You can publish original data in a blog post without a research budget by counting, surveying or measuring things you already have access to: your own customer records, support questions, project timelines, a small survey of your audience or a structured test. Original numbers, even small ones, give readers and AI assistants something they cannot find elsewhere, and that is the main reason posts with them tend to earn links and mentions.
The conditions are honesty and method. You must say how many cases you counted, how you counted them and what the limits are. This guide shows seven sources of original data, how to run each cheaply and how to present results without overclaiming.
Why does original data matter?
Most blog posts restate what other posts say. Search engines and AI answer systems have many near-identical pages to choose from, so a page with a unique fact stands out. A statement such as “In our 40 quote requests this year, 31 asked about price before anything else” is something only you can say. Other writers who want to cite it must link to you.
It also makes your content more credible. A claim with a source and a method is easier to trust than a general opinion, a point we made in our post on measuring blog ROI when AI answers take the click, where pages that carry unique information tend to earn value even when the click goes elsewhere.
Seven sources of data you already have
1. Your own customer records
Count what you have: orders by month, project length, most common request, typical budget range. Remove names and anything confidential. “Our average website project last year took 9 weeks from brief to launch” is a real, original number.
2. Support and sales questions
Read the last 100 emails or chat messages and tag them by theme. Report the top five questions and their share. This is both content and research into what to write next.
3. A small audience survey
A five-question survey sent to your email list or posted on LinkedIn can reach 30 to 100 people. That is too few to claim anything about an entire market, but it is enough to describe your audience. Label it clearly: “Survey of 64 subscribers to our newsletter, September 2026.”
4. A structured test
Test something and report what happened. Run the same ten prompts through three AI assistants and record what each says about a topic, or compare load times of five competitor pages. Define the method before you start.
5. Public data, recombined
Government statistics, public registries and platform transparency tools hold data that few people turn into insight. Combine two sources in a way no one has, and credit them. Statistics Finland, Eurostat and the Meta Ad Library, for example, offer data that you can summarise for a specific industry or region.
6. Your own advertising or website analytics
If you run campaigns or have a website with analytics, you hold conversion rates, costs and seasonal patterns. Publish aggregated and anonymised results. Our post on setting your own cost-per-lead target argues that your own numbers beat generic benchmarks, and the same applies to content.
7. Expert interviews
If you cannot produce numbers, produce first-hand quotes and judgments. A set of five interviews with practitioners, each asked the same three questions, is original research of a qualitative kind. Our guide to interviewing a subject-matter expert in 30 minutes shows how to do it efficiently.
How to design a small study that holds up
- Start with one question. “How long do projects really take?” is a good question. “What is the state of the industry?” is not.
- Decide what you will count before you look. Write down the definition: what is included, what is not and the time period.
- Record everything in one spreadsheet. One row per case, one column per variable.
- Keep the raw sample size visible. Always report “n=” in the text.
- Look for the interesting number. Averages are dull. Ranges, outliers and surprises are what readers quote.
- Check for anything you would be embarrassed by. Missing data, overlap and cases you excluded should be noted.
How to present small numbers honestly
The danger with small samples is overclaiming. Follow four rules:
- State the sample. “Of 42 projects we delivered between 2024 and 2026” is clear and fair.
- Say what it does not show. “This reflects our clients, mostly small e-commerce firms, and may not apply elsewhere.”
- Use words that match the evidence. With 12 cases, say “most” rather than “75%”. Percentages from tiny samples look more precise than they are.
- Never invent or round up to a better story. One discovered fabrication destroys the value of everything else you publish.
The same principle applies if you cite someone else’s data. Link the original, name the year and say what it measured. If you cannot find the original, do not use the number.
A worked example
Imagine a small web design studio wants a post on how long websites take. It exports the last 30 completed projects from its project tool and records the weeks from signed brief to launch. It finds a median of 8 weeks, a shortest of 3 and a longest of 21. It then tags the five slowest projects and notices that four of them waited more than a month for client content.
The post practically writes itself: “Across 30 website projects, the median was 8 weeks. The five slowest all had the same cause: waiting for content.” That is original, specific, quotable and costs an afternoon with a spreadsheet. The claim is bounded to the studio’s own work, which makes it defensible.
How to format data so readers and machines can use it
- Put the headline finding in the first two paragraphs as a plain sentence.
- Use a table for the main numbers, with clear column headings.
- Describe any chart in text as well, because AI systems and screen readers read text more reliably than images.
- Add a short “Method” section at the end with the date, the sample and the definitions.
- Update the post annually and say so at the top.
Once published, refer back to your data in later posts. A study that you cite in six articles becomes a reference point, which supports the kind of topical depth discussed in our post on whether blogging is still worth it in 2026.
Things to avoid
- Sharing client-identifiable data without permission. Aggregate or anonymise, and check contracts.
- Surveys that lead the respondent. “Don’t you agree that…” produces useless results.
- Collecting personal data without a lawful basis. If your survey asks for names or emails, follow GDPR rules and say what you will do with the answers.
- Claiming causation from a coincidence. “Projects with a kick-off call finished faster” is not the same as “kick-off calls speed things up”.
The short version
You do not need a research department to publish something original. Look at what you already hold, count it carefully, show your method and keep your claims inside what the numbers support. One honest, specific number per post is enough to separate your writing from the pile.
How often to refresh your data
Original numbers age. Set a reminder to rerun your count every year, and say in the post when it was last updated. A “2026 edition” that repeats your method with fresh numbers lets readers see change over time, which is itself new information.