How we measure AI citation
CitePack measures whether AI engines cite you by asking the engine your buyers’ question and checking whether your domain is among the sources it actually attributed. Three samples per question, reported per engine, with the date. Perplexity is asked natively. ChatGPT, Claude and Gemini are asked in web-grounded mode. For Google we ask Gemini as a proxy, and we do not measure AI Overviews. The free scan asks one of these engines, Perplexity. The other three come with the $49 report.
We publish this because a citation score is only worth trusting if you know exactly what it counts. Here is the whole method.
The measurement, step by step
We derive the questions your buyers actually ask
Starting from your URL and category, we form the real, natural-language queries a buyer would type into an AI engine — the questions where you'd want to be the cited source, not keyword strings.
We ask the real engines those questions
We send each query to the live engines — not a cached index or a simulation. Perplexity is queried natively; ChatGPT, Claude, and Gemini are queried in their web-grounded mode, where they retrieve and can attribute sources.
We check the actual cited sources for your domain
For each answer we read the engine's real cited sources and check whether your domain appears. "Cited" means your domain is one of the sources the engine attributed — a pure domain match, not a heuristic guess about whether you 'should' rank.
We sample across several runs
AI answers vary between runs, so a single check is noisy. We sample each query several times and take the majority result, so a one-off fluke doesn't read as a citation (or a miss). We report per engine, per question.
We measure the on-page signals — and prove the lift
Separately from the live probe, we score the seven GEO signals on your page. After you ship the Fix Pack, we re-score and show the before → after change deterministically — so you can see the on-page work landed, measured rather than estimated.
What each engine measurement means
Perplexity
Measured natively — it's retrieval-grounded by design and exposes its cited sources, so this is the most direct read.
ChatGPT
Measured in web-grounded mode, where answers retrieve and cite live pages.
Claude
Measured with web access enabled, where it retrieves and can attribute sources.
Google (Gemini)
We probe Gemini as a proxy for how Google's generative surfaces read a page. We do NOT measure Google AI Overviews directly, and we don't claim to.
Two distinct numbers, never conflated
CitePack reports two different things and keeps them separate on purpose. The on-page GEO score rates the citation signals on your page and is deterministic — the same page scores the same way. The live citation check asks the real engines your buyers’ questions and reads their actual cited sources. We never present an on-page score as a real citation, and we prove score changes before → after rather than estimating them.
The honesty line
No tool can promise a specific engine will cite a specific page — engines don’t publish their logic and their answers vary between runs. So we don’t promise a citation. We measure the signals that make citation likelier, prove your on-page score moved, and probe the live engines honestly, labelling Gemini as a proxy for Google’s generative surfaces rather than pretending to measure AI Overviews.
Frequently asked questions
What does 'cited' mean in a CitePack scan?
It means your domain appears in the engine's actual cited sources for that query — a direct domain match against the sources the engine attributed in its answer, not a heuristic estimate of whether you ought to rank. If your domain isn't among the cited sources, you're not cited for that question, even if you rank well in classic search.
Why sample each question several times?
AI answers are non-deterministic — the same question can return different sources on different runs. Sampling across several runs and taking the majority result gives a stable read, so a single lucky (or unlucky) run doesn't misrepresent whether you're actually cited.
Do you measure Google AI Overviews?
No. We probe Google's Gemini model as a proxy for how Google's generative surfaces read a page, and we label it 'Google (Gemini)'. We do not measure AI Overviews directly and don't claim to — that's an honesty line we hold in the product and in our reports.
Can CitePack guarantee I'll get cited after the fixes?
No honest tool can guarantee a specific engine will cite a specific page — engines don't publish their logic and their answers vary. What CitePack proves is that the on-page citation signals it adds moved your score before → after, measured deterministically. The live citation probe then tells you, honestly, whether the engines' cited sources change over time.
Is the on-page score the same thing as the live citation check?
No — they're two different measurements. The on-page GEO score rates the signals on your page (entity clarity, structured data, extractability, and so on) and is deterministic. The live citation check asks the real engines your buyers' questions and reads their actual cited sources. We keep them distinct so we never dress up an on-page score as a real citation.
Run it on your site
The free scan reads your page, tells you which of the seven signals need work, and asks Perplexity one question about your category. No signup. The other three engines and the three question families come with the $49 report.