Original Data and Statistics

Original data is information only you can publish: numbers from your product, customers, research or analysis. Presented with a clear method and a quotable sentence, it gives writers, journalists and AI engines a reason to cite you. Most content repeats what's already online. A credible, specific statistic stands out because no one else has it.

In this lesson

• Why original data earns citations
• Where to find data you already have
• Methods that make numbers trustworthy
• Writing statistics people can quote
• Exercise and quiz

Why does original data matter for GEO?

The GEO research paper tested several ways of changing content and measured visibility in generative engine responses. Adding statistics was among the methods that improved visibility in its experiments, alongside adding quotations and citing sources. That's one study with its own setup, so treat it as evidence, not a guarantee.

There's also a simpler reason. When AI assistants answer questions with numbers, the numbers come from somewhere. If your page is the original source of a useful figure, other pages link to it and AI answers may cite it.

Where can you find original data?

Always check privacy, consent and contractual limits before using customer data. Publish only aggregated figures.

What makes data trustworthy?

1. State the method: what you measured, how, when and the sample size.
2. Report uncertainty: with small samples, give ranges, such as a confidence interval (NIST).
3. Avoid cherry-picking: publish the result even if it's less exciting than you hoped.
4. Separate fact from interpretation: "42% said X" is a fact; "so X is a trend" is your view.
5. Make it reproducible where you can, by sharing questions or a dataset.

How do you write a quotable statistic?

A quotable statistic fits in one sentence and includes:

• The number
• What it measures
• Who or what it covers
• When
• The source, linked

Weak (illustrative): "Lots of our users love automation."

Strong (illustrative): "In a survey of 412 ExampleCo customers in September 2026, 38% said they use automated invoicing every week (method)."

Place that sentence near the top of the page, repeat it in the relevant section, and add a short "How we got this number" note.

How do you get the data seen?

• Publish it on a dedicated, permanent URL
• Include a chart, with the numbers also in text
• Share it with journalists and newsletter writers in your niche
• Update it annually, keeping the same URL and noting the date

What should you avoid when publishing data?

• Leading survey questions that push respondents toward the answer you want
• Tiny samples presented as trends: 30 responses can't support a broad claim
• Mixing definitions: if "active user" means different things in different years, the comparison is meaningless
• Hiding the denominator: "500 customers saw results" means little without knowing how many were measured
• Over-precise numbers: reporting 37.42% from a sample of 80 implies false accuracy; round sensibly

A good test is to imagine a sceptical journalist reading the method. If they'd find an obvious hole, fix it before you publish.

A quick example

Illustrative. An HR platform analyses anonymised, aggregated scheduling data and finds the median time from job offer to start date in its customers' companies. It publishes the figure with the sample size, period and method, plus a table by industry. Trade publications cite the page, and later AI answers about "average time to hire" reference it as a source.

Original-data checklist

• [ ] Data is yours, legal and privacy-safe
• [ ] Method, sample size and dates stated
• [ ] Uncertainty reported where relevant
• [ ] One quotable sentence per key finding
• [ ] Numbers also in text, not only charts
• [ ] Permanent URL and update schedule
• [ ] Nothing invented or exaggerated

Exercise

Find or create 1 original stat and write it as a quotable sentence.

1. List three data sources you already have.
2. Pick one question your audience asks that the data can answer.
3. Calculate the figure, noting the method and sample size.
4. Write it as a single sentence with number, measure, scope, date and source.
5. Ask a colleague whether they'd understand it if they saw it quoted alone.

Quiz

1. What did the GEO paper find about statistics?

a) They hurt visibility b) Adding them improved visibility in its experiments c) They were not tested Answer: b

1. What must you do before using customer data?

a) Check privacy, consent and contract limits b) Nothing c) Publish raw records Answer: a

1. What should a trustworthy statistic include?

a) Method, sample size and date b) Only the number c) A headline Answer: a

1. Why put numbers in text as well as charts?

a) Text is easy to read and quote b) Charts are banned c) It's shorter Answer: a

1. What should you do if the result is less exciting than hoped?

a) Change the number b) Publish it honestly c) Hide the method Answer: b

Next: Lesson 6: Refreshing existing content

Sources

• Aggarwal et al., GEO: Generative Engine Optimization
• NIST/SEMATECH, Confidence intervals for proportions