Measuring AI visibility
Why this lesson
Section titled “Why this lesson”You can’t improve what you can’t see, and AI visibility is genuinely hard to see. Traditional SEO hands you Search Console with impressions, clicks and positions. AI search hides most of that — some platforms strip referral data, citations appear and vanish without notice, and much of the impact is a brand impression inside a conversation that never produces a click at all. This is the lesson that closes the loop: how to track what you can measure, how to combine those signals into one AI Visibility Score you can move month over month, and how to produce the before/after evidence that satisfies this level’s gate — get a brand cited by AI, and prove it.
The explainer
Section titled “The explainer”Pillar 1 — AI referral traffic (and why it lies)
Section titled “Pillar 1 — AI referral traffic (and why it lies)”The first thing to track is AI referral traffic: visits that arrive when someone clicks a link inside ChatGPT, Perplexity, Claude, Gemini or Copilot. In GA4 you isolate it by building a custom channel group — Admin → Data display → Channel groups — with a new “AI” channel whose source matches a regex covering chatgpt/openai, perplexity, gemini.google.com, copilot.microsoft.com, claude.ai and deepseek. Then read it under Reports → Acquisition → Traffic acquisition.
Treat the number as a floor, never the truth. Platforms handle referrers inconsistently: ChatGPT’s search-result links pass data, but in-content links on paid accounts use a no-referrer attribute; Perplexity tracks on web but not its desktop app; Copilot tracks on web but not Windows; Grok passes nothing. Stripped referrers land in GA4 as direct traffic, so what you see is an undercount. Use it for the trend — which engines are sending people — not a precise count.
Pillar 2 — bot activity, Pillar 3 — self-reported attribution
Section titled “Pillar 2 — bot activity, Pillar 3 — self-reported attribution”Second, track the AI bots themselves in your server logs or a Cloudflare bot report. Crawlers hit your pages far more than humans do, and the pages a citation bot (ChatGPT-User, OAI-SearchBot) fetches repeatedly are your strongest citation candidates; important pages bots never touch signal a discovery problem. Third — and Sam Oh calls this the most important — self-reported attribution: add a “How did you hear about us?” question with an “AI assistant / ChatGPT / Perplexity” option to your signup or checkout. It catches the conversion where someone heard your name from AI, then typed your URL directly (which analytics files as “direct”). At Ahrefs it surfaced ~3% of conversions that were otherwise invisible, converting well above organic.
The composite: one AI Visibility Score
Section titled “The composite: one AI Visibility Score”Traffic and citations are the outcome; you also want a leading-indicator score you control. The GEO framework rolls four measurable inputs into one AI Visibility Score (0–100):
AI_Visibility = Citability·0.35 + BrandMentions·0.30 + CrawlerAccess·0.25 + llms.txt·0.10
The weights encode the lessons of this whole level: citability (0.35 — can an AI quote your pages? lesson 3.2) matters most, then brand mentions (0.30 — the off-site trust of lesson 3.8), then crawler access (0.25 — can bots even reach you? lesson 3.3), and llms.txt last (0.10 — ship it, don’t lean on it, lesson 3.7). Compute it once as a baseline, ship your fixes, recompute. That single number, tracked over time, is your scorecard.
Citation tracking and the ongoing rhythm
Section titled “Citation tracking and the ongoing rhythm”Beyond your own analytics, you need citation tracking across engines — running a fixed prompt set (the 20–50 real questions your customers ask) through ChatGPT, Perplexity, Gemini, AI Overviews and Copilot and recording whether you’re mentioned, cited, and how you’re described. Tools like Ahrefs Brand Radar and the Semrush AI Toolkit do this at scale, reporting mentions, citations, impressions and share of voice versus competitors. Watch four things monthly, deep-dive quarterly: share of voice, new cited domains, topic coverage closed, and mention sentiment — because AI is vulnerable to misinformation (planted fake sources got repeated in up to 37% of Gemini/Perplexity answers in Ahrefs’ test), so fill every gap with specific, official, dated content.
The Level 3 gate
Section titled “The Level 3 gate”This is where you prove it. The capstone: run the full AI-visibility audit on a real site, ship the fixes (a citability rewrite, crawler access, schema/sameAs, one answer-target page, a brand-mention push), then demonstrate a new citation or mention in at least one engine — AI Overviews, ChatGPT, Perplexity, Gemini or Copilot — with before/after evidence. Screenshot the “before” (the engine not naming you for your prompt), do the work, screenshot the “after.” That before/after is the deliverable. Take it to the practice hub to log the gate.
- Build the GA4 AI channel group with the regex above; confirm AI referral sessions start appearing.
- Add a “How did you hear about us?” question with AI options to your signup or contact form.
- Write a prompt set of 15–25 real customer questions. Run each through two engines and record: mentioned? cited? described how?
- Score the four AI Visibility inputs (citability, brand mentions, crawler access, llms.txt) and compute your composite baseline.
- Screenshot one prompt where the engine doesn’t cite you — that’s your “before.”
- Log the baseline and the score in the Level 3 workbook; schedule a monthly recheck.
Terms introduced
Section titled “Terms introduced”Check yourself
Why is AI referral traffic in GA4 always an undercount?
In the composite AI Visibility Score, which component carries the most weight?
Which single measurement best connects AI visibility to actual revenue?
You can move on when you can… set up AI referral tracking in GA4, run a prompt set to check citations across engines, compute an AI Visibility Score, and produce before/after evidence that a brand earned a new AI citation — the Level 3 gate. Log it at /practice.
Go deeper
Section titled “Go deeper”- Ahrefs — Brand Radar and Semrush — AI Toolkit / AI Visibility Checker: the two mature ways to run a prompt set across engines and track share of voice without doing it by hand.
- Ahrefs — “The state of AI referral traffic”: the source of the ~0.25%-of-traffic-but-far-higher-conversion framing. Read it before deciding how much to invest.
- Next: 4.1 · How SEO gets sold: models & pricing — turn all of this into something you can package and sell.