Publish numbers that people can check

A number without a source is just a claim. A number with a source, a date and a method is something a reader can use and something another writer can safely cite. If you have real figures about your business or your market, publish them in a way that others can check.

The short answer

For every statistic on your site, show four things next to it: what exactly was counted, when, how, and where the data came from. Put your best figures in one table on one page. Drop numbers you cannot back up.

What counts as a number worth publishing

Good numbers describe something real and specific:

• How many orders you shipped last year, and how many arrived late.
• The median time to reply to a support message, over a stated month.
• The average price of a job, across a stated set of jobs.
• Results of a survey you ran, with the number of respondents.

Vanity numbers sound large and say little. "Thousands of happy customers", "trusted by leading brands", "number one in quality". No reader can check these, and a careful reader will discount them. Google's advice on helpful content asks whether a page offers original information or analysis, and whether it presents information with clear sourcing. A vague boast does neither.

Step 1: write the claim precisely

Turn a loose claim into a sentence with a unit, a period and a population.

Loose: "We reply fast."

Precise: "In September 2026, the median first reply to an email from a customer was 3 hours 10 minutes, across 412 emails (Example Studio support inbox, working days only)."

The numbers above belong to a made-up business. Use your own real ones.

Step 2: record the source and the method

Each statistic needs:

1. What was measured. The exact thing, with its unit.
2. Period. Start and end dates.
3. Population. All orders? Only paid ones? Only in one country?
4. Method. How you counted. A report from your system, a survey form, a manual count.
5. Source. Where the data came from. If it is another organisation's study, link to it and read it first.
6. Date published, and date last checked.
7. Known gaps. What was left out and why.

If a figure comes from someone else, cite the original study, not a news article about it. If you cannot find the original, do not use the number.

Step 3: build one table people can cite

Put your key figures in a single table on a stable page, for example example.com/numbers. A table is easy to read and easy to copy accurately.

| Measure | Value | Period | Method | Source |
| --- | --- | --- | --- | --- |
| Orders shipped | 5,230 | Jan-Dec 2025 | Count of orders marked shipped | Our order system export |
| Late deliveries | 214 of 5,230 | Jan-Dec 2025 | Delivered after the promised date | Our order system and carrier tracking |
| Median first reply | 3 h 10 min | Sep 2026 | Time from email in to first human reply, working days | Support inbox export |

Below the table, add a short "Notes and limits" section and the date of the last update. Keep older versions of the table with their dates, rather than overwriting them silently. A reader who sees last year's numbers and this year's side by side can trust both.

Optional: mark it up as a dataset

If your table is a real dataset, such as a downloadable file, Google describes Dataset markup that helps people find it in Dataset Search. Its required fields are a name and a description, and a content URL for any download. Recommended fields include a licence, a creator and the period covered. Google's general structured data rules also say not to mark up content that readers cannot see. This is optional and is for datasets, not for ordinary pages with a few statistics.

A licence matters if you want others to reuse the figures. Say what they may do, for example "You may quote these figures with a link to this page."

Dates and updates

Show when the page was last updated, and keep it honest. Google's guidance on dates says to show a clear, visible date, keep any date in markup consistent with it, and never use a future date. Change the date only when a figure changes.

Avoid these traps

• Rounding up. "Nearly 10,000" when the count is 8,300.
• Mixing periods. A year's total next to a month's average with no label.
• Choosing the best month and calling it typical.
• Counting something different from what the sentence claims.
• Quoting an old study as if it were current. State the year.
• Showing a percentage with no base number. "40 per cent" of 5 people is not the same as of 5,000.

How to check it worked

1. Give the page to a colleague. Ask them to recompute two numbers from the stated method. If they cannot, add detail.
2. Read each sentence and ask: could a stranger find out if this is true?
3. Check that every external figure links to the original and that you opened the link.
4. Ask an AI assistant questions on your topic a few times. See if your figures are used or cited, and whether it states them correctly. Our free check shows which sources assistants cite for buyer questions.

What this won't do

• It does not make an assistant cite you. Assistants and search engines choose their own sources.
• A neat table does not improve weak data. If the sample is small, say so, or do not publish it.
• It does not replace a proper study. A count from your own system describes your business, not the whole market.
• Markup is not required, and it does not guarantee any special display.