Skip to main content

How CiteCycle measures AI visibility

Which engines we ask, how questions are chosen, what mentioned, cited and position mean, how we separate real changes from noise, and what we can't see.

AI answers change from one run to the next. A tool that shows a single score without saying how it got there is asking you to trust it. This page sets out exactly what we do, what each number means, and what we can't see.

What we ask, and where

Every week we put your buyers' questions to:

  • ChatGPT, Gemini and Claude, through each engine's API with its own web search switched on
  • Google: the top 10 organic results and, when Google shows one, the AI Overview

Each question is sent word for word, with no system prompt, no chat history and no account. That is the closest repeatable stand-in for what a buyer sees. It isn't identical to the consumer apps, which can personalise answers for logged-in users.

Which questions

We track the questions your buyers ask, not keywords and not your brand name. At setup we read your homepage and suggest 15–25 unbranded questions across the buying journey, and you choose which ones to track. After that, new questions come from four places:

  • searches the engines run while answering that keep coming up and aren't tracked yet (for Claude, the searches it says it would run)
  • Search Console queries where you rank 11–20 with real impressions, rewritten as natural questions
  • full questions AI tools searched for on their way to your pages, which show up in Search Console as prompt-shaped queries (Questions → Asked by AI). These are tracked word for word
  • questions you add yourself

What the numbers mean

  • Mentioned: the answer text names your brand, or one of the aliases you set. This is an exact, case-insensitive word match, not an AI judgement, so you can check every hit yourself.
  • Cited: one of the answer's sources is a page on your domain or a subdomain.
  • Ghost citation: cited but not named. The engine used your content and recommended someone else.
  • Position: where your brand first appears, relative to the competitors you track. 1 means named first.
  • Rates: mentioned or cited answers divided by all answers, per engine and in total. Failed calls, and Google results pages without an AI Overview, are left out of the count rather than counted as misses.

There is no composite "visibility score". Every number opens up to the stored answer and its sources.

Each question gets one answer per engine per week, so any single week can move by chance. Before we call a change real, we run a two-proportion significance test at the 95% level, with at least 10 answers on each side. Smaller moves are shown as noise, not as wins or losses.

Before and after

When you mark an action as done (a page refreshed, a list joined), we compare the mentioned and cited rates for the questions linked to it, before and after that date. With Search Console connected, we add clicks and impressions for the pages you published and for searches that include your brand name.

This is correlation, and we label it that way. AI engines change on their own, so none of this is a controlled test.

Clicks, not impressions

Where we use Search Console, we report clicks. Google has confirmed that Search Console misreported impressions, average position and click-through rate from 13 May 2025 to 27 April 2026, and it will not restate that history. Clicks were not affected. A before-and-after comparison built on impressions across that window can show growth or collapse that never happened.

In our own study of 5,771 AI-assisted articles, impressions rose more than fourfold between late 2025 and spring 2026 and then fell by more than half, while clicks to the same articles moved far less. So we treat impressions as context, never as the result. Read the study.

Location and language

Google results and AI Overviews use your target country and language. AI engine calls carry no location, so local intent comes only from the wording of the question, for example "best dentist in Denver".

What we can't see

  • Answers personalised by a buyer's history, memory or account
  • Differences between the API models and the consumer apps
  • Engines we don't track: Perplexity, Copilot, Grok, Meta AI and Google AI Mode
  • AI referral traffic on its own: Google counts AI Overview clicks within normal Search Console totals

Free checker vs weekly tracking

The free checker sends 5 questions to each engine once, and counts your brand name or your domain as a mention. It is a snapshot. Weekly tracking matches your brand name and aliases only, and builds a trend over time.