Original research can give readers information unavailable elsewhere. To make it useful, publish enough detail for another person to understand how the findings were produced. A claim of originality alone does not establish quality or guarantee citations.
Make claims verifiable
A statistic with a defined population and method is easier to assess than a vague claim. The GEO paper tested additions such as statistics and sources, with improvements under its experimental conditions. That evidence does not establish the return from every research project.GEO: Generative Engine Optimization (Aggarwal et al., KDD 2024). Effectiveness varied by domain, so read the reported maximum as a result observed in a benchmark, and expect drift as engines evolve.
Google’s content guidance asks publishers to demonstrate relevant experience and explain authorship. These practices help readers evaluate a source. They should not be presented as a known formula for every AI product.Google's guidance on helpful, reliable, people-first content, the document that defines E-E-A-T.
Author, evidence, method, and date
Provide the author or analyst, the evidence, the method, and the collection date. State qualifications where relevant. Explain missing data and limitations. Use stable source links so readers can check the claim.
Where credited citations actually point
Publish results on an accessible page with the method nearby. If a dataset is offered, document its fields and permitted uses.Research sources and limitations. Identifies the available report data and the analysis still needed to support earlier draft figures.
Independent coverage can help relevant readers find the research. Its effect on future model training is uncertain. Evaluate publication opportunities for their audience, accuracy, and editorial relevance.
Original research on a small team's budget
Choose a question you can answer with data you are permitted to use. A small, well-described benchmark or survey can be useful. Explain who or what was measured and avoid generalizing beyond that population.
Publish the sample size, collection dates, calculations, and limits with the findings. When contacting a publisher, explain the result and why its readers would care. Do not claim a broader conclusion than the method supports.
Maintaining a statistics page
A statistics page can organize findings on one subject. Give each number its source, date, definition, and denominator. Distinguish your research from third-party findings and update entries when their source changes.
Publish the calculation behind a headline
A research page should let a reader identify the measured population and reproduce a headline number from the supporting table. Include collection dates, units, exclusions, scoring rules, and the calculation. Describe generated prompts as generated prompts.
If 9 of 18 completed answers mention a company, publish both numbers with the 50% rate. State that two other requests failed. A headline based on “half of AI answers” would imply a broader population than this test supports.
Keep a versioned copy of the data used for the article. If a correction changes the sample, recompute the table and any related chart. Explain the correction instead of silently substituting a new percentage.
Check whether the data can be shared. Remove private customer details from publication copies while preserving the internal records needed to audit the work. Describe any omissions that affect reproducibility.