Why this signal matters for Google
A number only you have cannot be copied. Every other page on your site can be rewritten by someone else in an hour. Your data cannot. A plumber who logs the price of their last 200 jobs publishes an average nobody else knows. An online store that counts its real return rate over twelve months holds a figure its competitors do not have. That is what original research means in practice: a statistic, a study or a survey you produced yourself. It does not take a lab. It takes counting something only you are in a position to count. And it pays: in the annual Orbit Media Studios survey (2026) of 1,042 content marketers, publishing original research raises the odds of strong results by 50%, yet fewer and fewer marketers do it. The first number I pull out of a new law firm client is the average delay between the first call and the first appointment: it has been sitting in their calendar for years and nobody ever published it.
Google has its results reviewed by human evaluators equipped with a rulebook anyone can download: the Quality Rater Guidelines, or QRG. Section 3.2 covers the main content, meaning the part of the page the page exists for. Four criteria keep coming back: effort, originality, talent or skill, and accuracy. Data you gathered yourself checks all four at once. An article that repeats numbers from a study already quoted everywhere adds nothing new, and that is exactly the case the section describes as content made without effort. Content effort is decided there first.
Data without a method is just a claim. Write "68% of contractors never call back the same day" and the reader immediately wants to know who counted, across how many cases, and when. Section 2.5.2 of the QRG asks the evaluator to identify who is responsible for the site and who wrote the page. Section 3.4 judges expertise on real knowledge of the topic. Both are answered in the same place: the method box. Number of cases, collection period, how you counted, and the name of the person who ran it. A scientific publication has always done this, and a blog post can do it in five lines.
AI engines quote numbers, not opinions. An answer from ChatGPT, Perplexity or a Google AI Overview is built from short, attributable sentences: a percentage, an average, an amount, with a named source. A paragraph of general advice gives them nothing to grab. A line like "across 312 quotes analyzed in 2025, the average reply time was 4.2 days" travels as is. That is how generative answers work: your name moves with the number. A well-phrased original figure becomes a sentence other people reuse, with your brand attached to it. The industry knows it and still does not do it: in the BuzzSumo and Mantis Research survey (2020) of nearly 650 marketers, 94% believe original research strengthens a brand's authority, yet only 39% had published any that year.
The NeuroEEAT scoring grid (0-10)
Assess what your site contributes in hard numbers, from a page that only quotes others to a study others quote.
| Score | Description | Concrete indicators |
|---|---|---|
| 0-2 | No data of your own | Every number on the site comes from somewhere else. Nothing separates your pages from the ten others repeating the same sources, which section 3.2 reads as content made without effort. |
| 3-4 | A number with no method | You state a percentage or an average, but nothing says where it came from: no case count, no period, no author. Section 2.5.2 is left unanswered. |
| 5-6 | Documented data | The number exists and the method is written out: case count, period, how you counted. It is buried in a paragraph and nobody can reuse the underlying data. |
| 7-8 | Reusable study | Visible method box, readable table or chart, raw file available for download, named author. A journalist can cite you without having to email you first. |
| 9-10 | Cited and repeated | The study gets picked up by third parties, is rerun on a regular schedule, is marked up in the code, and the numbers on the page match the published file exactly. |
How to audit it on your site
What to check to find out whether your site produces data or only consumes it.
- 01
List what you already count
Open your invoices, your support tickets, your forms, your analytics. You already measure things without thinking about it: reply time, average order value, the most common cancellation reason. That is the raw material of a study, and it is already in your hands.
- 02
Separate borrowed numbers from produced ones
Go through every page and tag each number one of two ways: taken from elsewhere, or produced here. Many sites discover they have nothing in the second column. That is the honest starting point.
- 03
Check the published method
For every in-house number, make sure four things are visible: how many cases, over what period, how you counted, and who ran the measurement. Section 2.5.2 asks precisely that anyone be able to identify who stands behind the content.
- 04
Test whether it can be quoted
Try to pull one sentence containing the number, the unit and the source. If you cannot, neither a journalist nor an AI engine will. A readable table and a figure written as text rather than baked into an image fixes it.
- 05
Check the freshness
Look at the collection date of each figure, not the publication date of the article. A 2021 number written in the present tense turns false over time. Decide right away when you will run the measurement again.
How to activate or improve it
The path, step by step, from a site that only quotes others to a source others quote.
Get a first number out
- Pick something you already measure with no extra work: reply time, average price, cancellation rate.
- Count across twelve months and at least thirty cases, which is enough to say something honest.
- Write three lines of method directly under the number: how many cases, what period, who counted.
- Publish that number inside an article you already have instead of waiting for the perfect study.
Turn the number into a study
- Run a short survey among your customers or your industry: forty responses are enough if you say so.
- Add a simple table or chart, with the values written out as text beside it so machines can read them.
- Sign the study with a named author and their role, never with the company name alone.
- Date the collection and announce the next edition in the same sentence.
- Spell out the limits: small sample, single sector, any bias you already know about.
Make the study reusable
- Publish the anonymized raw file as a CSV, alongside a detailed methodology page.
- Mark the page up in JSON-LD, a block of code invisible to the reader, using the Dataset type with the author as a Person.
- Rerun the same measurement every year with the same method: the series is what keeps citations coming.
- Send the study to journalists and industry sites, with the key figure in the subject line.
- Track who cites you and publicly correct any misquoted figure.
5 common mistakes to avoid
Sector specifics
Health advice site
A number about a symptom or a treatment affects the reader's health. Section 3.4.1 expects real expertise here: the data must be signed by a qualified professional, the method published, the limits written out. An internal survey does not replace a clinical study, and the text has to say so rather than let the reader guess.
Online store
Your sales data is a resource nobody else holds: return rate by category, average product lifespan, the gap between the size ordered and the size kept. Aggregate it, anonymize it, publish it. A publication writing about your market has nowhere else to find figures like these.
Software vendor or agency
An annual report built on your usage data creates an industry benchmark. The trap is bias: your users are not the whole market. Say it on the first page. Naming the limit strengthens credibility rather than weakening it.
Trade or neighborhood business
A local business can count what nobody counts: average job price in its city, the real gap between the first call and the appointment, the months when demand spikes. A local newsroom will happily pick up that kind of figure, and the link that follows is worth more than ten directories.
Signals to activate in parallel
Original data earns its value when it is published cleanly and connected to the other proof you produce.
FAQ
What is original research when you have no budget and no team?
It is counting what you already have. Your invoices, support tickets and forms hold measurements nobody else can take. Thirty to fifty cases over twelve months are enough for a first figure, as long as the method sits right underneath it.
How many responses does a survey need to be credible?
There is no magic threshold. Forty responses reported as forty responses beat three hundred with no method. What makes a survey credible is stating how many people answered, who they are, and what the number does not prove.
Can you publish figures drawn from customer data?
Yes, if the data is aggregated and anonymized and no customer is identifiable, even indirectly. Check your own terms of service and the privacy rules that apply to you before publishing. When in doubt, ask the customers concerned for explicit permission.
Does original data really earn links?
It is the format that earns the most, because other people need a figure to cite and they cite the one they can find. Nothing is guaranteed: a study nobody sees earns nothing. The outreach work matters as much as the collection work.
Should the study be republished every year?
Yes, as soon as the measurement can be repeated. A three-year series becomes a reference journalists start waiting for, and each edition brings citations back to the earlier ones. If you cannot rerun it, keep the collection date visible and stop writing the numbers in the present tense.
Case study
Before · regular posts, no data pages
After · data pages that answer journalists' searches
Source: Backlinko (Brian Dean), 2022, https://backlinko.com/reverse-outreach · Figures published by Backlinko about its own pages. These are compiled data presented cleanly; the mechanics are the same for data you collect yourself.