The Top 100 Most-Cited Domains in AI Search (Q2 2026)

The finding

We ran 75 brand-neutral buyer-intent prompts through ChatGPT, Claude, Gemini, Google AI Overview, and Perplexity on 2026-05-18, extracted every cited URL from every response, and counted the result by root domain. The dataset contains 2,583 total citations spread across 1,119 distinct domains.

Three properties of the resulting distribution surprised us, and they each have direct strategic consequences.

First, the head of the distribution is small. The top 10 domains together carry only 8.1% of all citations. The top 25 carry 14.1%. The top 100 carry 32.8%. To cover 80% of cited-prompt volume you need 810 distinct domains. AI citation is a power law with a fatter tail than the Google-results distribution most SEO teams reason about.

Second, cross-engine consensus is rare. Only 12 of the 1,119 distinct domains (1.1%) appear in every engine's citation universe. The #1 domain by raw citations (youtube.com) appears in only 3 of the 5 engines. The "AI authority list" looks different depending on which engine the question is asked through.

Third, vendor first-party pages are scarce at the very top. The first nine ranks are aggregators, video, and forums. The first vendor domain (salesforce.com) does not appear until rank 10. When AI engines decide who to attribute, they reach more often for the reviewer than for the brand.

The top 25

Ranked by total citations across the full 375-response dataset, with ties broken first by distinct prompts cited and then by engine breadth. The "Engines" column counts how many of the 5 engines cited the domain at least once.

The shape of the top 25 is a single mixed set: a video platform (youtube.com) anchoring the list, business and tech press (Forbes, CNET, TechRadar, The Independent), niche publisher hubs (Healthline, Sleep Foundation, Outdoor Gear Lab, The Good Trade, Better Trail, CleverHiker, Upgraded Points, MySubscriptionAddiction), listicle content farms (Email Vendor Selection, Insider One, Venture Harbour, Email Tool Tester), review aggregators (G2, CRM.org), one forum (Reddit), one blog platform (Medium), institutional outlets (Consumer Reports, NCOA), and a single vendor first-party domain (Salesforce at rank 10, Grow Therapy at rank 13). Aggregators and niche publishers carry the top of the distribution; vendor first-party pages compete with them only in narrow categories.

The 12 domains cited by all five engines

These are the only domains that earned at least one citation from ChatGPT, Claude, Gemini, Google AI Overview, and Perplexity in the full dataset. They are the closest thing AI search has to a cross-engine canon. Note that this list excludes the most-cited domain by raw count (YouTube) and the most-cited aggregator (Forbes). High-citation and cross-engine-consensus are nearly orthogonal properties.

The shape of the cross-engine consensus set is different from the shape of the top 25. Niche publisher hubs dominate (Healthline, Outdoor Gear Lab, Sleep Foundation, CleverHiker, The Run Testers). Vendor first-party domains punch above their top-25 weight here (Salesforce, Monday.com, Klaviyo, Expedia, PostHog, Postmark). The set is heavily weighted toward category-specific authority: domains that own one vertical deeply rather than domains that span many verticals shallowly. Universal cross-engine authority is a small set, and it is earned through depth, not breadth.

The long tail is bigger than the head

The full citation distribution is more long-tailed than the Google-results distribution most SEO teams plan against. To carry 80% of cited-prompt volume, you need 810 distinct domains. To carry 90%+ you need most of the 1,119 unique domains in the dataset. The reason the tail is so fat is that 896 of the 1,119 distinct domains (80.1%) appear in exactly one prompt and were never cited again. Only 20 domains were cited in five or more prompts; only 2 in ten or more.

Of the 1,119 distinct domains across the dataset, 621 (55.5%) are cited by exactly one engine. The cross-engine drop-off is steep: by the time you get to "cited by 4 of 5 engines" the count falls to 87, and "cited by all 5 engines" is only 12 domains. Inside the top 100 the same property holds in compressed form: 9 top-100 domains touch all 5 engines, 52 touch 4, 30 touch 3, and 9 touch only 2.

The vendor question

The first 9 ranks are aggregators, video, and a forum. The first vendor first-party domain in the top 100 is salesforce.com, at rank 10. That is the single most important structural finding for a B2B SaaS marketing team reading this report.

Vendor first-party domains do appear in the top 100; in fact 43 of the 100 entries (43.0%) are vendor first-party. But none of them carry enough cross-prompt, cross-engine reach to crack the top 9. The reason is structural. Aggregators get cited because they answer multi-vendor comparison queries; vendor first-party pages get cited because they answer questions about that specific vendor. The first kind of question is far more common in a buyer-intent prompt set than the second.

The 43-vendor block inside the top 100 has its own shape worth reading, and two orderings sit on top of each other inside that block. By rank, the leading vendor first-party domains are Salesforce (10), Grow Therapy (13), GoodRx (33), and Gusto (34). By cross-engine breadth, six vendor first-party domains touch all five engines at least once: Salesforce, Monday.com (rank 46), Klaviyo (60), Expedia (70), and PostHog (89), plus Postmark from outside the top 100. Most other top-100 vendors are category-specific: PostHog in dev tooling, Grow Therapy in mental-health services, Custify and Avoma in narrow vertical SaaS niches. Vendor first-party citation in AI search looks more like deep category ownership than broad brand recognition.

Each engine has its own top 10

Aggregating across all 5 engines hides how little overlap the per-engine top lists have with each other. Here are the top 5 cited domains for each engine, ranked by distinct prompts cited.

The Gemini and Google AIO lists overlap almost completely (both inherit Google's search index and ranking). The other three diverge sharply: Claude leans on Healthline and Fortune, Perplexity leans on YouTube and niche category publishers like Email Vendor Selection, ChatGPT leans on tech press (TechRadar) plus institutional outlets (Consumer Reports, NIH.gov) plus a single vendor first-party (HubSpot). The Gemini-AIO axis is one cluster. Each of the other three engines is its own cluster.

The niche-publisher story

The most actionable surprise in the top 25 is the cluster of category-specific publishers that almost no one outside of their own vertical recognizes by name. Email Vendor Selection at rank 5, Insider One at rank 12, Venture Harbour at rank 18, Email Tool Tester at rank 25, plus product-category publishers like Better Trail (rank 20), CleverHiker (rank 15), MySubscriptionAddiction (rank 21), Outdoor Gear Lab (rank 7), The Run Testers and Sole Review (cited across all 5 engines), and Upgraded Points (rank 17).

These sites share a profile. Single-digit-thousand monthly traffic. Almost no consumer brand awareness. Owned by one or two operators rather than a media company. But each one owns one category deeply: structured listicle pages, tight scoring rubrics, regular updates, and a clear point of view. AI engines reach for that combination ahead of mainstream press when the prompt is "best X for Y." Getting included in three or four of the niche publisher hubs that own your category is, on this dataset, a higher-leverage AEO move than a single Forbes feature.

Ranks 26 to 100

The full top 100 below. Each row is one domain. The "Engines" column shows how many of the 5 engines cited the domain at least once; the "Type" column shows the lens collapsed from the 15-category classifier.

The first of each kind

The leader inside each of the eight lenses, for quick reference.

• Video: youtube.com (rank 1, 52 citations, 36 prompts)
• Aggregator: forbes.com (rank 2, 33 citations, 18 prompts)
• Forum / UGC: reddit.com (rank 9, 12 citations, 8 prompts)
• Vendor first-party: salesforce.com (rank 10, 12 citations, 6 prompts)
• Institutional: consumerreports.org (rank 11, 12 citations, 6 prompts)
• Blog platform: medium.com (rank 4, 17 citations, 9 prompts)
• Marketplace: chewy.com (rank 32, 8 citations, 3 prompts)

Implications for AI search strategy

Four conclusions follow from the data.

The category leaderboard is short and contested. Only 12 domains have universal cross-engine reach. Most of those 12 won that reach through deep category ownership, not broad brand recognition. The fastest path into that small set is to dominate one vertical thoroughly (one buyer category, deep structured content, consistent updates, clear point of view) rather than to chase mention surface area across many categories.

Aggregator inclusion outperforms aggregator imitation. 47 of the top 100 domains are aggregators. The realistic move for a vendor brand is not to compete with Healthline or Forbes on their own turf, but to be included in the listicles and category roundups those aggregators publish. The 47-aggregator block is a distribution channel, not a competitor set.

Vendor first-party still earns 43 of 100 slots, but only with depth. The vendors that show up in the top 100 are ones with deep, structured, comparison-ready product pages. The two that crack the overall top 25 (Salesforce at rank 10, Grow Therapy at rank 13) carry multi-decade SEO foundations or a category-defining institutional position. HubSpot is a separate case worth flagging: it does not crack the overall top 100, but it anchors ChatGPT's per-engine top 5, which is the kind of single-engine concentration the per-engine table above is built to surface. Building first-party AEO depth is harder than getting one aggregator placement, but it carries you across more engines once achieved.

The right per-engine playbook depends on which engine matters most for the audience. A team that cares mostly about Perplexity should invest in video, niche category publishers, and Consumer Reports placement. A team that cares mostly about Gemini and AIO should invest in Forbes, CNET, Medium, and Reddit presence. A team that cares mostly about ChatGPT should invest in tech press (TechRadar), institutional outlets (Consumer Reports, NIH.gov), and the brand's own first-party canonical pages. There is no single answer that wins in all five.

Caveats and limits

One response per (engine, prompt) cell is a small sample at the per-prompt level. The headline distributional findings (top 100 = 32.8% of citations, 12 all-5-engine domains, first vendor at rank 10) are robust against this because they aggregate across 75 prompts and 5 engines. The lower half of the top 100 (ranks 50 to 100) is more volatile run-to-run and should be read with a wide error band; the top 25 is much more stable. The prompt set is biased toward buyer-intent comparison queries by design (the parent benchmark was built to measure category authority, not general-knowledge retrieval); informational and definitional prompts would tilt the distribution toward institutional and encyclopedic sources. The data was collected on a single day (2026-05-18) and engine retrieval indexes drift; this artifact will be refreshed in Q3 2026.

Reproducibility

Every number on this page is derived from the same raw response JSONL used by the Q2 2026 reference benchmark, plus the hand-curated 15-category domain taxonomy and a deterministic aggregation script. Identical inputs produce identical outputs.

• Parent reference dataset: AI Search Citation Benchmark, Q2 2026
• Aggregated CSV (domain × engine × vertical × category): /research/citation-benchmark-2026-q2.csv
• Companion analysis (ChatGPT vs. AIO domain divergence): The Same Prompt, Two Different Webs
• Companion analysis (aggregator vs. vendor citation share): When AI Engines Cite the Reviewer vs. the Brand
• Companion analysis (buyer intent reshapes citations): Buyer Intent Reshapes AI Citations

Want to see how your own URL's AEO readiness scores against the structural patterns these top-100 domains share? Run a free Foglift Technical Audit. No signup required, 8 to 15 seconds end-to-end, same 8-dimension rubric Foglift uses internally.