Observation · ChatGPT

How do AI search engines choose comparison pages as sources?

AI search engines choose comparison pages based on how useful, trustworthy, relevant, and easy to extract information from those pages.

15 November 2026, 21:52 CSTen-US (English answer; zh-CN interface)Run 1 of 3Method v0.1
Question

Recorded prompt

“How do AI search engines choose comparison pages as sources?”
Product
ChatGPT
Surface
ChatGPT web with web search
Model
ChatGPT automatic model (model version not exposed)
Locale
en-US (English answer; zh-CN interface)
Observed answer
Open session ↗
Observed
15 November 2026, 21:52 CST
Fresh conversation
Yes
Answer snapshot

What the answer said

: AI search engines choose comparison pages based on how useful, trustworthy, relevant, and easy to extract information from those pages. Unlike traditional search engines, which primarily return a list of links ranked by relevance, AI search engines often retrieve several sources, extract information from them, compare the claims, and use selected sources to construct an answer. For example, if someone searches "Best project management software compared", an AI search engine might look for: Independent comparison articles. Product-versus-product reviews. Detailed feature comparison tables. Official product documentation. Expert reviews and customer feedback. The important distinction is that ranking highly in traditional search does not automatically guarantee being cited by an AI search engine. 1. The six stages of source selection Query understanding The engine interprets the question and identifies what needs comparing: price, features, performance, limitations, use cases, etc. Query expansion It may generate several smaller searches. For example: Product A vs Product B Product A pricing Product B integrations Best alternatives to Product A Candidate retrieval It retrieves potentially relevant webpages from its search index or other available sources. Passage evaluation It examines individual sections of pages to determine whether they contain useful information answering the question. Evidence selection It favours useful, credible, relevant and sufficiently current information, while potentially excluding duplicate or weak sources. Citation selection It generates the answer and chooses which sources to display as supporting citations. This is a simplified model. Individual search engines use different systems, and their precise selection algorithms are not public. This distinction matters: a page can be discovered without being selected, and a page can influence an answer without appearing as a visible citation. The AI Search Handbook +2 2. What makes comparison pages attractive sources? Here are the main characteristics that can make a comparison page useful to AI search systems. Characteristic Why it matters Direct relevance Matches the exact comparison being requested Detailed comparisons Provides evidence about multiple alternatives Clear tables Makes differences easier to extract Specific facts Includes prices, features, specifications and limitations Independent evidence Supports claims with verifiable information Topical authority Demonstrates knowledge of the subject Freshness Keeps changing information current Clear structure Makes individual passages understandable outside their original context One particularly important factor is semantic relevance. A page titled "The Complete Guide to Business Software" might be less useful for a particular query than a page titled "Product A vs Product B: Features, Pricing and Limitations". The second page more directly addresses the information need. Research published in October 2026 also suggests that matching the search engine's rewritten queries can be more important than superficial page-formatting features. arXiv +1 3. Do AI engines prefer comparison websites over official websites? Not necessarily. They serve different purposes. For a comparison query, an AI engine may combine several types of sources: Source type What it contributes Independent comparison sites Side-by-side evaluations Official product websites Features, specifications and pricing Review publications Hands-on testing and expert opinions Customer reviews Real-world experiences Technical documentation Verified technical capabilities For example, an AI engine comparing two software products might use an independent review for its evaluation, but consult both manufacturers' websites to verify their pricing and features. This helps reduce reliance on a single source. However, comparison websites can have commercial incentives, particularly where affiliate commissions are involved. A well-presented comparison is not automatically an unbiased one. 4. What does an ideal AI-citable comparison page look like? Consider this example structure: Example structure Product A vs Product B (2026) An independent comparison of features, pricing, integrations and limitations. 1. Direct answer A short explanation of which product is better suited to different needs. 2. Side-by-side comparison Feature Product A Product B Pricing £20/month £30/month Integrations 40 65 Free plan Yes No API access Yes Yes Illustrative figures, not actual product data. 3. Individual product analysis Main strengths Weaknesses Important features Ideal use cases 4. Evidence and methodology Links to official documentation Date pricing was checked Explanation of testing methodology Sources for factual claims 5. Final verdict Explain which product is suitable for which type of user, rather than declaring one universal winner. This format makes the information easier to interpret, verify and reuse. 5. Do different AI search engines select sources differently? Yes. Search engine General approach ChatGPT Search Retrieves web information and uses relevant sources to support answers Google AI Overviews Uses Google's search infrastructure and AI-generated summaries Perplexity Searches the web and presents answers with supporting citations Microsoft Copilot Uses web search and grounding to support generated responses Their exact source-ranking and citation-selection formulas are not publicly disclosed. SEORecheck +2 6. The most important distinction: ranking vs citation There are three separate outcomes: Visibility: Does the comparison page appear in the search engine's candidate results? Selection: Does the AI retrieve and use information from that page? Citation: Does the AI actually display the page as a source? A website can achieve the first without achieving the second or third. This is why measuring traditional SEO rankings alone doesn't tell you how visible a comparison page is in AI-generated answers. 7. What this means for comparison-page publishers The practical lesson is to create pages that function as useful evidence, not simply pages that mention lots of competing products. The strongest approach is generally to: Answer specific comparison questions directly. Include genuinely useful side-by-side data. Make factual claims independently verifiable. Explain how comparisons were conducted. Keep time-sensitive information updated. Clearly disclose commercial relationships. Build a reputation for reliable coverage of the subject. One final caveat: No publisher can guarantee that ChatGPT, Gemini, Perplexity or another AI engine will cite a particular page. Their retrieval and citation processes are dynamic. For further reading, the Ofcom research paper on the era of answer engines provides useful background on how generative search systems retrieve and present information. The key takeaway: AI search engines don't simply choose the best-looking comparison page. They try to find retrievable information that helps answer the particular question, then decide which evidence is useful enough to support the generated response.

Linked citations

1 recorded sources

Limitations

What this observation cannot show

Authenticated free-account ChatGPT browser run in a fresh conversation. The product did not expose its complete retrieval configuration. The source panel reported 1 sources and 1 unique source records were captured. Results can vary by session, locale, account state and date.