Industry research

Foglift Research

Original, fully-sourced research on AI search visibility: quarterly benchmarks, citation studies, and query-fanout anatomies. Free to read, free to cite, with downloadable data.

Flagship reportUpdated 2026-08-17

AI Search Citation Benchmark: Q3 2026

Across 375 responses to 75 frozen buyer-intent prompts, five AI engines produced 1,510 distinct cited domains. Of the 92 domains in the combined engine top-25 lists, 75% appeared in only one engine, and none appeared in all five.

Foglift sent the same 75 brand-neutral buyer questions across 25 verticals to ChatGPT, Claude, Gemini, Google AI Overview, and Perplexity from 2026-08-01 through 2026-08-10 using live web retrieval. Cited URLs were normalized to domains and counted once per response; the frozen Q2 prompt set was retained, and quarter-over-quarter differences remain observational because provider models and indexes changed.

AI answers · citations · fan-outs · crawlers

Read the flagship report

Turning the research into an operating plan? Start with the AI search visibility software comparison, use the AI search monitoring workflow to track the same engines in production, then review monitoring cadence and engine coverage before choosing a plan. For an engine-by-engine source taxonomy, read the citation content-types report. For an evidence map of where AI engines already cite Foglift, use the Foglift citation map.

Source layer

Research studies

Engine comparisons, citation benchmarks, and methodology studies behind the flagship report.

2026-08-19 · v2026-Q2-to-Q3 · Q2 to Q3 2026 (the same 75 questions across five engine lanes; 375 answers per quarter)

AI Engine Citation Drift 2026: How Fast Do AI Sources Change?

Foglift's frozen 75-question benchmark found 77.0% to 86.7% same-question citation-set turnover across five AI engines from Q2 to Q3 2026.

For each frozen prompt and engine, Foglift compared the Q2 and Q3 cited-domain sets with Jaccard similarity, excluded empty-to-empty pairs, and reported one minus the mean as turnover. Provider models and search indexes changed, so the result measures observed source drift rather than a single causal mechanism.

2026-08-04 · v2026-08 · 2026-07-17 to 2026-08-04 UTC (854 Technical Audits; latest eligible result for each of 194 distinct domains)

AI Readiness Across 194 Websites: The Median Site Scores 60/100

Across 854 Technical Audits using Foglift's current scoring contract, the latest score for each of 194 distinct domains has a median AI Readiness Score of 60. Median SEO on the same sites is 90. The 30-point gap shows why a clean SEO baseline still needs an AI extraction layer.

Foglift aggregated 854 public and authenticated Technical Audits that used the current AI Readiness field from July 17 through August 4, 2026. URLs were normalized to hostnames and deduplicated to the latest eligible result for each of 194 domains. The report compares AI Readiness and SEO scores stored on the same audit; domain names, URLs, account identifiers, and individual results are excluded.

2026-08-01 · v2026-08 · Publisher-excluded public benchmark. Original panel: 2026-05-26 to 2026-06-10 (13 unbranded buyer-intent prompts x 5 AI engines; 59 usable answers). Current source-layer panel: 2026-07-18 to 2026-07-31 (9 high-intent prompts; 343 answers across 5 engines).

Which AI Search Tools Do ChatGPT, Perplexity, Gemini & Google AI Recommend?

Among the 15 most-cited source domains, 1,061 citation URL occurrences split 59% across AI visibility vendor domains and 41% across off-vendor platforms and publishers.

This publisher-excluded benchmark monitored 13 unbranded buyer-intent questions across five AI engines, then analyzed 343 answers from July 18 through July 31 for source-domain and engine patterns. Source counts are citation URL occurrences; publisher observations, the publisher domain, and Google's redirect wrapper are excluded from public rankings while answer and engine denominators remain intact.

2026-07-21 · v2026 · 2026-03-20 onward, revalidated daily

Generative Engine Optimization Statistics: 2026 Research Report

More than 24,000 successful production AI answers reveal how citation inclusion, source density, query fan-out, and crawler purpose differ across the generative search ecosystem.

Foglift Research publishes anonymized aggregates from production AI answers, workspace-scoped query fan-outs, and classified crawler requests. Every statistic includes its exact current sample and date window.

2026-06-20 · v2026-06 · 2026-06-16 to 2026-06-20 UTC (1,373 production monitoring rows; 533 latest prompt-engine rows; 62 complete five-engine prompt sets)

AI Engines Agree on Brands More Than Sources

Across 1,373 post-fix production monitoring rows and 62 complete five-engine prompt sets, Gemini and Google AI Overview were the closest pair: 98.4% brand-mention agreement and 0.643 citation-domain overlap. Most other engine pairs still agreed on the brand answer more than 90% of the time, while their cited source domains diverged sharply.

Foglift aggregated anonymized production geo_results after the Perplexity false-negative fix and the ChatGPT / Claude citation extraction backfill. The study keeps the latest row per workspace, prompt, and engine, then measures brand-mention agreement and cited-domain Jaccard overlap across complete five-engine prompt sets. No customer, workspace, prompt, or response text is published.

2026-05-24 · v2026-Q2 · 2026-05-18 (375 buyer-intent responses across 75 prompts × 5 engines; 2,583 citations, 1,430 structurally classified)

Five AI Engines, Five Content Diets: A Q2 2026 Citation-Type Breakdown

Across 1,430 structurally classified citations from five production AI search engines, ChatGPT cites the vendor's own first-party site 68% of the time. The other four engines run 46 to 52%. Only Perplexity cites video meaningfully (9.7%, almost entirely YouTube). Community discussion (Reddit, Quora) appears in zero ChatGPT, Claude, or Perplexity citations and only 1 to 2% of Gemini and Google AI Overview citations.

Derived from the Q2 2026 AI Search Citation Benchmark. Each cited URL is tagged with one of fifteen structural domain categories defined in the Day 1 domain taxonomy (vendor first-party, niche publisher hub, listicle / content-farm, business press, tech press, lifestyle media, institutional, video, community UGC, review aggregator, marketplace, personal blog, developer platform, search-engine cache). Hand-classification covered the top-volume domains; the long-tail unclassified portion is reported transparently. Per-engine percentages use the engine's classified subset as denominator. Surfaces the engine-level content-type lens that the Day 2 binary aggregator-vs-vendor view collapsed away.

2026-05-22 · v2026-Q2 · 2026-05-18 (375 buyer-intent responses across 75 prompts × 5 engines)

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

Across 375 buyer-intent responses from ChatGPT, Claude, Gemini, Google AI Overview, and Perplexity, the top 100 domains account for 32.8% of all citations. The other 67.2% come from a long tail of 1,019 domains. The #1 domain (YouTube, 52 citations, 36 prompts) is cited by only 3 of the 5 engines. Only 12 domains across the entire dataset are cited by all 5.

Derived from the Q2 2026 AI Search Citation Benchmark. Flat domain frequency count across all 375 responses, ranked by total citation count with ties broken first by distinct prompts cited then by engine breadth. Each top-100 entry is annotated with the 15-category taxonomy label, per-engine prompt counts and citation counts, per-intent prompt counts, and top three vertical exposures. The aggregator-vs-vendor lens reuses the parent benchmark's taxonomy and adds first-class lenses for community UGC (Reddit, forums), video (YouTube), institutional (Consumer Reports, .gov, .edu), marketplace, blog platforms, and dev platforms. Surfaces the canonical AI-search authority list and the structural long-tail of the citation distribution.

2026-05-21 · v2026-Q2 · 2026-05-18 (75 buyer-intent prompts × 5 engines; this artifact pairs the ChatGPT and Google AI Overview responses)

ChatGPT vs. Google AI Overview: The Same Prompt, Two Different Webs

Across 75 buyer-intent prompts the average Jaccard overlap between ChatGPT's and Google AI Overview's cited domain sets is 4.1%, and 64% of prompts share zero cited domains at all. The two engines are running on almost completely disjoint slices of the web even when answering the identical question.

Derived from the Q2 2026 AI Search Citation Benchmark. For each of 75 prompts we pair the ChatGPT and Google AI Overview responses and compute domain-set Jaccard similarity, intersection size, and engine-exclusive domains. We also break the per-prompt overlap down by buyer intent and category, and tally global engine-exclusive domains (domains that appear in one engine's universe but never in the other's). Surfaces the structural divide between two production AI-search products built on the same web.

2026-05-19 · v2026-Q2 · 2026-05-18 (375 buyer-intent responses across 75 prompts × 5 engines)

Buyer Intent Reshapes AI Citations: Discovery, Shortlist, and Variation Cite Different Webs

When the buyer's intent shifts from discovery to shortlist to variation, the set of cited domains turns over almost completely. Across 25 verticals the average pair-wise overlap is just 13.4%, and only 2.2 of a typical 50.8-domain union appears under all three intents.

Derived from the Q2 2026 AI Search Citation Benchmark. For each vertical the cited-domain union is computed for discovery, shortlist, and variation prompts independently, then Jaccard similarity is computed between each intent pair within the vertical and averaged across all 25 verticals. A per-intent aggregator-vs-vendor lens reuses the parent benchmark's domain taxonomy. Surfaces three different domain personalities AI engines reach for at each funnel stage.

2026-05-18 · v2026-Q2 · 2026-05-18 (375 buyer-intent responses across 75 prompts × 5 engines)

AI Search Citation Benchmark, Q2 2026

Q2 2026 reference data: 375 buyer-intent responses across 75 prompts × 5 engines. 1,119 distinct cited domains.

Reference dataset: 75 brand-neutral buyer-intent prompts across 25 verticals (tech SaaS, consumer services, CPG/retail) sent to ChatGPT, Claude, Gemini, Google AI Overview, and Perplexity using each engine's grounded production model with live web search. Cited URLs were extracted, normalized to root domains, and aggregated into reference tables (overall top 25, top 10 per engine, top 5 per vertical, citation-type distribution, URL-depth distribution, cross-engine Jaccard, long-tail concentration).

2026-05-18 · v2026-Q2 · 2026-05-18 (375 buyer-intent responses across 75 prompts × 5 engines)

The Foglift AI Citation Map: Four Authority Patterns in Q2 2026

Every industry falls into one of four AI citation patterns. A fifth quadrant, Vendor Hegemony, is empty by design.

Analytical paper introducing the four-pattern framework (Contested Lead, Editorial Capture, SEO Capture, and Open Frontier) that classifies how AI search citation authority is distributed across 25 industries. Interprets the Q2 2026 benchmark data into a leader-vs-challenger strategy matrix for each pattern.

2026-05-18 · v2026-Q2 · 2026-05-18 (375 buyer-intent responses across 75 prompts × 5 engines)

When AI Engines Cite the Reviewer vs. the Brand: A 25-Vertical Split

CPG / retail responses cite the vendor's own site only 22.7% of the time; tech SaaS responses cite it 92.7%, a 70-point gap. Aggregator share runs the opposite direction (80% for CPG, 44.7% for SaaS), reshaping how a brand should think about AI search authority.

Derived from the Q2 2026 AI Search Citation Benchmark. The 15 structural citation-type categories are collapsed into a binary aggregator-family vs. vendor-first-party lens and broken down by category (tech SaaS, consumer services, CPG / retail), all 25 verticals, three buyer intents, and five engines. Shows the largest single structural divide in the dataset: a 70-percentage-point gap between tech SaaS and CPG / retail in how often the vendor's own first-party domain is cited.