How to measure your AI citation rate
How to measure whether AI engines name and cite your pages, set a baseline across four engines, and re-prove it over time as content and engines drift.
A screenshot of ChatGPT naming your brand once is not a citation rate. It is a single answer, on a single engine, on a single run, and the same prompt tomorrow might name someone else entirely.
People show that screenshot around like proof. It proves nothing. A rate is a pattern across prompts, engines, and time, and a pattern is the only thing worth trusting.
What this guide does
It shows you how to measure whether AI engines actually name and cite your pages, set a baseline across four engines, and re-prove it as things move. The aim is a stable signal you can act on, not a lucky answer you can brag about.
What a citation rate actually is
It is how often engines name or cite your page when asked the questions that matter to your business, read across engines and across prompts.
Two words carry the weight. Across engines, because ChatGPT, Perplexity, Gemini, and Claude read and answer differently, so one engine is not the story. Across prompts, because a single question is noise and a set of real questions is signal. The rate is the share of those that surface you, not one screenshot.
Step one, pick prompts that match real intent
A rate is only as meaningful as the questions behind it. Measure against prompts a real buyer would ask, in their words.
If you do emergency HVAC, the prompt is "who fixes a furnace at night near me," not "premier HVAC solutions." Vanity prompts you would never be asked inflate a number that means nothing. Honest prompts give you a rate you can act on.
Step two, check the four engines
Run those prompts across ChatGPT, Perplexity, Gemini, and Claude. Each one reads pages differently and weights different signals.
A page named on one engine and ignored on three is telling you something a single-engine check would hide: the page is readable to some engines and not others, which is a fixable structural gap, not bad luck.
Step three, set a baseline
Write down where you stand now. Which engines name you, on which prompts, which pages they cite. That is your baseline.
Without it, every later number floats. With it, you can say whether you moved, on which engine, after which change. The baseline is what makes the rest of the loop mean anything.
Step four, re-prove over time
Here is the part that makes a rate honest: you re-prove it. Engines update how they parse. You publish and edit. Competitors tighten their pages. A rate measured once and quoted for six months is fiction.
So re-run the same prompts on a schedule and watch the rate move. Citations are measured, never promised, and a measurement you do not repeat is just a memory. The re-prove habit mirrors keeping schema valid after edits: prove, watch, re-prove.
Old way versus new way
The old way was the screenshot: ask an engine once, catch a good answer, call it evidence. One run, one engine, frozen in time.
The new way is a rate: many prompts, four engines, a baseline, and a re-prove on a schedule. The screenshot tells you what happened once. The rate tells you what is true now, and lets you see when that stops being true. You can read the full case for this in measuring vs fixing AI citation tools.
The damaging admission
Measuring your citation rate does not raise it. We will not pretend it does. Measurement is a thermometer, not a cure. It tells you the temperature honestly and changes nothing on its own.
And no one can guarantee the rate will climb, including us. Engines are probabilistic and they shift on their own schedule. What measurement gives you is the truth about where you stand and whether your changes moved it. The rate is proof of readiness, never a promise of a citation. Anyone selling the promise is selling something.
There is also a fit limit. If you do not depend on being found and recommended, tracking a citation rate across four engines every week is more discipline than the situation deserves, and we will say so.
How the watch works
On WordPress, with the Citedon plugin connected, the watch re-proves your readiness as content and engines drift, and lets you apply a fix after you approve it when the rate slips for a readable reason. Watching and fixing are the paid part. The first scan is free.
On other platforms, the scan still measures and re-proves on demand, and you apply any structural fix by hand.
See where you stand today
You cannot improve a rate you have never measured, and a single screenshot is not a measurement. Get a real baseline.
Run a free scan to see whether ChatGPT, Perplexity, Gemini, and Claude can read and name your page right now. Then keep the underlying markup honest with how to keep schema valid after edits.