--- url: 'https://seoafterai.com/ai-search/which-passages-ai-answers-cite-and-which-they-use-unnamed/' title: 'Which passages AI answers cite, and which they use unnamed' author: name: Nadia Sorensen url: 'https://seoafterai.com/author/n-sorensen/' date: '2026-09-29T07:30:26+00:00' modified: '2026-09-29T07:30:26+00:00' type: post summary: 'Bart Magera, writing on the Advanced Web Ranking blog, coded 112 passages from 40 SEO queries. Passages that AI answers linked to were more often dated and named their subject outright. Passages used without a link were mostly restatements of common knowledge. The sample is small, and the author say' categories: - AI Search tags: - AI Overviews - Citations image: 'https://seoafterai.com/wp-content/uploads/2026/09/cover-264-5466831f.png' published: true --- # Which passages AI answers cite, and which they use unnamed Bart Magera, writing on the Advanced Web Ranking blog, coded 112 passages from 40 SEO queries. Passages that AI answers linked to were more often dated and named their subject outright. Passages used without a link were mostly restatements of common knowledge. The sample is small, and the author says so. ## Table of contents - [What this is about](#what-this-is-about) - [What the source says](#what-the-source-says) - [What the data shows](#what-the-data-shows) - [What it does not mean](#what-it-does-not-mean) - [What to do](#what-to-do) - [FAQ](#faq) **Key takeaways** - An AI Overview appeared on all 40 informational SEO queries tested on July 29, 2026. Fully captured answers cited 16.8 sources on average; Bing Copilot Search cited 4.9. - 119 of 201 AI Overview citations, or 59%, went to Reddit, YouTube or LinkedIn. - Cited passages showed a 2025–2026 date 80% of the time, against 53% for passages Google showed but did not cite. - Only 17 uncited passages were coded. Differences of a few points in that group rest on one or two passages. ## What this is about In a September 28, 2026 study, Magera looked below the page level. He asked which passages an AI answer credits with a link and which it uses without naming the source. For site owners, it suggests that ranking gets a page read, but common-knowledge wording may not earn the credit. ## What the source says Magera separates two outcomes. A cited passage appears in the answer with a visible link. An absorbed passage supplies information the answer uses, with no link. He is clear that absorption is “inferred, not proven”. He cannot see inside Google’s systems. He can only see an answer that closely matches a page shown on the same screen, with no citation to it. His clearest case is the query “what is semantic SEO”. The AI Overview’s definition tracked the first organic result, from WordLift, closely. The citations went to Reddit, Search Engine Journal and SE Ranking. In Magera’s coding, the WordLift passage had no visible date and read much like five other definitions. He also offers an explanation from linguistics: an engine links to a passage when the link adds something the answer cannot carry alone. That is his reading of the data. The study does not test it directly. ## What the data shows The collection ran on one day, July 29, 2026. Magera used logged-out browsers set to US English, from a Thailand IP address. He ran 40 prompts across five SEO topics and two engines, Google AI Overviews and Bing Copilot Search. He logged 265 citations. Only 12 AI Overview answers and 13 Copilot answers were captured in full, so the citation counts rest on those 25. | Passage attribute | Cited (n=95) | Shown but uncited (n=17) | | --- | --- | --- | | Fresh date visible (2025–2026) | 80% | 53% | | Subject named at first mention | 96% | 82% | | Pure restatement of common knowledge | 61% | 82% | | Direct answer in first sentences | 76% | 71% | | Written as a definition | 60% | 65% | | Contains a hard number | 6% | 0% | | Contains a new claim | 7% | 0% | Three rows show a clear gap: date, naming the subject, and common-knowledge wording. Two rows show almost none: answering early and writing a definition. Both groups did those equally. The size of the uncited group matters. Seventeen passages means each one is about six points. The 0% for hard numbers is zero passages out of 17. At the cited group’s rate of 6%, you would expect about one. So that row says little. The same caution applies to new claims. Two other findings stand out. On 11 of the 12 fully captured AI Overviews, every organic result visible on the page was also cited. And Copilot cited lower-authority sites than Google. Its cited passages were more often direct answers (82% against 70%) and more often recent (89% against 72%). On the seven method points this site asks of any study, this one does well on disclosure. The date, region, engines, prompt set and failures are all stated, and the materials are public. It gives no repeat runs, so answer-to-answer variation is unknown. One person did all the coding. ## What it does not mean **It does not show that adding a number gets you cited.** Only 6% of cited passages had one. The other 94% were cited without it. **It does not contradict large-sample studies.** Magera found AI Overviews citing almost every visible ranked page. Conductor, across 167.9 million citations, found a first-place page cited only 24.9% of the time, as we covered in [most AI Overview citations come from outside the top 10](https://seoafterai.com/ai-search/most-ai-overview-citations-come-from-outside-the-top-10/). Twelve informational SEO answers and every US search across all industries are very different samples. **It does not prove Google used a page it did not cite.** A close match on the same screen is suggestive. It is not a view into the system. **It is not neutral.** Magera founded Mojo Links, a link-building agency. That does not make the data wrong, but readers should know it. ## What to do 1. **Name the subject in the first sentence under each heading.** Write “Topical authority is…”, not “This approach is…”. Cited passages did this 96% of the time. 2. **Show a real updated date, and update the page behind it.** This was the largest gap in the table, 80% against 53%. Our look at [content recency in AI search](https://seoafterai.com/analytics-data/content-recency-in-ai-search-what-47097-citations-show/) found the same direction at far larger scale. 3. **Say one thing the standard answer does not.** Your own data, a named case, or a dated claim. The evidence here is thin, but it costs little to try. 4. **Check your key queries in more than one engine.** In this study, the same Backlinko page was cited by Copilot and left uncited by Google on the same day. ## FAQ What is an “absorbed” passage? It is text an AI answer appears to use without linking to its source. Magera infers it when the answer closely matches a page shown on the same results page and that page gets no citation. Why did Reddit and YouTube get so many citations? The study cannot say. Magera’s reading is that Google cites several independent sources to show that an answer is backed up. That is an interpretation, not a measured result. How much weight should this study carry? Treat it as a lead worth testing, not a rule. It covers 40 queries on one day, 25 fully captured answers and 17 uncited passages. The author calls it directional and publishes his materials so others can repeat it. Our [guide to how answer engines cite](https://seoafterai.com/ai-search/ai-search-how-answer-engines-cite/) gives the wider picture.