How to Publish Original Research That Others Can Check and Cite

Our own example is the 64-site machine-access study, measured on 18 September 2026, with a row per site you can download as CSV or JSON. The rule we followed: a figure leaves the page only with its date, its sample and the file it was counted from.

Publish original research around a question readers need answered, with enough method and data for someone else to check the result. A useful chart is only one part of the work. Readers also need the population, measurement date, definitions, exclusions and limitations behind the headline.

In short

• Start from a question you can measure directly, and write the claim down before you collect data.
• Publish the method, the sample, the definitions, the failures and the limits next to the result.
• Offer the data in a form others can check, with a stable URL and a date.
• Original data is useful, but no study shows engines cite it more; measure that separately.

Vercatus’s 64-site access study is an owned example. It observes access responses from our network. It does not measure how often assistants cite the sampled sites, and it is not a representative census of the web.

Choose a question you can measure directly

Write the proposed claim before collecting data. Identify the unit: a site, page, request, answer or person. Then ask which observation could support the claim without substituting a convenient proxy.

A successful request supports a statement about delivery under that request’s conditions. It does not establish a citation. A saved answer with a source link supports a dated citation observation. It does not establish customer adoption or incremental sales.

Define what would count as missing, refused, challenged and failed before the measurement starts. Retain those cases rather than removing them to improve a percentage.

Make the sample understandable

State how participants or sites were selected, which geography or niche they represent and what was excluded. A hand-picked list of recognizable brands can support an exploratory study, but its results should not be presented as an estimate for all businesses.

For repeated measurements, preserve the roster and identify replacements. If a site disappears, report that event. A changing sample can shift the headline without any sampled site’s behavior improving.

Publish a data dictionary

Item	What a reader needs
Population	Who or what the study concerns
Selection	How the measured sample was chosen
Unit	What one row and one observation represent
Date and context	Time window, origin and relevant settings
Variables	Definitions, units and possible states
Missingness	How failures and unavailable results enter totals
Derivation	How each headline is calculated
Limits	What the measurement cannot establish

In our study, one downloadable row represents a site, with page-level observations represented in fields. The CSV and JSON expose the data and method. The purpose-specific interpretation matters because the measured crawler names do not all represent search discovery.

Give every chart an inspectable claim

Use a sentence that names the sample, result and date. Keep denominator and missing cases near the percentage. Provide a readable table or downloadable source behind a graphic so the evidence does not depend on interpreting colors or pixels.

Avoid moving a footnote beyond the point where the headline would be misleading without it. In an access study, “requests from our network” belongs near the conclusion, because a verified vendor may receive a different response.

Prepare an honest citation path

Give the study a stable URL, clear title, author or publisher, observation date and reuse terms. Offer a short suggested attribution and link directly to the data. A journalist should be able to distinguish a quotation from your interpretation and a historic result from a newly measured one.

Appropriate Dataset metadata can describe the published asset, but it should match the visible method and downloads. Do not assert a license that you have not chosen or an external review that did not occur.

Keep corrections and follow-up visible

Maintain a correction channel and revision history. Preserve the original measurement date when changing wording. If you rerun the study, explain comparability and any method changes before presenting a trend.

What the sources say

• Specific, sourced content is more visible. In the GEO benchmark (Aggarwal et al., KDD 2024), rewriting sources to add statistics, quotations or citations raised their visibility in generative-engine answers by roughly 30 to 40% on position-adjusted word count; Google’s helpful-content guidance asks whether content provides original information, reporting, research or analysis. No study located shows that original data specifically is cited disproportionately (original data and citability, grade D). Read that as a case for evidence-rich writing, not a guarantee for studies.
• Earned coverage is where many citations land. Across more than 25 million links cited by ChatGPT, Claude and Gemini, one analysis found 84% of citations went to earned media and 0.3% to paid or advertorial content (earned media, grade C). A study that journalists can quote may travel further than the same facts buried in a product page, but that is a possibility to test.
• Readers look for who wrote it. Google asks whether pages carry bylines that lead to background about the author and whether content shows first-hand expertise or original research (author experience, grade A). That is why this study and its author page are linked.

A template for the write-up

1. Title and one-sentence finding. Name the sample and the date in the sentence.
2. Summary. The result, the denominator, the main limit.
3. Question and claim. What you set out to test, written before the data.
4. Method. Selection, origin of requests or observations, dates, tools in plain words.
5. Results. A table beside every chart; counts beside percentages.
6. Limits. What the data cannot say.
7. Data and reuse. Downloads, a data dictionary, suggested attribution, the licence you have actually chosen.
8. Corrections. How to report an error and a visible revision history.

Pre-publication checklist

• Could someone repeat the measurement from your description?
• Are failures and missing cases in the totals, not deleted?
• Does every headline number have its denominator next to it?
• Is the date of observation visible and separate from the publication date?
• Does the study avoid saying more than it measured (for example, “served” is not “cited”)?
• Would the headline still be fair without the chart?

A worked example, with invented details

A fictional analyst studies 50 city council websites to see whether the agenda page loads without scripts. The claim, written first: “Of 50 sampled council sites, the agenda page returned its main text in the initial HTML on N.” The write-up names how the 50 were picked (one per region from a public list), when and from where the pages were fetched, and what counted as “main text”. It reports the pages that timed out as a separate row, not as passes. It publishes a CSV with one row per site. The numbers are invented; the structure is what to copy.

Common mistakes

• A hand-picked list presented as a representative sample.
• Removing failed requests to improve a percentage.
• A chart with no table behind it.
• Claiming a licence or external review that does not exist.
• Rerunning a study with different methods and presenting a trend.

Questions people ask

Do I need a large sample?

You need an honest one. A small, well-described sample is exploratory and should say so.

Will publishing research get me cited?

It gives writers evidence to cite. Citations are an outcome to track, not to promise.

Should the data be downloadable?

Yes, with a dictionary. It lets others check the claim, and it protects you from misquotation.

A public study can give writers and editors useful evidence, but it does not guarantee coverage, backlinks or assistant citations. Track actual external uses and source links separately from page visits. Use the measurement guide to retain evidence of citations and the freshness guide to maintain accurate dates.