Zeover Citation Intelligence Report: Social Citation Trends by Model

Social citations now shape how AI engines describe categories, products, and brands. Zeover tracks which social surfaces AI engines cite for monitored queries, then shows where content operations need stronger evidence. See the citation map.

Zeover’s Citation Intelligence Report for May 2026 shows a split that brand teams can no longer treat as background noise. Reddit, YouTube, and LinkedIn aren’t interchangeable social surfaces. Each model weights them differently, and May’s partial data through May 18, 2026 makes the routing more visible.

Most AI marketing analytics still stops at whether a brand appeared in an answer. This report goes one layer deeper: which social posts for GEO are being cited, which platforms are gaining or losing weight, and which query families appear to pull from each source.

TL;DR

• Reddit is the most consistent social citation surface across the May slice. It leads openai/gpt-5.4 at 47.7%, x-ai/grok-4 at 33.7%, and x-ai/grok-4.3 at 43.4%.
• YouTube is still the highest-volatility surface. perplexity/sonar-pro moved from a near-total YouTube concentration in March and April to a split May profile: 44.1% YouTube, 39.9% Reddit, and 9.7% LinkedIn.
• LinkedIn matters most when the model is resolving professional authority, company context, and category definition. openai/gpt-5.4 weighted LinkedIn at 41.1% in April before Reddit took the May lead.
• X/Twitter is explicitly retrievable for Grok, but citation weight isn’t the same as retrieval access. x-ai/grok-4 kept X/Twitter near the top across all measured months, while x-ai/grok-4.3 barely used it in May.
• Query promotion is mostly implicit. The CSV measures platform share by model-month, not individual query intent, so query-level routing below is an inference from platform mix, content type, and official retrieval behavior.

Method

The dataset is Zeover’s filtered social citation export for March, April, and May 2026. May is partial through May 18, 2026. The filter keeps only model-months with enough social citation volume to avoid reading noise as trend.

Official product documentation supplies the retrieval context. OpenAI’s ChatGPT Search launch note says search responses include links to web sources and rely on search providers and partner content. Perplexity’s Sonar documentation describes Sonar as web-grounded. X’s Grok help page says Grok can decide whether to search public X posts and run real-time web search. xAI’s X Search documentation goes further, describing keyword search, semantic search, user search, and thread fetch on X.

Those sources explain possible retrieval paths, not why a specific citation was selected. The percentage tables show what Zeover observed in the filtered model-months.

Model-Level Trend

openai/gpt-5.4

perplexity/sonar-pro

x-ai/grok-4

x-ai/grok-4.3

The chart uses the same percentage shares as the tables. Missing points mean the platform didn’t appear as a visible row in the filtered export for that model-month.

The stable pattern isn’t “social is rising” in the abstract. A sharper point is that social citation share is sorting by content type: Reddit owns community validation, YouTube owns explainers and demos, and LinkedIn owns professional authority. X/Twitter is strongest when a model explicitly has a live-social path, but the May x-ai/grok-4.3 slice shows that an available path can still receive little citation weight.

openai/gpt-5.4 moved from a LinkedIn-Reddit balance to a Reddit-led May. The YouTube share also climbed each month, from 3.2% in March to 10.4% in May. That doesn’t make YouTube the dominant OpenAI surface in this slice, but it does show that video-derived evidence is gaining room inside a model-month where Reddit still leads.

perplexity/sonar-pro changed the most. In March and April, YouTube effectively swallowed the social citation mix. May is different: YouTube still leads at 44.1%, but Reddit is close at 39.9% and LinkedIn appears at 9.7%. For a web-grounded answer engine, that shift matters. It suggests that the model is no longer resolving social evidence mostly through video pages in this filtered slice.

x-ai/grok-4 is the most balanced profile. Reddit holds roughly a third of social citation share each month, YouTube sits in the low twenties, and X/Twitter remains material while trending down. LinkedIn also gained share from March to May, which points to a broader evidence mix than the simple “Grok equals X” story.

x-ai/grok-4.3 complicates the X assumption even more. In May, Reddit leads at 43.4%, YouTube follows at 27.7%, LinkedIn reaches 14.5%, and X/Twitter is nearly absent at 0.2%. The model can access X Search, but the observed citations in this slice are routed elsewhere.

Query Promotion Map

The CSV doesn’t expose individual query text, so this section maps query families to the social surface that the model appears to weight for that intent. “Explicit” means official docs describe a dedicated retrieval path. “Implicit” means the platform share is high enough to behave like a preferred evidence surface. “None” means the observed share is too small to treat as a promotion signal.

Community validation, reviews, and “best” comparisons: implicit promotion. Reddit reaches 47.7% on openai/gpt-5.4 in May, 39.9% on perplexity/sonar-pro in May, and 43.4% on x-ai/grok-4.3 in May.

Tutorials, explainers, demos, and product walk-throughs: implicit promotion. YouTube reaches 98.4% and 96.4% on perplexity/sonar-pro before settling at 44.1% in May. x-ai/grok-4.3 gives YouTube 27.7% in May, and x-ai/grok-4 holds YouTube near a stable low-twenties weight across all measured months.

Professional authority, company context, and B2B category definition: implicit promotion. LinkedIn reaches 41.1% on openai/gpt-5.4 in April, 14.5% on x-ai/grok-4.3 in May, and 9.7% on perplexity/sonar-pro in May.

Real-time commentary, launch reaction, and debate tracking: explicit for Grok, implicit or none elsewhere. x-ai/grok-4 gives X/Twitter 27.2% in March and 20.3% in May, but x-ai/grok-4.3 gives it 0.2% in May.

Local reputation, community pages, and event context: implicit but secondary. Facebook reaches 7.7% on x-ai/grok-4.3 in May and 5.9% on x-ai/grok-4 in May.

Visual consumer proof and creator-led lifestyle content: none to weak. Instagram, TikTok, and Pinterest remain low-share across the filtered model-months, with isolated pockets but no durable model-wide routing.

In practice, query wording doesn’t promote a source by itself. Promotion appears when the query asks for evidence that a platform is structurally good at supplying: peer comparison on Reddit, long-form demonstrations on YouTube, and role-linked professional context on LinkedIn.

That makes the weighting operational. Social strategy shouldn’t allocate effort evenly across platforms. It should map each priority query family to the platform that the model already treats as credible evidence for that family.

Weight Interpretation

Reddit is the default promoted source for judgment queries. When a prompt asks which product, service, company, or category option is better, the May weights imply that Reddit has the strongest cross-model pull. The signal is implicit, not documented as a model rule, but it’s broad: every May model in the filtered set gives Reddit at least a third of social citation share.

YouTube is the promoted source for procedural queries. The strongest signal is perplexity/sonar-pro, where YouTube remains first in May after leading the prior two months. For OpenAI, YouTube is still a secondary surface, but the month-to-month direction is up.

LinkedIn is the promoted source for professional-identity queries. It’s strongest on openai/gpt-5.4 in March and April, then remains material across the May slices for all four models. That pattern points to implicit weight on named people, employer context, company pages, and category commentary.

X/Twitter has explicit retrieval support for Grok, but explicit access doesn’t guarantee citation weight. The split between x-ai/grok-4 and x-ai/grok-4.3 is the clearest warning in the report: real-time social content can be available to a model and still lose to Reddit, YouTube, and LinkedIn when the query wants evidence rather than recency.

Facebook is a secondary promoted source for local and community queries. The May weights aren’t high enough to make Facebook a first-priority GEO surface for most brands, but they’re high enough to matter for restaurants, venues, local services, events, and community-shaped reputation.

Instagram, TikTok, and Pinterest should be treated as weak or unpromoted for citation-first work in this filtered set. TikTok’s 8.5% May weight on openai/gpt-5.4 is the only remarkable pocket. It isn’t enough to justify a broad model-wide claim.

What Changed In May

May’s partial period shows two important moves.

First, Reddit strengthened across models. It moved into the lead for openai/gpt-5.4, stayed first for x-ai/grok-4, and led x-ai/grok-4.3 by a large margin. It also emerged as a serious second source for perplexity/sonar-pro after being absent from the visible March and April mix. That makes Reddit the broadest social evidence layer in this report, not just a forum artifact.

Second, YouTube lost monopoly behavior on perplexity/sonar-pro but stayed highly relevant. A drop from 96.4% to 44.1% looks dramatic, yet the May result still leaves YouTube first for that model. The change is diversification, not disappearance.

LinkedIn is the quieter movement. It doesn’t always lead, but it appears in the strongest professional slots: OpenAI in March and April, Perplexity in May, and both Grok models in May. For B2B brands, that’s enough evidence to treat LinkedIn as a citation surface, not just a distribution channel.

Engine Notes

openai/gpt-5.4 is using a trust split. LinkedIn and Reddit together account for most of the social citation share in every measured month. The model appears to prefer either professional authority or community validation depending on the month, with YouTube rising but still behind the top pair.

perplexity/sonar-pro is the volatility story. The model begins as YouTube-heavy, then diversifies in May into Reddit and LinkedIn. Because Sonar is documented as web-grounded, this looks less like a social-platform preference and more like a retrieval mix that can shift quickly when answer types or indexed sources change.

x-ai/grok-4 is balanced but still X-aware. Reddit leads while YouTube stays steady, and X/Twitter remains material enough to show that the explicit X path matters without becoming the only path.

x-ai/grok-4.3 is the caution case. Official xAI docs show that X Search exists, but the May citation mix doesn’t center X/Twitter. Reddit, YouTube, and LinkedIn take the weight. Any Grok strategy that only publishes to X misses the dominant observed citation surfaces in this slice.

Operating Implications

Brands should assign query families to social evidence surfaces.

The practical value of this citation intelligence is prioritization. A team deciding where to invest in social posts for GEO shouldn’t spread effort evenly across every channel. The model-month data shows which surfaces are already acting like evidence layers.

For community proof and comparison queries, Reddit deserves the first audit. The question is whether third-party discussions describe the brand accurately, use current product language, and compare the brand against the right category alternatives.

For tutorial and explainer queries, YouTube remains the highest-upside surface. The citation value comes from transcript-ready substance: clear titles, factual descriptions, chaptering, and demonstrations that answer the query without requiring a human viewer to infer the point.

For professional authority queries, LinkedIn needs named experts and company context. Company-page posts help, but posts from identifiable operators can carry the kind of role and employer context that AI engines can parse.

For real-time queries, X/Twitter is worth maintaining for Grok visibility, but the May data argues against over-investing there as a universal answer. X access is explicit in the docs. Citation dominance isn’t.

For local and consumer reputation queries, Facebook and Instagram are secondary but not irrelevant. They should be kept accurate and active where the brand’s category depends on place, community, visual proof, or events. TikTok and Pinterest don’t show enough weight in this filtered set to justify citation-first prioritization.

What To Measure Next

The next analysis should join model-month platform share to query labels. That would separate measured query promotion from inferred promotion and show whether Reddit’s May strength comes from review queries, commercial comparison queries, support queries, or a broader shift in retrieval. It’d also clarify whether x-ai/grok-4.3’s low X/Twitter share is temporary, query-mix driven, or a real citation-policy change.

For now, the finding is specific enough to act on: AI search optimization should be model-aware and query-aware. Reddit, YouTube, and LinkedIn carry the main citation load, but the order changes by engine and month.

Zeover turns this kind of model-month citation data into an operating view: which query families pull from which social surfaces, where the brand is absent, and which content investments are most likely to improve brand visibility in AI.