Why this signal matters for Google
A dataset that other people reuse makes you the primary source. When a journalist, a researcher or a competitor writes "according to data from…" and names you, they admit the information starts with you. They cannot cite anyone else: you are the beginning of the chain. That is the hardest form of authority to fake, because every citation is written by someone you do not control. It is exactly what Google is trying to measure. Human evaluators rate the quality of websites using a public manual, the Quality Rater Guidelines (QRG for short). Section 3.3.1 tells them to do reputation research: look for what outside sources say about you. A site that claims "we are the reference in our industry" adds nothing. The same site whose annual salary table is picked up by three trade outlets and two university papers, with its name each time, adds proof the evaluator finds on their own.
Publishing a dataset is not the same as publishing a number. Many sites run original research and show only the conclusions: a chart, three percentages in a blog post. Nobody can reuse a chart. A dataset is the raw file behind the study: the full table, row by row, with a description of every column and the collection method. A florist who records the price of ten bouquets at twenty shops every week for a year owns a dataset. So does an accounting firm that compiles payment delays across 300 anonymized invoices. That file only has value for others if they can open it, understand the columns and reuse it without asking your permission. Hence the three conditions of the signal: a clear license, an open format and a permanent identifier.
Every reuse creates a citation you did not write. When a blogger rebuilds a chart with your numbers, they write "source: your company, 2026 dataset" and, most of the time, add a link. That is an authority backlink you never asked for, from a site that needed you. When a researcher cites your DOI, the permanent identifier of your file, the citation enters academic databases. When an outlet picks up one row of the table, your name appears in a press article. These pickups are exactly what you ask of others when you cite your sources: you become everyone else's source. On your own site, the evaluator needs to find the dataset page, its method and the list of those who reused it. Without that page the citations exist, but nothing ties them back to you.
AI tools look for data, not opinions. Ask ChatGPT, Claude or Perplexity "how much does … cost on average" or "what share of …" and the answer cites the source that published the number first, along with its method. A post that comments on someone else's numbers is almost never cited. The original file, well documented, with a clear title, a date and a license, often is. A 72 signal analysis typically finds this signal empty on most sites, even those that own unique data: it sits in an internal spreadsheet. The day you publish it properly, you have no competitor on that number: you are the only one who can be cited.
The NeuroEEAT scoring grid (0-10)
Rate what you have published, how easily it can be reused, and who has picked it up.
| Score | Description | Concrete indicators |
|---|---|---|
| 0-2 | No data published | You may own unique data, but nothing is online. Or only a chart in a post, with no file behind it. Or a number taken from elsewhere and presented as yours. |
| 3-4 | File online, but unusable | A file exists: a PDF, a screenshot, a table pasted into a page. No license, no column descriptions, no method. Nobody knows whether they are allowed to reuse it. |
| 5-6 | Clean dataset, no pickups | An open file (CSV), a clear license, a page that describes the method. But nobody has cited it yet, or you do not know, and it has no permanent identifier. |
| 7-8 | Cited dataset | Open file, license, method, DOI or permanent address. A few real pickups by sites you do not control, and a section on your page that lists them. |
| 9-10 | Primary source of your industry | A regular release, yearly or quarterly, picked up by trade press and academic work, marked up in structured data, with a version history. Your name is the reference whenever someone looks for that number. |
How to audit it on your site
Five checks to find out whether your data is a source or a forgotten file.
- 01
Find the data you already own
Open your tools: point of sale, invoicing, CRM, forms, field logs. Note every table nobody else has: recorded prices, turnaround times, sizes, survey answers. Data is worth publishing when it is unique, collected with a method and can be anonymized. If you find nothing, the audit stops here and the work starts at the activation stage.
- 02
Search for what you have already put online
Type your company name into Google with "data", "study" or "figures". Look at what comes up: a post with a chart, a PDF, a dedicated page? Open each page and ask whether someone could download the raw file and understand every column without emailing you.
- 03
Check the license and the format
On the dataset page, a license must be spelled out, for example Creative Commons BY, which allows reuse as long as you are credited. The file must open in any spreadsheet: CSV or JSON, not a screenshot or a PDF. Without those two elements, a careful journalist will not reuse it.
- 04
Count the pickups you do not control
Search for the exact title of your dataset in quotes, then your name with "according to" or "source". List every site, outlet or paper that cites you, with the date and the link. This is the search an evaluator runs by following section 3.3.1 of the QRG. Zero results means the signal does not exist yet, even if the file is perfect.
- 05
Ask AI tools where the number comes from
In a fresh session, ask ChatGPT and Perplexity the question your data can settle, for example "what is the average price of …". If the answer names your company and your file, you are the source. If it cites someone who reused your numbers without naming you, your page is missing a clear title, a date or a license.
How to activate or improve it
Three levels, from the first published file to the yearly release your whole industry waits for.
Publish a first clean dataset
- Pick a table only you own that answers a question your customers ask: "how much does it cost", "how long does it take", "which size should I choose".
- Remove anything that identifies a person: name, address, email, customer number. Anonymized data can be published; personal data never can.
- Save the file as CSV with a clear header row, and write a short note next to it explaining every column, the unit and the collection period.
- Create a page on your site with the dataset title, the date, the method in five lines, the license (Creative Commons BY works in most cases) and the download link.
Make the file easy to cite
- Deposit a copy in a public repository that issues a DOI, such as Zenodo or Figshare: a permanent identifier researchers and journalists know how to cite, and that survives a redesign of your site.
- Write the ready-to-copy citation on the page: "Company (2026), Dataset title, version 1, DOI". The easier it is to cite, the more you get cited.
- Publish a post that tells one or two surprising findings from the file, with the chart and the link to the data. The post attracts, the file retains.
- Send the dataset to three outlets or bloggers in your field who lack numbers, with the direct link and a suggested angle. Do not ask for a link: give them data.
- Add a "Who used this data" section to the page, with the link to every pickup, and update it with every new citation.
Become the yearly source of your industry
- Publish on a fixed date, every year or every quarter, with the same title and a version number: regularity turns you into an appointment, and outlets come back on their own.
- Mark up the page in JSON-LD, Google's structured data format, with the Dataset type: name, description, license, date, distribution format and author. Google Dataset Search, Google's search engine for datasets, only sees pages marked up this way.
- Keep every version online, with its date: a 2024 pickup must still lead to the 2024 file.
- Watch for pickups with an alert on the dataset title and on your name followed by "according to". When a site reuses your numbers without naming you, write and ask for the credit and the link.
- Cross your data with that of an association or a university in your field and publish the file under both names: two source names are worth more than one, and the partner cites the file on its side.
5 common mistakes to avoid
Sector specifics
Clinic or practitioner
A topic that carries risk for the reader, which Google calls YMYL, for Your Money or Your Life. Here the data concerns patients: the only rule is full anonymization and compliance with your profession's rules on health data. A physiotherapy clinic can publish the average number of sessions per type of injury, never one row per patient. Describe the method carefully: on a health topic, a number without a method does more harm than no number.
Firm or advisor
Turnaround times, costs and case outcomes are data everyone looks for and almost nobody publishes. A firm that publishes the average processing time of 200 anonymized cases, by type of request, becomes the source outlets cite. Check what your professional rules allow, and remove anything that could identify a client.
The field survey nobody else has
A tradesperson or a shop has field data: prices recorded in its area, real response times, seasonal demand. A plumber who publishes the average price of twenty common jobs, recorded every year, gets cited by local outlets and advice sites. The file is small, but it is unique.
Anonymized usage data
A software vendor sees thousands of uses go by. Published in aggregate and anonymized, for example the average duration of a process by company size, that file becomes an industry reference picked up by analysts and by competitors themselves. Date every version and keep the method stable from one year to the next, otherwise the comparisons no longer hold.
Signals to activate in parallel
A dataset extends original research and feeds other authority signals. These go with it:
FAQ
What is a dataset?
A dataset is a structured table: one row per observation, one column per variable, most often in a CSV file. For this signal, what matters is not the size but the origin: a dataset you collected yourself, that nobody else owns, and that others can download and reuse. A chart or a PDF is not a dataset: nobody can rerun their own calculations on it.
How can I create my own dataset?
From data you already collect. Open your point of sale, invoicing, CRM or field logs and pick one table nobody else has: recorded prices, turnaround times, survey answers. Remove anything that identifies a person, save it as CSV with a clear header row, and write a note describing every column, the unit and the collection period. A 200 row table you collected yourself is a dataset; a 100,000 row table copied from a public source is not yours.
How to publish a dataset online?
Two places, same file. On your own site, a dedicated page with the title, the date, the method in five lines, the license and the download link: this is the page evaluators and AI tools will find. In a public repository such as Zenodo or Figshare, a copy that gets a DOI, a permanent identifier researchers know how to cite. Mark up your page with the Dataset type in JSON-LD so Google Dataset Search can list it, then tell three outlets in your field that the data exists.
Which dataset license should I choose?
In most cases, Creative Commons BY: anyone can reuse the data, including commercially, as long as they credit you. That is exactly the effect you want. A license that is too restrictive, with no commercial use or no modification, scares off outlets and researchers, and with them the citations. Spell out the license on the page, not only inside the file.
How to cite a dataset, and how do I make mine easy to cite?
A dataset citation needs the author, the year, the title, the version, the publisher or repository, and a permanent identifier, ideally a DOI. That is the minimum most style guides and repositories ask for. Write that exact line on your dataset page, ready to copy, and keep the DOI stable across versions. The easier it is to cite, the more you get cited, and every citation is a pickup an evaluator can find.
Case study
Before · data asleep in a spreadsheet
After · a source that others cite
Illustrative figures meant to show the typical progression of the signal. They do not come from a real client case and must be replaced with your own measurements before any external communication.