before you build a report, separate four signals that dashboards often blend together. Each one lives in a different system and answers a different question.

• AI mention. An AI answers names or cites your brand for a tracked prompt. Prompt tracking tools sample these answers, so a mention tells you about exposure and nothing about a visit.

• Agent visit. A crawler such as GPTBot, OAI-SearchBot, ClaudeBot, or PerplexityBot fetches a page. Agents skip JavaScript, so only server or CDN logs record them, which is why Limy's Agent Monitor reads CDN request data instead of a page tag.

• Human referral. A person clicks a link inside an AI answer and lands on your site. Most analytics tools only label it as AI traffic when the referrer or a utm_source=chatgpt.com tag survives the trip. Limy's AI Traffic Monitor pixel focuses on these visits and records what each visitor does next. 

• Conversion. The visitor books a demo or reaches checkout on your site. Limy Attribution counts that action against the AI session through events you configure, and conversions that happen later or offline live in your CRM.
  
  
  

User-triggered fetchers sit in a gray zone. OpenAI's ChatGPT-User and Perplexity-User fetch a page while someone waits for an answer, so they signal live demand, yet they still belong in your agent column.

One more trap catches many SEO leads. Treat any Google-Extended hits in your logs with suspicion, because Google uses it as a robots.txt control token with no user agent of its own, as Search Engine Roundtable reported.
Keep the four signals in separate columns, and your measurement setup gets much easier to build.

How to measure AI Search traffic in five steps

You can build a workable setup two ways: stitch it together yourself, or use a platform built for AI traffic. The steps below follow Limy's setup, and the table after them compares both routes.

1. Install the AI Traffic Monitor. Add the Limy pixel through Google Tag Manager. It identifies visits from AI engines and records page views plus the events you configure, so AI sessions no longer hide inside your general traffic.

2. Connect your CDN for Agent Monitor. Limy reads request logs from Cloudflare, Akamai, Fastly, CloudFront, Vercel, Netlify, and others to show which AI agents fetch which pages. A WordPress plugin covers sites without a CDN connection. Agent Monitor adds a crawl health score, flags failing requests, and links each crawled page to the prompts it relates to.

3. Define your conversion events. Pick the CTAs that matter, such as a demo form or a checkout step, and give each one a monetary value. On ecommerce sites, Limy reads checkout and purchase events straight from your data layer, so there's nothing to set up. For lead-gen CTAs such as demo or trial forms, our team configures Attribution with you. Either way, you see which events come from AI sessions.

4. Tag your prompts as the set grows. Add new prompts whenever buyers start asking new questions in your category, and tag each one by topic or by the date you added it. Filtering by tag lets you compare visibility and conversion trends on the same set of questions, while newer prompts build their own baseline.

5. Ask buyers directly. Add a "How did you first hear about us?" field to your demo form. Self-reported answers catch the buyer who read a ChatGPT answer on their phone and typed your URL a week later. Save the answer to a CRM field and review it monthly next to the AI sessions Limy recorded. Deals that name an AI tool but never showed an AI session are demand your analytics be missed: report them as AI-influenced pipeline, and use the gap to show leadership how much AI traffic your tracked numbers undercount.
   
   
   

Google's AI Overviews and AI Mode stay a blind spot for every tool, because Google counts those clicks as organic search. Limy's prompt tracking still shows where you appear on both surfaces, and Search Console's Search Generative AI report adds impressions by page, with no clicks.

Once these five pieces are in place, you have the raw material for a report leadership can read in two minutes.

The two routes compare like this:

Which AI Search conversions and metrics to report

Leadership wants one page, so group your metrics by layer and say plainly what each one can't show. The table below works as a template for a monthly AI Search report.

Two habits make the report hold up under questioning. First, compare AI-referred conversion rates with your site average for the same page types, since a demo page converts differently from a blog post.

Second, track branded search and direct traffic to decision pages next to your visibility trend. In Similarweb's June 2026 study of US desktop traffic, brands ChatGPT recommended were 2.5x more likely to get a visit within seven days, and 55.9% of that traffic came through branded search. No referral report credits those buyers, so the proxy line matters.

Conversion quality deserves its own row too. Adobe reported that AI traffic to US retail sites converted 42% better than non-AI traffic in March 2026. B2B buying cycles run longer, so set your own baseline before you quote anyone else's.

What the Owl Labs example shows about measurement

Owl Labs, which sells video conferencing hardware, gives a useful look at the measurement chain in practice. During a 30-day initiative, the team put 10 Limy recommendations live, six of them on the two topics it cared about most.

According to the Owl Labs case study, AI-attributed activity over that period moved like this:

• Live demo requests up 73%

• Checkout events up 143%

• Add-to-cart events up 33%

• AI visibility up 10%, with citations up 26%

Read those numbers with their definitions attached. A checkout event marks a step in the checkout flow, so it does not equal a completed purchase or revenue. Demo requests count live demo form submissions, before any sales qualification.

The 10% visibility gain describes a relative change in the underlying rate.

The case study also leaves out raw event counts and the attribution window, so a large percentage could rest on a modest base. And with no control group, seasonality and ChatGPT's May 2026 link changes could account for part of the lift.

Treat Owl Labs as a method you can copy, and set your own expectations for results. Owl Labs sells hardware online, while most SaaS buyers move through longer cycles with more stakeholders.

The part worth borrowing is the structure: fixed topics, a fixed window, a prior-period comparison, and on-site events tied to AI sessions.

What AI Search attribution can and cannot prove

Honest limits make your report stronger, because finance teams trust numbers that come with boundaries. AI Search attribution can show you:

• Which sessions came from a known AI referrer and then fired a key event

• How a metric trends against a stated baseline

• Which pages AI crawlers reach, and how often

• Whether visibility on tracked prompts moves alongside on-site outcomes
  
  
  

It cannot show what buyers typed into a private chat, and no tool, ours included, sees those prompts. It also cannot prove causation without a holdout, and it misses buyers who read an answer and return days later through branded search or a bookmark.

Long B2B cycles add another layer. A buyer might read a ChatGPT answer, browse a review site, sit through a webinar, and take two sales calls before an opportunity opens, so single-touch credit will either flatter or shortchange AI Search.

So report AI-sourced and AI-influenced pipeline as separate lines, and publish your lookback window next to them. Label any probabilistic match as an estimate, then check your analytics and self-reported answers against your visibility trend.

Do that, and you'll hand your CFO a number that holds up when they poke at it.

See your AI Search traffic and pipeline in one place

Limy brings the four signals into one workflow, so your team spends less time stitching exports together. A lightweight pixel tracks human sessions from ChatGPT, Gemini, Perplexity, and other engines in Limy Analytics, along with landing pages, devices, and countries. A CDN integration records which AI bots and agents hit which pages.

On the conversion side, you assign events to any CTA on your site, such as a demo form or checkout button, and give each one a monetary value. Limy Attribution then connects AI sessions to those actions, while visibility tracking shows the prompts and topics where you appear against your category. The Attribution panel lists each event with its count, revenue, and change, a revenue overview totals the value of converting actions, and a toggle compares AI traffic with all traffic.Annotations to add: the provider split across ChatGPT, Gemini, and Perplexity, total revenue and converting actions, the Attribution panel with its AI traffic toggle, and the Agent Monitor bot ranking.

We can't see private prompts, and we don't attribute every conversion. What you get is a clear split between what Limy observes and what it estimates, in a format you can take to your next budget review.

Your next step takes about 20 minutes. Check how much of last month's traffic landed in Direct, then count how many recent demo requests named ChatGPT or another AI tool in the survey field.

When you're ready to see those signals side by side for your own site, book a Limy attribution demo.