AEOForgedAEOForged
Get started

Original research · measured 2026-07-27

Reddit and YouTube Top 30% of Answer-Engine Citations on AEO/GEO Trust Questions, With a 100% Same-Day Domain Flip Rate Across Repeat Panels

Original research · 91 engine checks across 24 questions, measured on 2026-07-27 · Methodology · Dataset

Executive summary

Across 91 recorded engine checks spanning 24 category questions and 3 answer engines (ChatGPT, Gemini, Perplexity), measured on 2026-07-27, the single most striking finding is a 100% same-day citation-set flip rate: all 10 of 10 same-day variance-panel repeat groups returned a different cited-domain set than their baseline run. No repeat group held its source list stable even within the same day. That instability is the defining structural fact about how answer engines currently address trust and legitimacy questions for AEO/GEO services.

Who Gets Cited

Reddit.com leads the domain leaderboard, appearing in 28 of 91 checks (30.8%, 95% CI 22.2%–40.9%), followed immediately by YouTube.com at 27 of 91 checks (29.7%, 95% CI 21.3%–39.7%). LinkedIn.com ranks third at 21 of 91 checks (23.1%, 95% CI 15.6%–32.7%). The top two domains are user-generated platforms, not institutional or journalistic authorities, which sets the character of what answer engines surface when users ask whether AEO/GEO is a scam or a legitimate discipline.

Editorial sources dominate by volume: 690 citations, representing 84.8% of the total source economy. Social platforms (LinkedIn, Medium combined) account for 4.2% (34 citations), while reddit.com and youtube.com each hold distinct classification slots at 3.4% (28 citations) and 3.3% (27 citations) respectively. Review sites contribute only 5 citations (0.6%), and local directories contribute 1 citation (0.1%).

Engine-Level Divergence

The three engines behave in sharply different ways. ChatGPT returned sources in only 6 of its 24 checks, averaging 0.6 sources per answer — the lowest source density by a wide margin. Gemini returned sources in all 23 of its checks, averaging 9.9 sources per answer, with discoveredlabs.com (7 appearances) and reddit.com (6 appearances) as its top two cited domains. Perplexity returned sources in all 44 of its checks, averaging 13 sources per answer, led by youtube.com (25 appearances), reddit.com (22 appearances), and linkedin.com (21 appearances). One of 92 planned calls failed and is recorded as a negative in the dataset.

Implications for the Category

The combination of a 100% variance-panel flip rate and a citation leaderboard dominated by community platforms rather than verified industry bodies means that no domain can be considered a stable citation anchor for this topic category. Every question produced at least one source on at least one engine, so answer engines are not silent on AEO/GEO legitimacy — but the specific sources they surface are highly volatile. Practitioners or researchers seeking to appear in these citations should understand that presence in a single check does not indicate persistent visibility, and that editorial content across a wide range of domains currently comprises the structural majority of what answer engines draw from when addressing trust concerns in this category.

Key findings

  • All 10 of 10 same-day variance-panel repeat groups changed their cited-domain set, producing a 100% flip rate on 2026-07-27.
  • Reddit.com was the most-cited domain, appearing in 28 of 91 checks (30.8%, 95% CI 22.2%–40.9%), followed by YouTube.com at 27 of 91 checks (29.7%, 95% CI 21.3%–39.7%).
  • Editorial sources accounted for 690 citations (84.8%) of the total source economy across 91 checks, while review sites contributed only 5 citations (0.6%).
  • ChatGPT returned sources in only 6 of 24 checks, averaging 0.6 sources per answer, versus Perplexity's 44 of 44 checks at an average of 13 sources per answer.
  • Gemini's top cited domain was discoveredlabs.com with 7 appearances across 23 checks, while Perplexity's top cited domain was youtube.com with 25 appearances across 44 checks.
  • One of 92 planned engine calls failed, and LinkedIn.com appeared in 21 of 91 checks (23.1%, 95% CI 15.6%–32.7%) as the third-ranked domain overall.

Findings

Citation Leaderboard

Across 91 recorded engine checks on 2026-07-27 — spanning 24 category questions and 3 engines (ChatGPT, Gemini, Perplexity), with 1 of 92 planned calls failing and recorded as a negative — two domains effectively tied at the top of the citation leaderboard. reddit.com appeared in 28 of 91 checks (30.8%, 95% CI 22.2%–40.9%) and youtube.com appeared in 27 of 91 checks (29.7%, 95% CI 21.3%–39.7%). The overlapping confidence intervals indicate no statistically meaningful separation between these two leaders. linkedin.com ranked third at 21 of 91 checks (23.1%, 95% CI 15.6%–32.7%).

The editorial cluster begins at rank four. discoveredlabs.com was cited in 14 of 91 checks (15.4%, 95% CI 9.4%–24.2%), and coursera.org in 13 of 91 checks (14.3%, 95% CI 8.5%–22.9%). Three domains — blog.hubspot.com, forbes.com, and yesoptimist.com — each appeared in 12 of 91 checks (13.2%, 95% CI 7.7%–21.6%), again with indistinguishable confidence intervals. medium.com followed at 11 of 91 checks (12.1%, 95% CI 6.9%–20.4%).

The lower portion of the top 15 is tighter still. aiadvantageagency.com was cited in 9 of 91 checks (9.9%, 95% CI 5.3%–17.7%). blitzmetrics.com, semrush.com, and siteimprove.com each appeared in 8 of 91 checks (8.8%, 95% CI 4.5%–16.4%). bighuman.com and memetik.ai each closed the leaderboard at 7 of 91 checks (7.7%, 95% CI 3.8%–15.0%). Every domain from rank 4 onward carries intervals that overlap substantially with adjacent ranks, meaning the ordering within that group should be treated as approximate rather than definitive.

A notable structural feature of the leaderboard is the gap between the top three domains — all community or social platforms with citation rates of 23%–31% — and the editorial cluster beginning at rank four, which ranges from roughly 8% to 15%. The top three are also the only domains whose lower Wilson bounds exceed 15%, marking them as reliably cited across repeated sampling rather than sporadically appearing.

Source-Type Economy

The source-type breakdown reveals a heavily skewed citation economy. Editorial sources — including blogs, agency sites, course platforms, and industry publications — accounted for 690 of total citations, representing 84.8% of all citation share. No other source type came close. The social category (which includes linkedin.com and medium.com) contributed 34 citations (4.2%). The docs category contributed 29 citations (3.6%). reddit.com as a standalone source type contributed 28 citations (3.4%), and youtube.com contributed 27 citations (3.3%).

Review sites represented a marginal presence at 5 citations (0.6%), and local directory sources appeared just once across the full sample (0.1%). Together, all non-editorial source types combined for 15.2% of citations, leaving editorial sources with the remaining 84.8%.

This concentration has a practical implication for interpreting the leaderboard: reddit.com and youtube.com rank first and second by domain despite belonging to source types that collectively account for only 6.7% of all citations. This apparent contradiction resolves when the counting unit is considered — domain-level citation counts measure how many of the 91 checks returned at least one citation from that domain, whereas source-type citation totals aggregate every individual citation link across all responses. Editorial sources dominate total citation volume because they appear in large numbers per response; reddit.com and youtube.com surface frequently across questions but typically contribute a smaller number of individual links per response.

Four of the 24 category questions recorded at least one engine check that returned zero sources, and the one failed call is carried as a negative throughout all calculations. These null results are part of the measured record and are reflected in the per-question denominators used throughout the study.

Engine behavior

Measurement date: 2026-07-27. Sample: 91 recorded checks across 24 category questions and 3 engines; 1 of 92 planned calls failed and is carried as a recorded negative throughout.

Source yield per engine

The three engines behaved in markedly different ways when answering the 24-question AEO/GEO trust-and-legitimacy inventory.

ChatGPT completed 24 checks but returned sources in only 6 of them, producing an average of 0.6 sources per answer. Eighteen of its 24 answers delivered zero cited domains — a measured negative, not an omission in recording. When sources did appear, no domain was cited more than once; the full cited set across all 24 checks consists of academy.wcoomd.org, brandpush.co, builtin.com, crklr.com, and en.wikipedia.org, each appearing exactly once. ChatGPT's near-zero citation yield makes it the least transparent engine in this sample for source attribution on this category.

Gemini completed 23 of its planned checks (the one failed call accounts for the missing 24th) and returned sources on all 23, averaging 9.9 sources per answer. Its most-cited domain was discoveredlabs.com at 7 appearances, followed by reddit.com (6x), with cjr.org, coursera.org, and semrush.com each appearing 4 times. Gemini's per-answer yield is roughly 16 times higher than ChatGPT's, and every completed check produced at least one citation.

Perplexity ran 44 checks — the largest per-engine count in the study — and returned sources on all 44, averaging 13 sources per answer. Its leading cited domain was youtube.com at 25 appearances, followed by reddit.com (22x), linkedin.com (21x), blog.hubspot.com (12x), and forbes.com (10x). Perplexity's citation volume exceeds Gemini's in both raw count and per-answer average.

Domain preferences by engine

The citation pools share minimal overlap. Reddit.com is the only domain appearing in both Gemini's and Perplexity's top-five lists (6x and 22x respectively), and no domain from ChatGPT's sparse cited set appears in either of the other engines' top lists. This divergence means the source landscape a user encounters depends heavily on which engine they query.

Test-retest stability

The variance panel ran 10 same-day repeat groups. All 10 changed their cited-domain set between repetitions, producing a 100% flip rate. No repeat group preserved an identical source list. For practitioners or researchers attempting to track which domains answer engines surface for this category, this finding indicates that a single query snapshot does not reliably represent what any engine will cite on a subsequent same-day check. Every question produced at least one source on some engine across the full study, but the identity of those sources cannot be treated as stable.

What this means

The citation economy recorded on 2026-07-27 across 91 checks sends a clear structural signal: editorial sources dominate the reference pool that answer engines draw from when users ask about AEO/GEO trust and legitimacy. Editorial content captured 690 citations — 84.8% of the total — while social content (LinkedIn and Medium combined) accounted for 4.2%, and Reddit and YouTube each contributed roughly 3% despite holding the top two spots on the domain leaderboard by raw check count.

That apparent paradox is worth unpacking. Reddit and YouTube lead on breadth — cited in 28 and 27 of 91 checks respectively — but they represent single domains drawing from fragmented, uncontrolled content. Editorial domains collectively cover far more surface area. No single editorial domain owns this category: the highest-ranked editorial source, discoveredlabs.com, appeared in only 14 of 91 checks (15.4%), and the confidence interval on that figure runs from 9.4% to 24.2%. The domain leaderboard flattens quickly; five editorial domains share citation rates between 13.2% and 15.4%, all with overlapping confidence intervals. That compression suggests genuine open ground rather than an entrenched authority holding unassailable position.

For teams operating in this category, the data points toward a specific implication: editorial formats — articles, guides, structured explainers — are the content type answer engines most consistently pull from when fielding trust-and-legitimacy questions about AEO/GEO services. Social posts and user-generated forum content are present but marginal in citation share despite their domain visibility. Review sites and local directories together represent less than 1% of citations across the entire sample.

The absence of a dominant editorial authority also means no single source has locked in the narrative on whether AEO/GEO is legitimate or not. Of the 24 category questions tested, multiple questions produced zero-source answers from at least one engine — a recorded negative that confirms some areas of the query space remain genuinely unserved by cited content.

Teams building editorial assets in this space may find it worth prioritizing the specific question clusters where cited-source density is lowest — that is where answer engines have the least material to draw from and where a well-structured, factually grounded piece of content has the most room to become part of the reference set. No outcome is guaranteed, but the structural gap is measurable and, based on this sample, currently unfilled.

Methodology and data

This study was produced with a deterministic measurement instrument (92 planned engine calls, 91 recorded, 1 failed — recorded as negatives). The complete instrument description, sample, refusal conditions, and limitations are in the methodology; every underlying measurement is in the dataset.