Why Google Analytics cannot measure AI visibility

When someone asks ChatGPT “who is the best SEO agency in London” and ChatGPT names three businesses, the user may:

  • Call one of them directly (no website visit)
  • Search the business name on Google and visit the site (appears as branded search in Analytics, not AI referral)
  • Visit the website from a ChatGPT link (appears as referral traffic, often misattributed)
  • Do nothing immediately but remember the name for later

Google Analytics tracks website visits. It cannot track the moment a user sees your business name in an AI answer and decides to act on it, especially when that action does not involve clicking through to your website.

This is the fundamental measurement gap. Traditional analytics tools were built for a click-based web. AI search produces answers, not clicks.

The four-part measurement framework

1. Citation rate

The percentage of relevant buyer queries where your business is named in the AI answer.

How to measure it: Define 12 buyer-intent queries that your target customers would ask an AI engine. Run each query across four engines: ChatGPT, Perplexity, Claude, and Google AI Overviews. Record whether your business is named in each response.

48 probes total (12 queries across 4 engines). Your citation rate is the number of citations divided by 48.

What good looks like: A business with no prior AEO work typically scores 5 to 15%. After one 90-day evidence cycle, 25 to 40% is a realistic target. The MOUVE case study achieved 60% across all four engines combined, with 92% on ChatGPT specifically.

2. Citation accuracy

Is the AI engine saying the right things about your business?

An inaccurate citation can be worse than no citation. If ChatGPT recommends your business but describes the wrong services, wrong location, or outdated pricing, the potential customer either contacts you with wrong expectations or dismisses the recommendation entirely.

How to measure it: For each citation, record what the engine says about your business. Compare against your actual services, location, pricing, and positioning. Flag inaccuracies.

How to fix inaccuracies: Update your website content, structured data, and llms.txt file with correct information. Use the corrections section in llms.txt to explicitly state what should not be said. Re-audit in 30 to 60 days.

3. Citation position

When your business is named, where in the answer does it appear?

AI engines typically name three to five businesses in a recommendation response. Being named first carries significantly more weight than being named last. Users read the first recommendation most carefully and often act on it without reading the rest.

How to measure it: For each citation, record your position: first named, second, third, or later. Track this over time. Moving from third-named to first-named for a key query is a meaningful improvement even if your raw citation rate does not change.

4. Citation sentiment

Is the engine recommending you positively, neutrally, or with caveats?

An AI engine might say “Business X is highly regarded for their thorough methodology” (positive) or “Business X offers these services, though their pricing is at the higher end” (qualified). Both are citations, but they produce different buyer responses.

How to measure it: For each citation, classify the sentiment as positive, neutral, or qualified. Qualified citations (those with caveats) may indicate that your messaging or third-party profile is creating mixed signals.

How to run a manual audit

You do not need expensive tools to measure AI visibility. A manual audit takes approximately two hours and produces a usable baseline.

Step 1: Define your queries. Write 12 queries your target customers would ask an AI engine. Mix service queries (“best [service] in [area]”), recommendation queries (“who should I hire for [task]”), and comparison queries (“[your service] vs [alternative]”).

Step 2: Run the queries. Open ChatGPT, Perplexity, Claude, and Google AI Overviews. Run each query and record the full response. Note which businesses are named, in which order, and what is said about each.

Step 3: Score. Calculate your citation rate, accuracy, position, and sentiment for each engine and overall.

Step 4: Identify gaps. Where are you missing? Is it specific engines, specific query types, or specific competitors displacing you?

Step 5: Baseline. Save the results. This is your starting point. Everything you do from here is measured against this baseline.

What a structured audit reveals that a spot-check misses

The most common mistake is checking one query on one engine and concluding “we are (or are not) visible in AI search”.

A structured audit reveals:

Engine variation. You might be cited on ChatGPT but absent from Perplexity, or visible on Google AI Overviews but not Claude. Each engine uses different signals and different crawling schedules. A single-engine check gives you one quarter of the picture.

Query variation. You might be cited for “best [service] in London” but not for “who should I hire for [service]”. Different query framings produce different results, and your customers use all of them.

Competitor patterns. A structured audit shows not just whether you are cited but who is cited instead. This reveals which competitors have stronger AI visibility and what they are doing differently.

How often to measure

Quarterly evidence cycles are the right cadence for most businesses. This gives enough time for optimisation work to take effect and provides comparable data points for trend analysis. The Qyliq methodology is built around this cadence.

Monthly spot-checks between quarterly audits can catch sudden changes (a competitor improving rapidly, an AI engine updating its model, or your citation dropping due to stale information). These are five-minute checks, not full audits.

A Visibility Briefing provides the initial structured audit with a prioritised action plan. Subsequent quarterly audits track progress and inform the next cycle of work.