JSON-LD and citations: 36.0% vs 10.5%, and why I will not call it a cause
Pages carrying JSON-LD were cited in 36.0% of runs. Pages without were cited in 10.5%. That is the strongest page-level signal in our dataset, and it is still not a cause.
The comparison
- With JSON-LD: 59/164 runs cited, 36.0%.
- Without: 11/105, 10.5%.
- 19 of the 48 pages had no structured data at all.
Why the two conditions are not the same experiment
We ran the same five queries under two conditions. In condition A the model saw only the title and the snippet of each source. In condition B it also saw the page body. The JSON-LD gap is almost entirely a condition-B phenomenon — when the model only reads a title, whether the page carries markup is invisible to it, so the two groups look the same.
That is the single most useful thing in this dataset, and it is not about JSON-LD at all. It says the page-level features only start to matter once the page is actually read. Which means the first question is never “what is on my page” but “is my page being read”.
Every tool in this category would print that comparison as a headline and stop there. The next three paragraphs are the reason we do not.
How strong is it, honestly
We looked at 15 page-level features against citation. 9 of them moved in the positive direction once the model could read the page body. The strongest is JSON-LD at a rank correlation of 0.391. At 48 sources the two-tailed threshold for “not distinguishable from noise” is 0.285 — so exactly one feature clears it, and it clears it by a modest margin.
Retrieval-crawler access is not in that table — it comes from the robots record rather than from the page, so we compute it separately: 0.162 on 43 sources, against a threshold of 0.301. It does not clear. That is lower than the rate comparison suggests, because access and citation are both influenced by how well-known a site already is. That is confounding, and it is why we print the correlation next to the rate instead of only the rate.
What would change my mind
A controlled test. Take pages that are otherwise comparable, add JSON-LD to half of them, and run the same queries again. If the gap survives, it is a cause. Nobody in this category has published that test — including us. Until someone does, “JSON-LD raises citations” is a hypothesis with a plausible mechanism and one correlational observation behind it.
So what should you do
Add it anyway — it is cheap, it is not harmful, and the mechanism is real: structured data is the easiest thing for a machine to lift verbatim. Just do not buy a dashboard because someone quoted a percentage at you. The percentage is a description of what we saw, not a promise about what you will get.
Questions this piece raises
Does JSON-LD increase citations?
We do not know, and we will not say it does. Pages with it were cited more often in our runs, and the effect only appears when the model reads the page body. That is a hypothesis worth testing, not a lever we can sell.
Why is the sample only 48 pages?
Because each additional source costs real retrieval runs, and we would rather publish a small honest number than a large invented one. The thresholds on this page are computed for this sample size, and they would move if the sample grew.
Which JSON-LD types did the cited pages use?
Mostly Organization, WebSite and FAQPage. We did not test whether the type matters — with this sample size, splitting by type would leave too few pages per cell to say anything.
Check your own page
Enter one URL and get the eight checks, including the ones that fail. Free, no signup, no score.