Research Guides
Publish Website Research People Can Check and Cite
A small website study can be useful when it answers a specific question and shows how the answer was produced. Before asking anyone to link to your research, make it possible for a reader to inspect the observations, understand the exclusions and repeat the calculation.
This guide provides a practical publication template and a worked example. The example is entirely fictional. It is not a survey of real businesses, a market benchmark or a report produced by a Colquhoun.shop scanning service.
Choose a question a page can actually answer
“How many selected homepages display a clearly labeled contact link?” is a researchable question. “Which businesses care about their customers?” cannot be established by counting links.
Write the observation rule before collecting results. For this English-language example, count a visible link labeled “Contact,” “Contact us” or “Get in touch,” ignoring capitalization. Include navigation and footer links. Do not count an unlabeled icon, a phone number without a link, or a contact page discovered through a search engine. Those may be useful to customers, but they answer different questions.
Fix the inspection conditions too. In this example, use a desktop browser at a 1280 by 900 pixel viewport, allow up to 15 seconds for loading, scroll to the footer and open any collapsed navigation menu once. A page that remains incomplete is not assessable. Another study may choose mobile conditions or different labels; disclose those choices rather than mixing methods midway through collection.
Choose the unit of analysis too. One homepage per business means a chain with 40 location pages does not receive 40 votes. A study of location pages could be valid, but its headline and denominator would need to say that.
Tell readers how the pages were selected
Keep the original selection list, its source and the inclusion rule. A handpicked list of businesses you already know is a convenience sample. Describe it that way and confine the findings to those pages. Calling the list “representative” requires evidence that this approach does not provide.
Disclose the collection dates, inspection method, researcher and sponsor. Explain any AI assistance and human checking. AAPOR's research-transparency guidance covers selection, measurement, exclusions and limitations, including content analysis. This small-study template draws on those disclosure principles; it is not an AAPOR certification or a substitute for a formal research design.
Record what happened to every selected page. Do not quietly replace an inaccessible page with an easier one. If access is blocked, stop and record that limitation rather than working around it. For practical interpretation of missing observations, see When a Page Cannot Be Read.
Keep one row per selected page
Use a spreadsheet or plain-text table with these columns. The proposed field names are a working convention for this example, not an industry standard.
- page_id: A stable local identifier, such as P01.
- business_id: An identifier linking the page to one business in the selection list.
- requested_url: The public homepage selected for inspection.
- final_url: The destination actually inspected after any redirect.
- checked_at: The observation date and time, including the time zone.
- inspection_method: Browser inspection or another precisely described method.
- assessment: present, not_observed or not_assessable.
- evidence_note: The relevant label and location, or the reason inspection failed.
- rule_version: The version of the contact-link definition used.
“Not observed” means the defined feature was not found during the stated inspection. It does not mean the business has no contact channel. “Not assessable” means the inspection could not answer the question. Keep these values separate when calculating results.
For a public release, include only evidence you are entitled to share. Do not expose customer records, private URLs, access tokens or personal contact details. A short observation and a public source link may be sufficient; republishing entire pages is unnecessary.
A fictional example: two honest percentages
Imagine 12 eligible homepages selected before inspection. Ten can be assessed and two cannot. These are invented teaching rows, with no real domains or measured findings:
- P01: present; “Contact” in the top navigation.
- P02: present; “Get in touch” in the footer.
- P03: not_observed; no qualifying label found.
- P04: present; “Contact” in the footer.
- P05: not_assessable; access blocked.
- P06: not_observed; a telephone number was visible, but no qualifying link.
- P07: present; “Contact” in the navigation menu.
- P08: not_observed; no qualifying label found.
- P09: present; “Get in touch” in the main page content.
- P10: not_assessable; page did not finish loading.
- P11: present; “Contact” in the footer.
- P12: not_observed; no qualifying label found.
The reconciliation is six present, four not observed and two not assessable. Six plus four plus two equals all 12 selected pages.
Among assessable pages, the share with a qualifying link is 6 divided by 10, or 60%. Among all selected pages, the share confirmed to have one is 6 divided by 12, or 50%. These percentages answer different questions. Label the denominator beside the number instead of choosing the larger percentage for the headline.
The two unknowns also let you show a simple boundary. If neither had the link, 6 of 12 would qualify. If both had it, 8 of 12 would qualify. That gives 50% to approximately 66.7% for this fixed list, assuming the ten completed assessments are correct. It is an unknown-outcome range, not a statistical confidence interval or an estimate for all businesses.
An accurate fictional summary would be: “Six of ten assessable homepages displayed a qualifying contact link. Twelve pages were selected; two could not be assessed.” Keep the fictional label if reusing these teaching numbers. Replace them only after doing and documenting a real study.
Publish the result with a compact method note
Put the count, denominator and collection dates near the main finding. Then make the method and supporting rows easy to find on the same page or through a clearly labeled link.
A useful method note has four short parts: the question and inclusion rule; the inspection procedure and dates; the treatment of missing or duplicate observations; and the limits on interpretation. Name the author and any sponsor separately. If all work was done by one person without an independent check, say so.
Copyable method-note template: “We selected [number] eligible homepages from [list and inclusion rule], with one page per business. [Researcher] inspected them on [dates] using [browser conditions and rule version]. We found [present count] qualifying links among [assessable count] assessable pages; [unknown count] pages could not be assessed for [reasons]. [Describe exclusions, quality checks and AI use, if any]. These observations describe the selected pages, not all businesses. Sponsor: [name or none].”
Before publication, ask a second reader to recalculate the headline from the rows and apply the rule to a few observations. Resolve disagreements in the definition, not by deleting inconvenient rows. Keep a record of corrections. This is a practical quality check, not proof that the sample represents a wider market.
Google's people-first content guidance encourages original information, clear sourcing and added value. A modest, checkable result is more useful than a dramatic claim whose evidence cannot be inspected. Neither the method nor the publication schedule guarantees search rankings or backlinks.
Make the study easy to cite and maintain
Give the report a stable URL, descriptive title, named author or organization, publication date and version. If the data change, add a dated correction or release note and preserve enough information to identify the earlier result.
Offer a suggested citation in this form: “[Author or organization]. [Study title]. [Version]. [Publication date]. [Report URL].” Link the supporting data and define the columns. Clearly identify any illustrative rows so they cannot be mistaken for observations.
Explain reuse permissions only for material you control. The FAIR Guiding Principles emphasize descriptive metadata, provenance and clear data-use licensing. A citation does not itself grant permission to republish someone else's text or images, and this guide does not apply a new license to any third-party material.
If you publish an actual dataset, Google's Dataset structured-data documentation explains how metadata can support discovery in Dataset Search. Describe the real dataset and its visible information accurately. Adding markup does not validate the research or guarantee discovery, citations or a special search appearance.
Share the finished report where it directly answers a relevant question, with the method and limitations intact. Let other publishers decide whether it merits a citation. The most useful outcome is a resource that saves someone else research time and gives them evidence they can check.