Every brand wants to appear in AI-generated answers. However, being found during an LLM’s research is different from being featured in the final answer users actually see. That distinction creates a new challenge for SEO and GEO: what makes a domain not just discoverable, but useful and trustworthy enough for an AI system to cite?
We started exploring this question while working on a recent study of 100,249 fan-out queries based on a dataset of 100,000 ChatGPT prompts. As an additional layer of analysis, we checked which domains appeared most often across our dataset and compared their presence in two places: the model’s search results versus its cited sources. We found that the lists overlap, but they are not identical. This post extends our original research by looking closer at these domains to better understand what separates appearing in the AI discovery process from actually being cited.
Contents
The gap between AI citations and AI search results
Main patterns across AI search results and citations
Key search-to-citation differences by category
1. Encyclopedias, dictionaries, education & government resources
2. Social, UGC, and review platforms
3. News & media publishers
4. Local, travel, ecommerce & delivery
5. Advice in real estate, personal finance, and career
6. Home, wellness & recipes
7. Entertainment and sports
8. Other (mixed category)
Conclusion
The gap between AI citations and AI search results
When we ask something in AI tools like ChatGPT, Perplexity, or Gemini, they may need additional or more recent information, and they browse the web to find it. During that research process, LLMs may discover many sources, but only a smaller group earns a visible mention. Essentially, two different layers of AI visibility are formed:
Search results – all relevant resources the model found. This indicates discoverability, but it’s weaker than being cited because users never see these results directly.
Cited sources – the list of sources selected to support the final answer. This is a valuable form of AI visibility because, in this case, a domain or page appears directly in the response.
The gap between these two is where valuable AI visibility is won or lost, so we decided to measure domain presence in AI search results and citations as separate signals.
How we measured the gap and other metrics
To understand the difference between being discovered by an AI system and being cited in its final answer, we focused on the top 200 cited domains in our ChatGPT-based dataset. While these are the top performers by citation count, they are not all cited equally. For example, Wikipedia and Reddit are chosen for citation over 17,000 and almost 8,000 times, respectively, while Dallasnews is referenced only 16 times, and Pinterest only 19.
When we compare how often each domain appeared in both discovery and citation stages, the results reveal interesting patterns about how AI search works behind the scenes.
Here are the exact metrics we used:
- AI Search Appearances – how many times the domain showed up in the model’s search results for prompts that triggered fan-out queries.
- AI Citations – the number of times the domain was selected as a cited source.
- AI Search Rank – the domain’s rank within the top-200 cited domains in our dataset based on how often it appeared in AI search results. A higher position means the domain surfaced more frequently during the model’s research process.
- AI Citation Rank – the domain’s rank based on how often it was cited. A higher position here means the domain was more frequently used as a source.
- Search-to-Citation Gap – the difference between a domain’s AI Search Rank and AI Citation Rank.
Importantly, we noticed that the LLM usually listed the same domains in both citations and search results in the same instance. However, this was not always the case, which is why these metrics should not be interpreted as a direct conversion funnel from search result to citation.
The Gap metric shows if a domain is over- or underrepresented in cited sources compared to how often it appears in search results. A negative gap means a domain is found often but cited less, while a positive gap means it is cited more than its search appearances suggest. So, this metric highlights the difference between domain performance in AI retrieval and citation.
Main patterns across AI search results and citations
One of the clearest patterns in the data is that AI citations are far more concentrated than AI search results.
Among the top 200 cited domains in our dataset, Wikipedia is the most important because it clearly illustrates the difference: it is not simply “popular” but the top attribution domain, ranking #1 as a cited source, while its AI search result rank is #106. Moreover, Wikipedia alone accounts for roughly 25.3% of all cited-source appearances on the list. The top 10 cited domains account for about 52.7% of all cited sources. In contrast, the top 10 domains in AI search results make up about 29.9% of all search result appearances. This shows that the model’s discovery is broad, but the citation layer is much narrower.
The gap becomes especially apparent when we look at domains that are much more prevalent in the discovery than in the citations. Pinterest is the most vivid example here, with a negative gap of 188, but the same pattern appears across several publishers and media domains, which we’ll explore in more detail later.
| Domain | AI Search Appearances | AI Citation Appearances | AI Search Rank | AI Citation Rank | Gap |
| pinterest.com | 2106 | 19 | 10 | 198 | -188 |
| foxsports.com | 807 | 40 | 44 | 189 | -145 |
| latimes.com | 1167 | 73 | 28 | 160 | -132 |
| foxnews.com | 1333 | 84 | 22 | 153 | -131 |
| fortune.com | 680 | 49 | 54 | 184 | -130 |
The opposite pattern shows that some domains appear less in AI search results but are cited much more often than their search visibility would suggest. AP News and Reuters are the clearest examples of citation authority outweighing AI search visibility. Reuters ranks only 197th in AI search but 19th in the list of cited sources. Similarly, AP News is ranking 200th in search appearances but 25th in citations.
| Domain | AI Search Appearances | AI Citations | AI Search Visibility Rank | AI Citation Rank | Gap |
| guinnessworldrecords.com | 225 | 173 | 198 | 77 | +121 |
| the-sun.com | 322 | 340 | 154 | 30 | +124 |
| homeguide.com | 250 | 321 | 192 | 33 | +159 |
| apnews.com | 117 | 394 | 200 | 25 | +175 |
| reuters.com | 231 | 438 | 197 | 19 | +178 |
Only a few domains are quite balanced and stable across both visibility stages, but these are rare. Most domains are either more visible as search results or as cited sources. Still, balanced domains are important because they succeed in both stages, and this combination is worth exploring and learning from.
| Domain | AI Search Appearances | AI Citations | AI Search Visibility Rank | AI Citation Rank | Gap |
| accio.com | 2511 | 770 | 6 | 10 | -4 |
| cbr.com | 791 | 235 | 46 | 50 | -4 |
| fandom.com | 6897 | 1293 | 2 | 6 | -4 |
| yahoo.com | 13903 | 1339 | 1 | 5 | -4 |
| bbb.org | 566 | 168 | 75 | 78 | -3 |
| alibaba.com | 6751 | 2771 | 3 | 3 | 0 |
| forbes.com | 4633 | 1559 | 4 | 4 | 0 |
| geeksforgeeks.org | 525 | 166 | 79 | 79 | 0 |
| inquirer.com | 289 | 62 | 172 | 172 | 0 |
| netflix.com | 341 | 91 | 148 | 148 | 0 |
| sfgate.com | 314 | 74 | 159 | 159 | 0 |
| economictimes.com | 304 | 70 | 166 | 165 | +1 |
| pbs.org | 371 | 101 | 136 | 135 | +1 |
| indeed.com | 1748 | 589 | 15 | 13 | +2 |
| jagranjosh.com | 288 | 64 | 173 | 170 | +3 |
Overall, the main insight here is that AI visibility drops sharply from search to citation. So, becoming a source that the model can confidently use in its answers is a major challenge.
As the next step, we’ll dive deeper into the main differences between domains by category.
Key search-to-citation differences by category
To understand what’s driving the gap between discoverability and actual visibility in AI answers, we grouped domains by category and calculated the median gap for each. We also counted the number of domains per group, then compared how often they appeared in AI search results versus how often they were cited. We used the median gap because it reduces the effect of outliers and gives a clearer view of what is typical in each category.
The table below not only shows which categories perform well, but also suggests what kinds of content AI tools seem to trust more and how they use different sources at the search and citation stages.
| Category | Domain count | Median Gap | Tendency |
| Encyclopedias, dictionaries, education & government resources | 23 | +19 | Citation-favored |
| Social, UGC, and review platforms | 26 | -6.5 | Mixed |
| News & media publishers | 57 | -11 | Retrieval-heavy |
| Local, travel, ecommerce & delivery | 10 | -32.5 | Retrieval-heavy |
| Advice in real estate, personal finance, and career | 19 | +39 | Citation-favored |
| Home, wellness & recipes | 21 | +42 | Strongly citation-favored |
| Entertainment and sports | 20 | -34 | Strongly retrieval-heavy |
| Other (mixed category) | 24 | +9.5 | Mixed |
The emerging pattern is pretty telling: LLMs tend to lean on structured, explanatory, and authoritative sources when they’re picking citations, but they reach for dynamic, user-generated, commercial, or entertainment-focused domains more often during retrieval. Put simply, the type of source and what it’s actually useful for matter a lot.
This tendency also aligns with our research data on prompt intent types. The leading intents are informational (64,482 total prompts) and commercial (12483), with commercial prompts triggering fan-out slightly more often than informational ones (50% vs 47%). Accordingly, domains offering content that satisfies these intents may be more likely to get cited.
As we zoom in on how the search-to-citation gap changes across individual categories, this gets even clearer.
Encyclopedias, dictionaries, education & government resources
Domain count: 23 | Median Gap: +19
The group of encyclopedias, dictionaries, education, and government resources leans heavily toward citations. These domains tend to define concepts, explain topics, and carry institutional weight, which seems to be exactly what AI models reach for when they need a source that feels stable and credible.
The clear standout is Wikipedia, which racked up more AI citations than any other domain in the dataset. Its broad coverage, structured format, and concise explanations make it almost tailor-made for citation use. Government and institutional domains also show positive gaps, suggesting that official resources can earn citation visibility even when they are not showing up most often in search.
| Domain | AI Search Appearances | AI Citation Appearances | AI Search Rank | AI Citation Rank | Gap |
| guinnessworldrecords.com | 225 | 173 | 198 | 77 | +121 |
| wikipedia.org | 421 | 17073 | 106 | 1 | +105 |
| rarest.org | 261 | 129 | 188 | 110 | +78 |
| nationalgeographic.com | 258 | 114 | 190 | 118 | +72 |
| harvard.edu | 283 | 134 | 176 | 105 | +71 |
| merriam-webster.com | 402 | 175 | 116 | 74 | +42 |
| iere.org | 492 | 238 | 88 | 49 | +39 |
| nps.gov | 313 | 109 | 160 | 124 | +36 |
| texas.gov | 301 | 101 | 167 | 136 | +31 |
| enviroliteracy.org | 951 | 610 | 37 | 12 | +25 |
| britannica.com | 1062 | 800 | 33 | 9 | +24 |
| biologyinsights.com | 1103 | 618 | 30 | 11 | +19 |
| dictionary.com | 372 | 109 | 135 | 123 | +12 |
| cornell.edu | 327 | 97 | 153 | 141 | +12 |
| ca.gov | 615 | 225 | 64 | 53 | +11 |
| collegevine.com | 316 | 92 | 156 | 147 | +9 |
| gameslearningsociety.org | 377 | 108 | 132 | 125 | +7 |
| jagranjosh.com | 288 | 64 | 173 | 170 | +3 |
| accountinginsights.org | 464 | 139 | 97 | 103 | -6 |
| nasa.gov | 396 | 105 | 123 | 131 | -8 |
| cambridge.org | 364 | 42 | 139 | 188 | -49 |
| collinsdictionary.com | 619 | 106 | 63 | 127 | -64 |
| fiveable.me | 426 | 45 | 104 | 186 | -82 |
In sum, this category shows that the strongest performers tend to be broad, structured, reference-oriented, and authoritative domains, while more specialized or narrower resources may be retrieved but are less often used for citations.
Social, UGC, and review platforms
Domain count: 26 | Median Gap: -6.5
One of the most interesting patterns here is that LLMs don’t treat all UGC domains equally The median gap is slightly negative, meaning the category is retrieval-heavy on average, but a dozen or so domains clearly overperform in citations.
Reddit stands out here with strong numbers on both sides: high retrieval visibility and very high citation visibility. Trustpilot, Consumer Reports, Rotten Tomatoes, Tom’s Guide, and TechRadar also perform well. These sites package user opinions, reviews, or product evaluations into clear, structured formats. They are not just collections of informal content; they often include ratings, summaries, rankings, and editorial framing, which likely makes them easier to cite.
| Domain | AI Search Appearances | AI Citation Appearances | AI Search Rank | AI Citation Rank | Gap |
| trustpilot.com | 293 | 211 | 170 | 58 | +112 |
| wired.com | 293 | 111 | 171 | 122 | +49 |
| knowyourmeme.com | 280 | 102 | 179 | 134 | +45 |
| reddit.com | 940 | 7972 | 39 | 2 | +37 |
| slashgear.com | 483 | 209 | 93 | 59 | +34 |
| techradar.com | 483 | 196 | 94 | 65 | +29 |
| consumerreports.org | 651 | 321 | 59 | 32 | +27 |
| rottentomatoes.com | 280 | 80 | 180 | 158 | +22 |
| tomsguide.com | 650 | 243 | 60 | 45 | +15 |
| yelp.com | 653 | 239 | 58 | 48 | +10 |
| geeksforgeeks.org | 525 | 166 | 79 | 79 | 0 |
| bbb.org | 566 | 168 | 75 | 78 | -3 |
| fandom.com | 6897 | 1293 | 2 | 6 | -4 |
| stackexchange.com | 318 | 70 | 155 | 164 | -9 |
| vocal.media | 450 | 114 | 100 | 117 | -17 |
| linkedin.com | 1242 | 239 | 27 | 47 | -20 |
| hubpages.com | 354 | 64 | 144 | 169 | -25 |
| medium.com | 1249 | 212 | 26 | 57 | -31 |
| tripadvisor.com | 813 | 131 | 43 | 108 | -65 |
| facebook.com | 2132 | 150 | 9 | 89 | -80 |
| youtube.com | 1305 | 131 | 23 | 107 | -84 |
| tumblr.com | 567 | 66 | 74 | 167 | -93 |
| instagram.com | 452 | 28 | 99 | 197 | -98 |
| scribd.com | 580 | 38 | 69 | 190 | -121 |
| answers.com | 753 | 55 | 49 | 178 | -129 |
| pinterest.com | 2106 | 19 | 10 | 198 | -188 |
Overall, user-generated platforms are good for finding information because they provide firsthand experiences, opinions, recommendations, and niche knowledge. However, this content can be fragmented, subjective, or hard to verify, so it seems more useful for context than as a main citation.
Platforms that organize user input into structured reviews or discussions are more likely to be cited. More visual, social, or informal sites usually stay in the background as retrieval sources.
News & media publishers
Domain count: 57 | Median Gap: -11
In the news category, there’s no single pattern either. The median gap is slightly negative, so the category is more retrieval-heavy, but, like in UGC, results vary. Some publishers are clearly citation-favored, some are balanced, and many are retrieved frequently but rarely cited.
This split indicates that citation behavior may depend on the publisher’s role. Wire services like Reuters and AP News are cited more often, likely because their concise, factual reporting provides the neutral, authoritative sources ChatGPT prefers. Another group that gets cited often is lifestyle and entertainment outlets like People, Vogue, Allure, Glamour, and Who What Wear. While this might seem unexpected, these sites provide specific, answer-ready information about people, products, trends, and releases. So for lifestyle or entertainment queries, they may be treated as relevant authorities.
| Domain | AI Search Appearances | AI Citation Appearances | AI Search Rank | AI Citation Rank | Gap |
| reuters.com | 231 | 438 | 197 | 19 | +178 |
| apnews.com | 117 | 394 | 200 | 25 | +175 |
| the-sun.com | 322 | 340 | 154 | 30 | +124 |
| people.com | 370 | 501 | 137 | 17 | +120 |
| businessinsider.com | 264 | 187 | 187 | 69 | +118 |
| nypost.com | 391 | 298 | 124 | 35 | +89 |
| allure.com | 271 | 134 | 184 | 106 | +78 |
| vogue.com | 361 | 197 | 140 | 63 | +77 |
| gbtimes.com | 306 | 146 | 165 | 95 | +70 |
| whowhatwear.com | 461 | 266 | 98 | 41 | +57 |
| glamour.com | 283 | 112 | 177 | 121 | +56 |
| gamesradar.com | 297 | 124 | 168 | 114 | +54 |
| axios.com | 489 | 292 | 89 | 37 | +52 |
| dexerto.com | 386 | 149 | 127 | 92 | +35 |
| bustle.com | 265 | 85 | 186 | 152 | +34 |
| time.com | 574 | 207 | 71 | 60 | +11 |
| stacker.com | 295 | 71 | 169 | 162 | +7 |
| economictimes.com | 304 | 70 | 166 | 165 | +1 |
| pbs.org | 371 | 101 | 136 | 135 | +1 |
| forbes.com | 4633 | 1559 | 4 | 4 | 0 |
| inquirer.com | 289 | 62 | 172 | 172 | 0 |
| sfgate.com | 314 | 74 | 159 | 159 | 0 |
| yahoo.com | 13903 | 1339 | 1 | 5 | -4 |
| indiatimes.com | 2055 | 549 | 11 | 16 | -5 |
| parade.com | 1159 | 319 | 29 | 34 | -5 |
| ndtv.com | 547 | 162 | 77 | 83 | -6 |
| prnewswire.com | 1870 | 427 | 14 | 21 | -7 |
| aol.com | 2744 | 549 | 5 | 15 | -10 |
| the-independent.com | 863 | 231 | 40 | 51 | -11 |
| businesswire.com | 313 | 58 | 161 | 176 | -15 |
| cnbc.com | 831 | 218 | 41 | 56 | -15 |
| houstonchronicle.com | 364 | 83 | 138 | 154 | -16 |
| theguardian.com | 1965 | 355 | 12 | 29 | -17 |
| nasdaq.com | 714 | 182 | 51 | 71 | -20 |
| investing.com | 377 | 80 | 133 | 157 | -24 |
| gamerant.com | 675 | 163 | 55 | 81 | -26 |
| thegamer.com | 397 | 89 | 122 | 149 | -27 |
| cbsnews.com | 2472 | 279 | 7 | 38 | -31 |
| hellomagazine.com | 465 | 106 | 96 | 128 | -32 |
| globenewswire.com | 402 | 88 | 117 | 151 | -34 |
| abc7.com | 316 | 35 | 157 | 191 | -34 |
| indianexpress.com | 335 | 49 | 150 | 185 | -35 |
| gamespot.com | 484 | 103 | 92 | 133 | -41 |
| aljazeera.com | 342 | 29 | 147 | 195 | -48 |
| thestreet.com | 489 | 92 | 90 | 145 | -55 |
| dallasnews.com | 349 | 16 | 145 | 200 | -55 |
| washingtonpost.com | 1506 | 162 | 17 | 82 | -65 |
| go.com | 517 | 88 | 83 | 150 | -67 |
| newsweek.com | 1251 | 143 | 25 | 97 | -72 |
| foxbusiness.com | 414 | 32 | 111 | 192 | -81 |
| patch.com | 497 | 61 | 86 | 173 | -87 |
| eonline.com | 576 | 55 | 70 | 179 | -109 |
| hindustantimes.com | 570 | 19 | 72 | 199 | -127 |
| independent.co.uk | 596 | 28 | 67 | 196 | -129 |
| fortune.com | 680 | 49 | 54 | 184 | -130 |
| foxnews.com | 1333 | 84 | 22 | 153 | -131 |
| latimes.com | 1167 | 73 | 28 | 160 | -132 |
To sum up, news content is often time-sensitive, repeated across outlets, or restricted by paywalls. Many publishers compete to report the same facts, especially on broad or fast-moving stories, so LLM’s citation funnel tends to pick sources based on reputation and factual reliability. Pages that are opinionated, localized, or focused on entertainment instead of giving clear factual answers seem less ideal for citation.
Local, travel, ecommerce & delivery
Domain count: 10 | Median Gap: -32.5
For local, travel, ecommerce, and delivery sites, the data shows a strong retrieval-heavy pattern. So, while they are helpful for discovering options, comparing availability, or finding local results, the LLM rarely uses them for final citations.
Our study findings may help explain this tendency. We found that prompts containing words related to food, travel, and local contexts are associated with a higher number of fan-out queries than average, likely because relevant information in these domains is highly dynamic, localized, and spread across many platforms.
At the same time, most of these platforms focus on transactions rather than explanatory content. Delivery apps, booking platforms, marketplaces, and map services are good for discovering what exists and where, like restaurants, products, prices, menus, or routes. As a result, ChatGPT appears to consult a wide pool of sources during retrieval to gather options, availability, and location-specific details, but it favors more stable and descriptive sources for citations.
For instance, AAA and Time Out are more editorial and offer travel advice, recommendations, and structured guidance, and they are cited more often. Interestingly, the example of Alibaba shows that a platform can achieve high AI visibility with enough scale and direct commercial relevance.
| Domain | AI Search Appearances | AI Citation Appearances | AI Search Rank | AI Citation Rank | Gap |
| aaa.com | 248 | 114 | 193 | 119 | +74 |
| timeout.com | 405 | 150 | 115 | 90 | +25 |
| alibaba.com | 6751 | 2771 | 3 | 3 | 0 |
| walmart.com | 485 | 142 | 91 | 100 | -9 |
| ebay.com | 338 | 52 | 149 | 181 | -32 |
| opentable.com | 360 | 60 | 142 | 175 | -33 |
| toasttab.com | 378 | 61 | 131 | 174 | -43 |
| ubereats.com | 407 | 70 | 114 | 163 | -49 |
| mapquest.com | 379 | 50 | 129 | 183 | -54 |
| doordash.com | 419 | 66 | 108 | 168 | -60 |
The main point is that platforms that help AI find options but are too focused on transactions and specific locations are often not on the main source list. This suggests that being useful for citation means being helpful for making choices, not making a purchase.
Advice in real estate, personal finance, and career
Domain count: 19 | Median Gap: +39
Resources that offer advice and guidance in real estate, personal finance, and career are clearly citation-favored. That makes sense because these topics are decision-heavy. People asking where to live, what salary to expect, or which financial option to pick are looking for explanations, comparisons, and practical guidance — and content on most of these sites is built exactly for that.
The strongest performers, such as WalletHub, Bankrate, NerdWallet, and Kiplinger, use rankings, calculators, comparisons, and recommendations to help users weigh their options. Real estate sites like Redfin, Realtor.com, Homes.com, HomeSnacks, and NeighborhoodScout combine location data with explanations, help users understand market conditions, neighborhood quality, affordability, and livability. This mix of data and interpretation is certainly useful for citations. Career and legal advice sites also perform well for similar reasons. Glassdoor, Salary.com, Indeed, FindLaw, and LegalClarity provide practical, structured information like salary ranges, employer reviews, legal explainers, and career guidance, and this kind of content can back up a specific claim.
| Domain | AI Search Appearances | AI Citation Appearances | AI Search Rank | AI Citation Rank | Gap |
| wallethub.com | 222 | 138 | 199 | 104 | +95 |
| redfin.com | 279 | 150 | 181 | 91 | +90 |
| kiplinger.com | 287 | 156 | 174 | 88 | +86 |
| homesnacks.com | 386 | 265 | 126 | 42 | +84 |
| realtor.com | 286 | 128 | 175 | 112 | +63 |
| bankrate.com | 329 | 149 | 152 | 93 | +59 |
| homes.com | 399 | 173 | 120 | 76 | +44 |
| nerdwallet.com | 589 | 408 | 68 | 24 | +44 |
| glassdoor.com | 440 | 198 | 101 | 62 | +39 |
| findlaw.com | 390 | 158 | 125 | 86 | +39 |
| neighborhoodscout.com | 356 | 112 | 143 | 120 | +23 |
| legalclarity.org | 1271 | 962 | 24 | 8 | +16 |
| salary.com | 984 | 410 | 36 | 23 | +13 |
| indeed.com | 1748 | 589 | 15 | 13 | +2 |
| zippia.com | 950 | 247 | 38 | 44 | -6 |
| gobankingrates.com | 495 | 128 | 87 | 111 | -24 |
| justia.com | 778 | 175 | 48 | 73 | -25 |
| bestneighborhood.org | 434 | 92 | 102 | 146 | -44 |
| fool.com | 524 | 105 | 80 | 130 | -50 |
The main takeaway is that real estate, finance, legal, and career content can have real consequences, so LLM tends to cite sources that present information clearly and help reduce uncertainty in important decisions.
Home, wellness & recipes
Domain count: 21 | Median Gap: +42
Home, wellness, and recipe content is generally practical, structured, and easy to connect to specific needs, which may explain its strong citation performance. Consider the questions we ask AI: how to fix something at home, what a symptom means, how long to cook something, or what to swap an ingredient with in a recipe. These questions need clear answers, and most domains in this category are designed for that, offering step-by-step instructions, guidance, explanations of treatments, cost estimates, and practical recommendations.
Wellness and medical content is especially citation-friendly, and citations are more important in health answers because the information can affect real decisions. So, LLMs appear to lean toward sources that are medically authoritative and well-organized, like CDC, Mayo Clinic, Cleveland Clinic, WebMD, Healthline, Drugs.com, and Medical News Today.
The retrieval-heavy exceptions are interesting as well. NIH, Eater, Mashed, The Daily Meal, and Good Housekeeping appear more in searches than in citations, but that doesn’t reflect their authority. For instance, NIH may simply get edged out when consumer-facing medical sites offer simpler, more direct answers.
| Domain | AI Search Appearances | AI Citation Appearances | AI Search Rank | AI Citation Rank | Gap |
| homeguide.com | 250 | 321 | 192 | 33 | +159 |
| cdc.gov | 244 | 166 | 194 | 80 | +114 |
| livestrong.com | 281 | 187 | 178 | 68 | +110 |
| homeadvisor.com | 243 | 146 | 195 | 96 | +99 |
| drugs.com | 233 | 125 | 196 | 113 | +83 |
| homesandgardens.com | 360 | 189 | 141 | 66 | +75 |
| medicalnewstoday.com | 381 | 185 | 128 | 70 | +58 |
| mayoclinic.org | 517 | 393 | 82 | 26 | +56 |
| chowhound.com | 261 | 101 | 189 | 137 | +52 |
| clevelandclinic.org | 602 | 434 | 66 | 20 | +46 |
| eatthis.com | 415 | 187 | 109 | 67 | +42 |
| tastingtable.com | 528 | 274 | 78 | 39 | +39 |
| webmd.com | 605 | 374 | 65 | 28 | +37 |
| chefsresource.com | 567 | 269 | 73 | 40 | +33 |
| healthline.com | 1003 | 1014 | 35 | 7 | +28 |
| angi.com | 707 | 293 | 52 | 36 | +16 |
| nih.gov | 2386 | 552 | 8 | 14 | -6 |
| mashed.com | 509 | 139 | 85 | 102 | -17 |
| eater.com | 670 | 173 | 56 | 75 | -19 |
| thedailymeal.com | 420 | 104 | 107 | 132 | -25 |
| goodhousekeeping.com | 521 | 130 | 81 | 109 | -28 |
The bottom line: home, wellness, and recipe content sits at the intersection of expertise and action. When people want answers they can use right away, the model rewards sources that make those answers clear, specific, and practical.
Entertainment and sports
Domain count: 20 | Median Gap: -34
Sports and entertainment domains are highly visible in AI search, but that visibility doesn’t translate into proportional citations. This is likely because these topics are more about discovery than evidence. People searching for entertainment or sports info are usually after a title, cast member, score, schedule, recap, or ranking. Many sources can provide this context, but the final answer does not need to cite all of them.
Sports websites show this clearly. ESPN, MLB, CBS Sports, NBC Sports, NFL, and Fox Sports have lots of stats, schedules, and recaps, and this gets them found often, but it does not help them earn more citations.
| Domain | AI Search Appearances | AI Citation Appearances | AI Search Rank | AI Citation Rank | Gap |
| netflix.com | 341 | 91 | 148 | 148 | 0 |
| cbr.com | 791 | 235 | 46 | 50 | -4 |
| comicbook.com | 310 | 62 | 162 | 171 | -9 |
| screenrant.com | 1339 | 335 | 21 | 31 | -10 |
| ncaa.com | 310 | 52 | 163 | 182 | -19 |
| collider.com | 822 | 196 | 42 | 64 | -22 |
| nme.com | 641 | 159 | 61 | 84 | -23 |
| as.com | 378 | 81 | 130 | 155 | -25 |
| si.com | 1495 | 239 | 18 | 46 | -28 |
| thewrap.com | 425 | 100 | 105 | 139 | -34 |
| sportskeeda.com | 734 | 158 | 50 | 85 | -35 |
| imdb.com | 1487 | 206 | 19 | 61 | -42 |
| statmuse.com | 343 | 31 | 146 | 194 | -48 |
| nfl.com | 399 | 53 | 121 | 180 | -59 |
| iheart.com | 794 | 114 | 45 | 116 | -71 |
| nbcsports.com | 400 | 31 | 119 | 193 | -74 |
| espn.com | 1907 | 142 | 13 | 99 | -86 |
| mlb.com | 566 | 69 | 76 | 166 | -90 |
| cbssports.com | 1022 | 72 | 34 | 161 | -127 |
| foxsports.com | 807 | 40 | 44 | 189 | -145 |
The main takeaway is that in fast-moving categories like sports and entertainment, being retrieved often means a site is broad and relevant, but being cited depends on offering unique evidence or a clearly attributable claim.
Other (mixed category)
Domain count: 24 | Median Gap: +9.5
This group includes a mixed set of domains across tech, automotive, religion, lifestyle, AI-related content, and other miscellaneous categories that do not fit cleanly into the main buckets. It shows a modest tilt toward citation over retrieval; however, the effect is relatively weak compared to strongly citation-favored categories like home and wellness or personal finance and other advice.
In summary, this group reinforces the broader pattern that citation is less about industry labels and more about content structure and extractability. Domains that provide clear, structured, and directly answer-ready information are more likely to be cited, while those with less structured or more heterogeneous content tend to remain within the retrieved results more.
| Domain | AI Search Appearances | AI Citation Appearances | AI Search Rank | AI Citation Rank | Gap |
| kbb.com | 253 | 142 | 191 | 101 | +90 |
| biblehub.com | 413 | 250 | 112 | 43 | +69 |
| sivo.it.com | 413 | 224 | 113 | 54 | +59 |
| luxwisp.com | 278 | 107 | 183 | 126 | +57 |
| caranddriver.com | 267 | 97 | 185 | 142 | +43 |
| cyberpost.co | 478 | 222 | 95 | 55 | +40 |
| aarp.org | 400 | 157 | 118 | 87 | +31 |
| microsoft.com | 426 | 178 | 103 | 72 | +31 |
| decentfoot.com | 279 | 81 | 182 | 156 | +26 |
| petscare.com | 783 | 425 | 47 | 22 | +25 |
| flavor365.com | 1080 | 485 | 32 | 18 | +14 |
| wellwisp.com | 314 | 95 | 158 | 144 | +14 |
| eathealthy365.com | 658 | 225 | 57 | 52 | +5 |
| accio.com | 2511 | 770 | 6 | 10 | -4 |
| droracle.ai | 375 | 99 | 134 | 140 | -6 |
| apple.com | 1671 | 384 | 16 | 27 | -11 |
| ahgautoservice.com | 511 | 143 | 84 | 98 | -14 |
| bible.com | 308 | 43 | 164 | 187 | -23 |
| autoevolution.com | 334 | 57 | 151 | 177 | -26 |
| achivx.com | 415 | 95 | 110 | 143 | -33 |
| wordpress.com | 1401 | 146 | 20 | 94 | -74 |
| statista.com | 626 | 100 | 62 | 138 | -76 |
| google.com | 707 | 105 | 53 | 129 | -76 |
| oreateai.com | 1098 | 121 | 31 | 115 | -84 |
Conclusion: AI search and AI citation visibility reward different things
Our analysis suggests that retrieval and citation are not just two stages of the same process. They appear to reward different types of content: retrieval rewards visibility, citation rewards usability, and authority. While retrieval appears to pull domains that are broadly indexed, popular, high-volume, or useful for discovery, ChatGPT’s source selection tends to favor domains that have reference-like, extractable, and trust-signaling content.
Read our full research for deeper fan-out analysis and prompt-level findings.
If you want to understand whether your brand is only being found by AI systems or actually cited in their answers, try DataForSEO AI Optimization API. It is designed for tracking AI search visibility, citations, and competitive presence across LLM responses.