My Search / Visibility

GEO / Content · 06 min read · Updated Sep 2026

Original data for AI search: numbers a model cannot find elsewhere

Original data is citable because it has a source. A repeated industry statistic with no method is not original, even if you typeset it nicely.

Publish original numbers with a method and a date, in text, on a URL you are willing to keep.

Publish the number and the method together

State what you measured, over what period, and what you did not measure. ‘Of 40 Hong Kong service sites we crawled in August 2026, 11 canonicalised every URL to the homepage’ can be quoted. ‘Most sites do SEO wrong’ cannot.

Put the date in the sentence. Undated numbers get reused long after they expire, and then get blamed on you.

Make the table the source

A simple table with labels beats a chart that traps the figures in an image. Repeat the key figure in a sentence so a text-only fetch still sees it.

Link the data page from the guides that use it. Those internal citations tell a crawler which URL is the home of the number.

  • Method beside the figure.
  • A date in the sentence.
  • Text, not only an image of a chart.

Expect to maintain it

If you will not update or clearly retire the figure, do not publish it as a living benchmark. Mark one-off snapshots as snapshots.

When the number changes, update the same URL and say what changed. A new URL every year splits the citations.

Questions

Do AI systems prefer original statistics?

They prefer facts they can attribute. Original, specific figures with a method are easier to cite than anonymous ‘studies show’ lines.

How large does a study need to be?

Honest about its size. A crawl of 40 sites is useful if you say it was 40. Pretending it is a market census is how citations become misleading.

Continue reading

This note is part of the Hong Kong field library from My Search / Visibility, built with Itehk.