One Source, Three Numbers: Check the Denominator
An AI citation statistic means little until you know what it was divided by: all citations, the top ten sources, or all answers. In SOLEDAD’s ChatGPT data, Reddit is 1.23% of cited links, 6.4% of the top ten domains’ links and present in 14.7% of answers, all from the same 155,520 links.
Mercy Tahitoe, Studio Head, White Wood, Jakarta.
1. Three ways to divide the same count
Each figure is correct; they answer three different questions. SOLEDAD, White Wood’s AI-monitoring engine, logged 155,520 source links in 6,510 ChatGPT answers across 66 brand projects from 18 May to 14 October 2026. Reddit accounts for 1,917 of those links, in 957 answers. Here is what changes with the denominator:
| Denominator | The question it answers | Reddit on ChatGPT |
|---|---|---|
| All cited links (155,520) | What share of everything ChatGPT cited was Reddit? | 1.23% |
| Links to the top ten domains (29,774) | How big is Reddit among ChatGPT’s main sources? | 6.4% |
| All answers (6,510) | How often does a ChatGPT answer include Reddit at all? | 14.7% |
The first is a share of citations. The second only divides among the leaders, so it is always larger. The third is a share of answers.
2. Why the answer share runs so much higher
One link is enough to put an answer in the count, and ChatGPT answers carry many links. In this pool a ChatGPT answer cites about 24 links on average. If one of them is Reddit, the whole answer counts as “includes Reddit”. That is how a source can be a small slice of citations and still appear in one answer out of seven.
Neither number is wrong. They suit different decisions. If you want to know whether buyers will see a source next to your category, the share of answers is closer. If you want to know how much of the evidence an engine leans on comes from that source, the share of citations is closer.
3. The engine changes the number too
Before comparing two statistics, check that they describe the same engine. In the same pool, Reddit appears in 14.7% of ChatGPT answers, 4.7% of Perplexity answers, 3.0% of AI Mode answers, 2.3% of AI Overview answers and 1.0% of Gemini answers, and in no Copilot answer. Across all six engines it is 0.51% of cited links, twentieth among domains. A figure quoted for “AI” without naming the engine can mean any of these.
4. Five questions for any statistic
Ask these before a number goes into your plan or your board deck:
- What was divided by what: citations, top sources, or answers?
- Which engine, or which mix of engines?
- Which language and market were the prompts in?
- Over which dates?
- Whose prompts: a vendor’s keyword list, or prompts your buyers actually ask?
We apply the same questions to our own reports. A share that moves from one month to the next may only mean the prompt set or the engine mix changed.
What this check cannot tell you
Our pool leans toward the categories our clients work in, and the prompts are in Bahasa Indonesia and English, so other markets may look different. SOLEDAD sees links shown in answers, not clicks. The “top ten” figure depends on which domains made the top ten in this pool. None of these numbers says whether being cited sends visitors or sales.
Declared interest: White Wood sells this work and measures it with SOLEDAD, our own AI-monitoring engine.
FAQ
Which denominator should I use?
Use the share of answers to judge exposure and the share of citations to judge weight. Report the one you use next to the number.
Why do two reports on the same source disagree?
Usually the denominator, the engine mix, the dates or the prompt set differ. Check all four before deciding one report is wrong.
Does Reddit matter for AI answers?
It depends on the engine. In our pool Reddit appears in 14.7% of ChatGPT answers and in no Copilot answer.
Sources
- SOLEDAD, White Wood’s AI-monitoring engine: cited links per answer, 66 brand projects on six engines, 18 May to 14 October 2026 (re-run 29 October 2026)