--- language: en title: "From Retrieved Page to Cited Source: How to Read AirOps’ Commercial Study" description: "AirOps analyzes 217,508 pages retrieved for commercial queries. Learn how to interpret structure, tables, and statistics without turning associations into formulas." canonical: "https://tropk.ai/en/resources/research/from-retrieved-page-to-cited-source-how-to-read-airops-commercial-study" last_modified: "2026-10-12T16:14:18.889Z" --- # From Retrieved Page to Cited Source: How to Read AirOps’ Commercial Study Reviewed on 2026-10-06. ## The question goes beyond appearing in search This Tropk editorial analysis, reviewed on October 6, 2026, covers research by AirOps published on April 1. The data and models are from AirOps. The study distinguishes between a page retrieved during search and one referenced in the final answer. For commercial content, this distinction helps investigate why information may be available but still not selected as a visible source. The useful takeaway is not to add a fixed number of tables or lists to every page. Instead, assess whether the content provides the information needed at that decision stage. An initial explanation, a comparison, and supplier validation serve different purposes. Improving structure can make content easier to understand, but it does not replace accurate facts, relevance, and access to the content. ## Key points - AirOps analyzes 217,508 pages already retrieved in ChatGPT from 7,500 commercial queries. - Being retrieved in search and receiving a citation in the final answer are different events. - Associations with statistics, sentences, tables, and lists vary by commercial decision stage. - The study’s percentages are not causal gains or required counts of editorial elements. - Adapt the format to the buyer’s task and test changes using verifiable facts, criteria, and sources. ## The analysis design and its denominator The [original report](https://www.airops.com/report/from-retrieved-to-cited-how-commercial-content-earns-citations-in-ai-search) analyzes 217,508 pages retrieved in [ChatGPT](/en/resources/glossary/chatgpt) from 7,500 commercial queries. There are four stages, with 1,875 queries each: discovery, list building, comparison, and validation. The study considers 25 page characteristics and uses logistic regression, random forest, and sensitivity analyses, including stratification by [domain authority](/en/resources/glossary/domain-authority). The article does not report exact collection dates, geographic distribution, or model version. The denominator is especially important: these are pages already retrieved. The study does not estimate the likelihood that any page on the internet will be cited. Teams that do not know whether a URL was retrieved should not use the findings to automatically explain its absence. Access, [indexing](/en/resources/glossary/indexing), topic coverage, and passage selection remain separate questions. ## The numbers and how to interpret them carefully AirOps reports associations with a higher likelihood of [citation](/en/resources/glossary/ai-citation): 20.3% for five to seven statistics during discovery; 18.8% for an average of up to ten words per sentence during list building; 25.7% for three tables during comparison; and 26.9% for eight list sections during validation. These are the results presented in the [AirOps source](https://www.airops.com/report/from-retrieved-to-cited-how-commercial-content-earns-citations-in-ai-search). These percentages are not percentage-point increases, editorial targets, or demonstrated causal effects. The analysis controls for some observable differences, but pages may differ in quality, brand, subject matter, and evidence not captured by the model. A table may be associated with well-researched comparative content; that does not mean inserting an empty table will produce the same result. Interpretation should distinguish an observed characteristic from an intervention that still needs testing. ## Define the page’s purpose before choosing its format During discovery, a page needs to explain the problem and help readers assess its importance. Data is useful when it includes a source, time period, and denominator; decorative numbers merely create the appearance of evidence. When building a list, readers need to understand options and criteria quickly. Clarity matters more than sticking to a word limit per sentence, especially when adapting a finding into Portuguese. During comparison, a table makes sense when alternatives can be assessed using the same criteria. Each cell should indicate known information, a condition, or missing data. Mixing prices with different terms and hiding limitations makes a comparison misleading, even if the format is easy to read. During validation, questions about integration, support, implementation, and scope may require concrete answers and examples—not eight invented lists to hit a count. A single page can serve more than one stage, but it needs an organization that preserves its main purpose. Before splitting content across many URLs, check whether there is a distinct need and enough information. Creating nearly identical variations increases maintenance work and can make it harder to find the right answer. The study does not show that increasing the number of pages improves citations. ## Applying the findings in Portuguese and in the Brazilian market The article does not report a Portuguese- or Brazil-specific analysis. Therefore, sentence limits and structural patterns should not be transferred as if they were local standards. Technical terms, integration names, and commercial terms may require longer explanations. A short sentence that omits a relevant condition is less useful than a slightly longer, precise one. Consider a software comparison for Brazilian businesses. Currency, tax information provided by the vendor, availability, support, and integration requirements can affect the choice. The team should look for current documentation and record the date it was checked. If information is not public, say it could not be found or must be confirmed; do not fill the table with inferences that look like facts. Adapting content also means distinguishing translation from research. An article from another country can help explain a category without addressing local conditions of use. Adding Brazilian examples only makes sense when evidence supports them. The goal is to reduce real buyer uncertainty without claiming that the same percentage gain will appear on the site after revision. ## A content experiment with a clear hypothesis Select comparison pages with accurate information that is difficult to find. The hypothesis might be that bringing equivalent criteria together in a table improves understanding and the page’s visibility as a source. Change only this aspect when possible; keep a record of other necessary changes. Choose similar pages for comparison and keep queries, search engines, and location consistent during observation. First assess whether people can find the answer, whether sources support the table cells, and whether conditions remain visible. Then track citations and the relevance of cited passages. An increase without evidence of prior [retrieval](/en/resources/glossary/information-retrieval) does not answer the study’s question about selection among retrieved pages. If the interface does not reveal retrieval, record that limitation rather than filling in the data based on assumptions. The test can inform the next revision even without demonstrating causality. If the table clarifies the decision, the work has editorial value independent of fluctuations in AI results. If it introduces repetition, outdated information, or false equivalences, revise the design. Reader quality must remain an approval criterion, not something to set aside when citation counts rise. ## Checklist for reviewing the conclusion - Does the page address the buyer’s decision stage and criteria, with verifiable facts and explicit conditions? - Are tables, lists, and statistics included because they help with the decision? Can each data source be checked? - Does the report distinguish association, an intervention hypothesis, and an observed result? Does it explain when retrieval and location are unknown? ## Further reading and work tools Explore [search intent](/en/resources/glossary/search-intent), [RAG](/en/resources/glossary/rag) and [content brief](/en/resources/glossary/content-brief). The [comparative article prompt](/en/resources/prompts/prompt-for-writing-a-comparison-article) helps define criteria, gaps, and sources; the [pillar guide prompt](/en/resources/prompts/prompt-for-writing-a-guide-or-pillar-page) supports comprehensive coverage without repetition. [Auditing](/en/product/audit) and [Strategy](/en/product/strategy) help organize issues and priorities. Editorial decisions should start with the observed need, keeping the study’s numbers tied to their original scope. ## Sources and references - [From Retrieved to Cited: How Commercial Content Earns Citations in AI Search]() — AirOps — Oshen Davidson ## Frequently asked questions ### What is the difference between a page retrieved and a page cited in an AI answer? A retrieved page was found during the search process used to build the answer. A cited page receives a visible reference in the final answer. AirOps’ commercial research analyzes selection among pages already retrieved; it does not estimate the likelihood that any page on the internet will be cited. ### What did AirOps’ research on commercial content and ChatGPT citations analyze? AirOps analyzed 217,508 pages retrieved in ChatGPT from 7,500 commercial queries, distributed equally across discovery, list building, comparison, and validation. The study examines page characteristics and their associations with citations. The report does not provide exact collection dates, geographic distribution, or model version. ### Does adding three tables or eight lists guarantee more AI citations for commercial content? No. AirOps’ research reports associations between characteristics of pages already retrieved and the likelihood of citation at particular decision stages. The findings do not demonstrate that inserting a fixed number of tables or lists causes citations to increase. The format should clarify real criteria using accurate data and comparable conditions. ### How can you apply AirOps’ commercial research to Portuguese-language pages targeting Brazil? First define the decision stage and the buyer’s question, then review local facts and conditions and choose tables, lists, or explanations when they aid understanding. AirOps’ research does not report a Brazilian or Portuguese-language analysis. Test specific changes while keeping queries, search engines, and location comparable, without promising the same percentages observed in the study. ## Continue a leitura - [Content brief]() - [Search intent]() - [Prompt for writing a comparison article]() - [Prompt for writing a guide or pillar page]() - [RAG]() - [AI Presence Audit]() - [AI discovery strategy]() ## About the author [Luiz Gardelin]() Luiz Gardelin is the founder of Tropk.ai and works in GEO/AEO, product vision, strategy, and growth, helping brands understand and improve their presence in AI responses. --- [All pages in Markdown]()