Thought leadership a machine can count: publishing your own numbers, with the method attached

What a machine can count, and what it cannot

"We are the most trusted roofer in the region" is an opinion, and nothing reading it can check it. "Of the [number] roof inspections we carried out between [month] and [month], [number] found storm damage to the flashing" is a fact: it has a count, a period, a subject and an owner. A passage like that can be lifted into an answer and still be true and still be yours, which is what this site's glossary calls citability.

The one controlled study we cite points the same way. Aggarwal et al. (KDD 2024) found on their benchmark that adding statistics, quotations and citations to a source raised its visibility in generated answers by up to about 40%. The paper measured sources in general, not businesses, so read it as the mechanism: specific, attributed numbers are the kind of text answers are built from.

Numbers a business already has

Most of them are in the job records, the booking system or the accounts, waiting to be counted. The test for each is the same: could you show the count to someone who doubted it?

• What you found: the causes behind the jobs you did, counted over a season or a year.
• How long things take: the median time from call to visit, or from order to delivery, measured from your own timestamps.
• When demand comes: the months your inquiries rise and fall, from two years of records.
• What things cost: the range your customers actually paid for a common job, with what moved it.
• What your panel found: if you measure how assistants answer your buyers' questions, what the answers said, over months.

The last one is how this site began. Our audit of 9,483 small business websites is a count of what a crawler received from each site, published with its method and its limits, and it is the only source of the figures this site quotes about small business websites.

Publish a number so it survives being quoted

A figure travels without its context unless the context is in the same sentence. Put the conditions beside the number, not in a footnote, so whoever quotes one quotes the other.

1. State the count and the base together: [number] of [number] jobs, not a bare percentage.
2. State the period: the first and last dates the figure covers.
3. State who is counted and who is not, in the same sentence or the next.
4. Say how it was counted and from what records, in a short methods note on the same page.
5. Name the author and date the page, and change the date only when the figures change.
6. Put the table in the page as HTML, and offer the rows as a download if there are many.
7. Keep the address stable, so a citation made today still leads to the figure next year.

Then say what the number does not show. A count of your own jobs describes your customers, not the market; a season's figures describe that season. Saying so is not modesty; it keeps the figure from being repeated as something it is not.

Dataset markup, when the numbers are a dataset

A table on an article page needs no special markup; the article's own markup and the table's text are enough. When the numbers are a dataset in their own right, a body of records with a method and a download, schema.org's Dataset type describes them, with properties such as creator, variableMeasured, measurementTechnique and temporalCoverage. Google's dataset documentation says that markup with a name, a description, a creator and a distribution helps a dataset appear in its Dataset Search.

{
  "@context": "https://schema.org",
  "@type": "Dataset",
  "name": "[What was counted, in plain words]",
  "description": "[What the records are, how many, over what period, and what they do not cover]",
  "creator": { "@id": "https://www.example.com/#business" },
  "temporalCoverage": "[YYYY-MM/YYYY-MM]",
  "measurementTechnique": "[How it was counted, and from which records]",
  "variableMeasured": "[The thing each row measures]",
  "isAccessibleForFree": true,
  "distribution": {
    "@type": "DataDownload",
    "encodingFormat": "text/csv",
    "contentUrl": "https://www.example.com/data/[file].csv"
  }
}

Every value in it has to be on the page, and the download has to contain what the description says. The study behind this site's figures carries a Dataset node of the same shape, with its method linked from measurementTechnique.

What not to publish

• A number you cannot show the records for. If a customer or a journalist asked, you should be able to point at the rows.
• A percentage without its base. "Most of our customers" means one thing out of five customers and another out of five thousand, and a percentage hides which.
• A survey of a handful of people presented as a finding about a market.
• Someone else's statistic without its primary source. Quote only the original, attributed and linked, and never a figure that has passed through three blogs.
• A comparison with competitors you have not measured. "Faster than anyone in town" is a claim; your own median response time is a fact.

Getting the figures in front of people who cite

A figure nobody reads is a figure nobody quotes. Send it to the people who write about your trade in your area: the local paper, your trade association's newsletter, the forum your buyers use, with the page it lives on. A mention of your figure on someone else's page is the kind of evidence you cannot write yourself, which is the subject of mentions and digital PR. Then watch your panel's cited sources for the page, and for other pages quoting it.