Original research · 92 engine checks across 24 questions, measured on 2026-07-27 · Methodology · Dataset
Executive summary
The single most striking finding from this study is a 100% same-day flip rate: all 10 of 10 same-day repeat groups changed their cited-domain set entirely within a single session. That instability is the central mistake brands make when treating AI citations as a stable placement — there is no stable placement. Any content strategy built on locking in a citation position is working against the medium's demonstrated behavior.
This study recorded 92 answer-engine checks across 24 active category questions on answer-engine optimization and AI search visibility, queried across three engines (ChatGPT, Gemini, Perplexity) on 2026-07-27. Zero of 92 planned calls failed. Every question produced at least one cited source on some engine, meaning the category is fully answerable by current AI systems — but which domains get credited shifts constantly.
Who gets cited. reddit.com leads the leaderboard, cited in 31 of 92 checks (33.7%, 95% CI 24.9%–43.8%). youtube.com ranks second at 21 of 92 checks (22.8%, 95% CI 15.4%–32.4%). blog.hubspot.com and linkedin.com share third at 20 of 92 checks each (21.7%, 95% CI 14.5%–31.2%). The editorial format dominates the source economy at 864 citations representing 89.5% of all citation share. Social, Reddit, YouTube, docs, and review sites together account for the remaining 10.5%, with review sites at the floor — just 4 citations, or 0.4%.
What structural patterns separate cited from uncited domains. The citation concentration is narrow: 15 domains account for all leaderboard positions, and the top domain's CI ceiling reaches only 43.8%, meaning no single domain has a commanding grip even at the upper confidence bound. Editorial content — structured, topically specific, independently published — is the dominant structural pattern across all three engines. Domains without an editorial footprint are functionally absent from the citation pool.
Engine behavior diverges sharply. ChatGPT returned sources in only 7 of 24 checks, averaging 1.9 sources per answer — the sparsest citation behavior of the three engines and a meaningful visibility gap for any domain relying solely on ChatGPT exposure. Gemini returned sources in all 24 of 24 checks, averaging 11 sources per answer. Perplexity returned sources in all 44 of 44 checks, averaging 14.9 sources per answer, making it the highest-volume citation engine in this sample. Domains appearing on Perplexity's leaderboard — reddit.com (24 appearances), linkedin.com (20), blog.hubspot.com (19), youtube.com (19) — reflect that volume directly.
The common brand mistake. The 100% flip rate on same-day repeats reveals that brands treating a single citation observation as proof of sustained visibility are measuring noise. Citation presence is probabilistic, not persistent. The structural pattern that recommends itself from the data is consistent editorial publishing across multiple platforms indexed by all three engines, not optimization of any single page or domain placement.
Key findings
- reddit.com led all domains with citations in 31 of 92 checks (33.7%, 95% CI 24.9%–43.8%) on 2026-07-27 across 24 category questions.
- Editorial content accounted for 864 citations — 89.5% of total citation share — making it the dominant structural format across all three engines.
- All 10 of 10 same-day repeat groups flipped their cited-domain set entirely, producing a 100% same-day variance-panel flip rate.
- ChatGPT returned sources in only 7 of 24 checks at an average of 1.9 sources per answer, versus Perplexity's 44 of 44 checks at 14.9 sources per answer.
- blog.hubspot.com and linkedin.com each appeared in 20 of 92 checks (21.7%, 95% CI 14.5%–31.2%), sharing third place on the citation leaderboard.
- Review sites registered only 4 citations out of 965 total — a 0.4% share — the lowest citation rate of any measured source type.
Findings
Citation Leaderboard
Across 92 recorded engine checks spanning 24 category questions and 3 engines (ChatGPT, Gemini, Perplexity), measured on 2026-07-27, citation share is concentrated at the top but drops off sharply beyond the first two positions.
Reddit.com leads all domains, cited in 31 of 92 checks (33.7%, 95% CI 24.9%–43.8%). This is the only domain to breach the one-in-three threshold and holds a meaningful gap over every other source. YouTube.com follows at 21 of 92 checks (22.8%, 95% CI 15.4%–32.4%). These two non-editorial domains occupy the top two positions — a notable structural pattern given that the source-type economy (see below) is otherwise dominated by editorial content.
The next tier is occupied by two domains tied at 20 of 92 checks each (21.7%, 95% CI 14.5%–31.2%): blog.hubspot.com, classified as editorial, and linkedin.com, classified as social. Their confidence intervals overlap substantially with youtube.com, meaning the gap between positions two through four is not statistically firm at this sample size.
authoritytech.io appears at 16 of 92 checks (17.4%, 95% CI 11.0%–26.4%), followed by hubspot.com at 15 of 92 (16.3%, 95% CI 10.1%–25.2%) and semrush.com at 14 of 92 (15.2%, 95% CI 9.3%–23.9%). All three are editorial. conductor.com reaches 13 of 92 checks (14.1%, 95% CI 8.4%–22.7%), also editorial.
A four-way cluster sits at 10 of 92 checks each (10.9%, 95% CI 6.0%–18.9%): searchengineland.com, siftly.ai, therankmasters.com, and ziptie.dev — all editorial. The confidence intervals for this cluster overlap with conductor.com above and with the final group below, so rank ordering within and around this band should be treated as approximate.
The leaderboard closes with otterly.ai and yotpo.com at 9 of 92 checks each (9.8%, 95% CI 5.2%–17.6%), and averi.ai at 8 of 92 checks (8.7%, 95% CI 4.5%–16.2%). All three are editorial.
The practical upshot is that the 15th-ranked domain is cited less than one-quarter as often as the first-ranked domain, and the confidence intervals at the bottom of the leaderboard are wide enough that individual rank positions carry limited precision.
Source-Type Economy
The citation economy is structurally skewed toward editorial content. Of all citations recorded across the 92 checks, editorial sources account for 864 citations, representing 89.5% of the total. Every other source type is marginal by comparison.
Social sources — principally linkedin.com — contribute 32 citations (3.3%). Reddit, counted separately from the social category, contributes 31 citations (3.2%), nearly matching social despite being a single domain. YouTube accounts for 21 citations (2.2%). Docs-type sources, which include official documentation and platform guides, contribute 13 citations (1.3%). Review sites account for the smallest recorded share: 4 citations, or 0.4% of the total.
The concentration finding is direct: nine of every ten citations in this category go to editorial-format content. Reddit and YouTube together account for roughly 5.4% of citations yet hold the top two positions on the domain leaderboard — indicating that while editorial volume dominates the aggregate, non-editorial formats can still achieve high per-domain citation rates when they are consistently referenced across questions.
Docs-type sources, despite covering authoritative platform material such as developers.google.com — which was the most-cited source on the "how to appear in Google AI Overviews" question — contribute only 1.3% of total citations. This reflects sparse coverage rather than low individual relevance; where docs sources appear, they often appear prominently, but they appear in few questions.
No source type outside editorial crosses the 4% threshold, and review sites at 0.4% are effectively at the margin of detectability in this 92-check sample.
Engine behavior
Measurement date: 2026-07-27. Sample: 92 recorded engine checks across 24 category questions and 3 engines.
The three engines tested — ChatGPT, Gemini, and Perplexity — exhibit substantially different citation behaviors across the same 24 category questions on answer-engine optimization and AI search visibility. These differences are not marginal; they reflect distinct retrieval architectures that have direct consequences for which domains gain visibility and which do not.
Source Volume Per Answer
ChatGPT returned the fewest sources by a wide margin. Across 24 checks, only 7 returned any sources at all, yielding an average of 1.9 sources per answer. The remaining 17 checks produced zero cited sources — a measured negative, not a data gap. These zero-source answers are findings in themselves: they represent questions where ChatGPT generated a response without attributing any external domain, making citation impossible regardless of content quality.
Gemini showed full source return: all 24 checks returned sources, averaging 11 sources per answer. Perplexity was the highest-volume engine across the largest check count — 44 checks, all returning sources, at an average of 14.9 sources per answer.
Domain Preference by Engine
Each engine favors a distinct domain set. ChatGPT's top cited domains across its limited source pool included ai.google, appvizer.co.uk, askvardi.ai, blog.hubspot.com, and cairrot.com, each appearing once. The narrow citation window means few domains appear at all.
Gemini's top cited domains were reddit.com (7 citations), authoritytech.io (6), hubspot.com (6), medium.com (5), and averi.ai (3). Discussion platforms and structured authority sites dominate this engine's citation pattern.
Perplexity cited reddit.com most frequently at 24 citations, followed by linkedin.com (20), blog.hubspot.com (19), youtube.com (19), and semrush.com (11). Perplexity's broader source appetite draws heavily from community platforms, professional networks, video content, and established SEO-adjacent publishers simultaneously.
No single domain dominated across all three engines. blog.hubspot.com and reddit.com appear on more than one engine's top list, but the overlap is limited — cross-engine citation consistency is the exception, not the rule.
Test-Retest Stability: The Flip Rate
The variance panel tested 10 same-day repeat groups to measure answer stability. All 10 of 10 repeat groups changed their cited-domain set — a 100% flip rate. This means no repeat query on the same day returned an identical set of cited sources. For brands attempting to measure or act on citation presence, this instability is a structural reality of how these engines currently operate. A domain appearing in one check cannot be assumed to appear in the next, even within the same day. Citation presence, as measured here, is probabilistic and volatile, not fixed.
What this means
Measured across 92 engine checks on 2026-07-27, the citation economy in the answer-engine optimization and AI search visibility category is heavily consolidated around a single content type: editorial. Editorial sources account for 864 citations, representing 89.5% of all recorded source attributions. Every other format — social (3.3%), reddit (3.2%), video (2.2%), docs (1.3%), and review sites (0.4%) — competes for the remaining 10.5% combined. For teams operating in this category, that distribution is the first structural signal worth internalizing.
What dominates citation lists in this category is structured, standalone editorial content: guides, explainers, and analysis pages hosted on dedicated domains. The top-cited editorial domains — blog.hubspot.com at 21.7%, authoritytech.io at 17.4%, semrush.com at 15.2%, and conductor.com at 14.1% — share a common pattern: they publish topic-specific pages that answer category questions directly, without requiring the engine to synthesize across multiple source types. Reddit achieves 33.7% citation frequency at rank one, but that reach reflects community discussion volume rather than a format teams can replicate directly. YouTube's 22.8% citation rate confirms that video transcripts and structured video descriptions can surface in engine responses, though video accounts for only 2.2% of total citation volume — meaning it reaches broadly but shallowly across questions.
The practical implication is straightforward: teams that have not yet published dedicated, structured editorial content on the core questions this category surfaces are operating without representation in the citation pool measured here. With the leaderboard showing that even the 15th-ranked domain, averi.ai, achieves an 8.7% citation rate across 92 checks, there is measurable open ground for domains that build topic-specific pages with clear answers, defined structure, and consistent crawlability.
Docs-format content (1.3% of citations) and review-site content (0.4%) are significantly underrepresented relative to editorial, suggesting these formats are less likely to be selected by engines when editorial alternatives exist. Teams relying primarily on product documentation or third-party review listings to generate AI visibility should treat those formats as supplementary rather than primary.
No format guarantees citation. But the data indicates that editorial content is the format class engines in this category consistently select, and teams without it are absent from a pool where presence is measured and unevenly distributed.
Methodology and data
This study was produced with a deterministic measurement instrument (92 planned engine calls, 92 recorded, 0 failed — recorded as negatives). The complete instrument description, sample, refusal conditions, and limitations are in the methodology; every underlying measurement is in the dataset.