A B2B buyer asks ChatGPT for vendor recommendations, uses Gemini to compare implementation approaches, turns to Perplexity to validate technical claims — and only then visits the companies that made the shortlist. Most of the decision happens before the session does. Green Hat's 2025 B2B Buyer Journey Research found up to 94% of buying-party members use LLMs during selection to validate, summarize and confirm. Wynter's 2026 research put CMO use of AI tools during vendor discovery at about 84%.
The reporting implication is simple: a stack that only measures what happens after the visit now sees half the journey. Rather than chasing perfect attribution, measure influence in layers across the journey. Each layer answers a different business question.
Before a brand can be recommended in an AI answer, AI systems have to find, crawl and understand the content. If LLMs cannot reliably reach your site — or do not consider the content worth retrieving — your products never enter the conversation.
That makes AI bot activity the earliest progress indicator. More verified AI bot visits do not guarantee visibility, but they do suggest AI systems are discovering and revisiting what you publish. Watch crawl frequency, crawl depth and crawl coverage.
Verification matters here more than with traditional crawlers, because user-agent strings can be spoofed. Validate with reverse DNS lookups, published IP ranges, or your CDN's verified bot service rather than trusting the user-agent alone. The point is not counting requests; it is establishing confidence that legitimate AI systems reach the content you are funding.
Once systems can access the content, the question becomes whether they use it. Measure mentions, citations and prompt responses — but inconsistent measurement turns this into noise. Running a handful of prompts after publishing is interesting and rarely directional.
Instead, build a standardized prompt library reflecting the questions buyers actually ask across the journey, and keep it stable. Stability is what lets you read trends instead of isolated wins. Mention rate, citation rate, Search Console impressions from AI Overviews and AI Mode, and Bing Webmaster Tools grounding queries each contribute a different angle. None tells the whole story alone; together they answer whether you are part of the conversation when buyers ask AI about your market.
This is where traditional attribution enters. GA4 captures AI assistant and AI referral traffic from recognizable LLM sources once a click lands, giving you direct-response outcomes much like organic or paid.
Be precise about the limits. AI Mode and AI Overviews are not included — they generally blend into Google organic, and can appear as Direct depending on how the click is passed. Those experiences influence discovery without being cleanly attributable. A buyer who found you through ChatGPT may never click the citation, then search your brand days later or return directly. None of that shows up as AI referral traffic. Treat revenue from AI assistant referrals as one component of the framework, not the definitive measure of AI's business impact.
Buyers commonly first encounter a brand through AI Mode, AI Overviews or ChatGPT, then return via branded search, direct, or email before converting. Analytics will not label those conversions AI-driven, but AI shaped the journey.
The practical read is branded clicks in Search Console alongside branded organic conversions in GA4. No single branded conversion proves AI created the opportunity. When AI visibility gains and branded demand rise together repeatedly, the evidence gets much stronger.
Executives fund AI search because they expect pipeline and revenue. Pipeline, closed-won and revenue remain the metrics that decide whether the strategy created value.
AI does not change the business outcomes you must measure. It changes the evidence you use to show progress. Instead of disconnected metrics, you can tell one coherent story: AI systems accessed the content, visibility inside AI experiences rose, branded demand grew, measurable AI-driven revenue increased, pipeline and revenue followed.
No single metric proves causation. Together they form a body of evidence that marketing is influencing behavior in channels traditional attribution cannot observe. For reading AI visibility against paid data, see AI Visibility Metrics Explain PPC Performance; for proving an improvement is real, see How to Design a Technical SEO Test That Actually Proves Something.
AI Mode and AI Overviews are not classified as AI assistant or AI referral traffic. They generally blend into Google organic and can land as Direct depending on how the click is passed, so GA4 alone cannot isolate their contribution. Pair it with Search Console impressions and branded demand.
Collect the questions real buyers ask at each journey stage and freeze them into a fixed set. Stability is the point — only a stable set lets you read mention and citation rate as a trend rather than a series of one-off checks.
Do not trust the user-agent, which can be spoofed. Use reverse DNS lookups, vendor-published IP ranges, or your CDN's verified bot service. The goal is confidence that real AI systems are reaching your content, not a request count.
No. Layers 1 and 2 alone already improve most teams' decisions. But executive reporting needs layers 4 and 5, so wire branded search clicks and pipeline into the same report early rather than retrofitting later.
To apply what you just read to your own site, start with a free audit of where things are now.
A strategist replies within 24 hours on business days.
An E-E-A-T checker was never automatable because there is no API for it. Here is how an AI coding agent read Google's quality guidance, crawled a page sample, and produced a scored audit — plus where it stops.
A GEO audit is not checking whether your brand appears in an AI answer. Here is a nine-point GEO checklist split into four on-site items, four off-site, and one measurement layer.
LinkedIn, YouTube, Reddit, and TikTok are building systems to identify low-value AI content. Here are the limits of watermarking and what content still gets distributed.