# Citation Stability by Source Role: How Market Databases, Analyst Firms, and Editorial Media Hold AI Citations Differently

MRI data across 6,020 domains reveals that citation stability varies by source role — editorial media holds citations most consistently while market databases generate more volume. Each AI engine shows different role preferences.

Canonical URL: https://machinerelations.ai/research/citation-stability-source-role-patterns-ai-search-2026
Published: 2026-07-14
Research type: Legacy Research

## Source Body

Market databases generate the most AI citations per domain, but editorial media holds citations more consistently over time. Across 6,020 domains and 17,540 citation events measured by the Machine Relations Index, temporal consistency — how many days per month a source stays cited — varies sharply by source role. Each AI engine also shows measurably different role preferences, which means the stability of your citation depends on both *what type of source* carries the mention and *which engine* retrieves it.

## What Temporal Consistency Measures

The MRI's temporal consistency component (0–10 scale) tracks how many days within a 30-day window a domain appears in AI-generated answers. A source cited on 30 of 30 days scores 10. A source cited once scores near zero.

This is distinct from citation volume. G2.com received 145 citations in the latest 30-day measurement window — the highest among market databases — but its temporal consistency score is 8.7, not 10. Gartner.com received fewer total citations (130) but scored a perfect 10, meaning it appeared every single day of the measurement period.

Independent research confirms the instability that temporal consistency quantifies. [SISTRIX's 82,619-prompt study](https://www.sistrix.com/blog/ai-citation-drift-how-stable-are-sources-in-ai-search-results/) found that Google AI Mode replaces 56% of cited sources every week, while ChatGPT replaces 74%. [Trakkr Research](https://trakkr.ai/trakkr-research/citation-decay) tracked 10,000 brands across 10 months and found that 73.2% of cited URLs appear only once — the median half-life of an AI citation is 30 days. A [530,875-citation study by GetMentions](https://medium.com/@anirudh.ag/ai-citation-volatility-a-530-875-citation-study-4d4eeb151ce3) reached the same conclusion: "AI visibility is not a rank you hold. It is a probability."

The MRI adds a layer these studies do not: measuring stability *by source role* to identify which types of sources beat the baseline volatility.

## Source Role Stability Rankings

The MRI classifies every cited domain into one of nine source roles based on its operational function — not its reputation or domain authority. When we aggregate temporal consistency scores across all domains in each role, the hierarchy is:

| Source Role | Domains | Total Citations | Top-10 Avg Temporal Consistency | All-Domain Avg | Top-10 Citation Share |
|---|---:|---:|---:|---:|---:|
| Community/Social | 20 | 712 | 5.1 | 2.73 | 98% |
| Wire/Distribution | 9 | 111 | 2.4 | 2.40 | ~100% |
| Vendor-Owned | 579 | 3,179 | — | 1.01 | — |
| Editorial Media | 545 | 2,594 | 6.9 | 1.00 | 23% |
| Market Database | 307 | 1,617 | 6.0 | 0.92 | 37% |
| Academic/Government | 163 | 696 | — | 0.84 | — |
| Analyst Research | 235 | 987 | 5.1 | 0.81 | 40% |

Two patterns emerge. First, the gap between top-domain stability and all-domain stability is extreme in every role. Editorial media's top 10 domains average 6.9 temporal consistency while the full 545-domain set averages 1.0 — a 6.9x ratio. Market databases show a similar pattern: top 10 at 6.0 versus 0.92 across all 307 domains. This confirms what SISTRIX calls the ["core versus carousel"](https://www.sistrix.com/blog/ai-citation-drift-how-stable-are-sources-in-ai-search-results/) structure: a small set of stable sources surrounded by a rotating pool.

Second, editorial media's top sources hold citations more consistently than any other third-party role. TechRadar (temporal consistency: 9.0), Meltwater (8.0), and ITpro (8.0) sustain daily citations at rates comparable to community platforms like LinkedIn (10.0) and Reddit (9.7) — but editorial media achieves this across a much wider base. Its top 10 domains hold only 23% of the role's total citations, while community/social's top 10 hold 98%.

## Why Market Databases Lead on Volume but Trail on Stability

Previous [MRI analysis of source type authority](https://machinerelations.ai/research/source-type-authority-ai-search-mri-2026) established that market databases earn more citations per domain and achieve better average citation positions than analyst firms. The temporal data reveals the trade-off: market databases generate more citations but hold them less consistently than editorial media or community platforms.

The mechanism is structural. Market databases like G2 and Crunchbase serve structured, query-specific data — product comparisons, company profiles, funding records. AI engines retrieve this data when a user asks a specific product or market question, but the query distribution rotates. Today's query might be "Datadog vs Dynatrace," tomorrow's might be "ESG reporting software." The source gets cited on both days but for different queries, and when query rotation slows, citation volume drops.

Analyst research shows a different pattern. Gartner achieves a perfect temporal consistency of 10.0 — cited every day of the 30-day window — because its content maps to recurring query categories (enterprise software comparisons, market sizing, technology evaluation) that AI engines encounter daily regardless of the specific query. McKinsey and Deloitte score lower (6.0 and 7.0 respectively) because their content is more thesis-driven and less structurally query-mapped.

The concentration data reinforces this. [MRI citation concentration research](https://machinerelations.ai/research/market-database-ai-citation-concentration-2026) showed that citation pools follow a power law within every source role. The temporal consistency data adds a time dimension: the domains at the top of the concentration curve are also the most temporally stable, while the long tail is both low-volume and inconsistently cited.

## Engine-Specific Role Preferences

Each AI engine shows measurably different preferences for source roles. This is not a subtle signal — the differences are large enough to change which source types a brand should prioritize depending on where its audience searches.

**ChatGPT concentrates on editorial media.** TechRadar received 131 of its 138 total citations from ChatGPT alone (95%). Axios received 63 of 69 (91%). ChatGPT also dominates wire/distribution citations (59 of 111 total, or 53%). It largely ignores community platforms — LinkedIn and Reddit received zero ChatGPT citations despite being the two most-cited community sources overall.

**Google AI Mode favors community and social platforms.** LinkedIn received 123 of its 251 citations from Google AI Mode (49%). Reddit received 44 of 189 (23%). Google AI Mode is also the leading engine for market databases (359 total citations in the role) and contributes 234 citations to community/social — more than any other engine for that role.

**Gemini distributes broadly but leads on analyst research.** Gartner received 54 of its 130 citations from Gemini (42%). Gemini also leads market database citations (427 of 1,617) and editorial media citations (610 of 2,594). It is the most balanced engine across source roles.

**Perplexity favors vendor-owned sources and community platforms.** Perplexity is the top engine for vendor-owned sources (565 of 3,179) and the second-highest for community/social (179 of 712). It contributes zero citations to several high-profile editorial media sources (TechRadar, ITpro, Meltwater) — a stark contrast to ChatGPT's editorial concentration.

**Claude shows distinctive market-database affinity.** Claude gives Crunchbase 20 citations — the highest of any engine for that domain. It also leads on certain niche sources (Shadow.inc: 15 of 54, Medium: 12 of 139) that other engines underweight.

| Engine | Strongest Role | Notable Pattern |
|---|---|---|
| ChatGPT | Editorial Media | 95% of TechRadar citations; ignores community platforms |
| Google AI Mode | Community/Social | 49% of LinkedIn citations; leads market databases |
| Gemini | Analyst Research | 42% of Gartner citations; most balanced across roles |
| Perplexity | Vendor-Owned | Zero citations for several top editorial media sources |
| Claude | Market Database | Highest single-engine share for Crunchbase |
| Google AI Overviews | Distributed | No single-role dominance; favors LinkedIn (62) and Reddit (54) |

## What This Means for Machine Relations Strategy

The practical implication is that source role and engine coverage interact to determine citation stability. A brand mentioned on Gartner holds that citation with near-perfect consistency (temporal score 10.0) but reaches only five of six engines — Perplexity never cites Gartner. A brand reviewed on G2 reaches all six engines but with lower temporal consistency (8.7) because citation volume depends on query rotation.

Three principles follow from the data:

**Match source role to target engine.** If your buyers use ChatGPT, editorial media placements (especially in technology publications) will generate the most citations. If they use Google AI Mode, community platforms and market database listings are more valuable. Multi-engine visibility requires multi-role coverage.

**Distinguish volume from durability.** Market databases deliver higher citation volume per domain but editorial media delivers more consistent day-over-day presence. For brands optimizing for sustained visibility rather than peak citation counts, editorial placements compound more reliably.

**Accept that most sources are unstable.** Across all roles, the average domain holds citations inconsistently — all-domain temporal consistency averages range from 0.81 (analyst research) to 2.73 (community/social). Stability is the exception, concentrated in fewer than 10 domains per role. Building earned media across multiple stable sources in multiple roles is the only structural defense against citation drift.

## FAQ

### Which source type holds AI citations longest?

Community and social platforms (LinkedIn, Reddit, Medium) show the highest average temporal consistency across their full domain set (2.73), but the pool is small (20 domains). Among broader third-party roles, editorial media's top sources hold citations most consistently, with a top-10 average temporal consistency of 6.9 on the MRI's 10-point scale.

### Does citation volume predict citation stability?

Not reliably. G2 leads market databases with 145 citations but scores 8.7 on temporal consistency, while Gartner has fewer citations (130) but scores a perfect 10. Volume measures how many times a source is cited; temporal consistency measures how many days it stays cited. They capture different dimensions of citation authority.

### Why does ChatGPT ignore community platforms like Reddit?

ChatGPT's retrieval architecture favors editorially produced content — publications with structured articles, bylines, and distinct URLs. Community platforms organize content as threads and comments, which ChatGPT's source selection appears to deprioritize. Google AI Mode takes the opposite approach, heavily citing LinkedIn and Reddit content. This is a measurable architectural difference, not a quality judgment.

### How should brands use this data?

Identify which AI engines your buyers use, then invest earned media in the source roles those engines prefer. For ChatGPT-heavy audiences, prioritize editorial media placements. For Google AI Mode audiences, ensure your brand has strong community presence and market database listings. Track temporal consistency — not just citation count — to distinguish visibility that compounds from visibility that decays.

---

*Methodology: Data from Machine Relations Index v1.1, 6-engine measurement across Perplexity, ChatGPT, Gemini, Claude, Google AI Mode, and Google AI Overviews. 6,020 domains, 17,540 citation events, 30-day window ending July 2026. Source roles classified by deterministic taxonomy based on domain function, not reputation. Temporal consistency scored 0–10 based on proportion of measurement days with at least one citation. Last updated July 14, 2026.*

## Additional source context

- This guide provides practical guidance on how to prepare citable material and instruct the model to format citations effectively, using patterns that are familiar to OpenAI models. ([Citation Formatting | OpenAI API (developers.openai.com)](https://developers.openai.com/api/docs/guides/citation-formatting)).
- With the advent of Large Language Models (LLMs), this risk has intensified: LLMs are increasingly used for academic writing, but their tendency to fabricate citations (“ghost citations”) poses a systemic threat to citation validity. ([GhostCite: A Large-Scale Analysis of Citation Validity in the Age of Large Language Models (arxiv.org)](https://arxiv.org/abs/2602.06718)).
- May 2026 study — sentence-level grounding analysis, freshness, readability, and per-platform citation behavior. ([danishashko/ai-citation-patterns (github.com)](https://github.com/danishashko/ai-citation-patterns), 2026).
- [LLM Citation Drift: Why Citations Change and Vanish](https://brandmentions.link/llm-citation-drift) provides external context for AI citation stability source role patterns over time.
- [Public Benchmarks for Citation Accuracy in AI-Authored Papers — clawRxiv](https://clawrxiv.io/abs/2604.02008) provides external context for AI citation stability source role patterns over time.
- [How AI Decides Which Sources to Cite | Cite Solutions](https://cite.solutions/blog/how-ai-decides-which-sources-to-cite) provides external context for AI citation stability source role patterns over time.
- [AI Citation Patterns: How AI Engines Cite Sources (2026) | Geodocs.dev](https://geodocs.dev/reference/ai-citation-patterns) provides external context for AI citation stability source role patterns over time.
- [How Different AI Platforms Cite the Same Source Differently – ZipTie.dev](https://ziptie.dev/blog/how-different-ai-platforms-cite-the-same-source-differently) provides external context for AI citation stability source role patterns over time.
- [News Publisher Citation Share in AI Answers 2026 | Presenc AI](https://presenc.ai/research/news-publisher-citation-share-in-ai-2026) provides external context for AI citation stability source role patterns over time.

## Attribution

This research was produced by AuthorityTech, the first agency to practice Machine Relations. Machine Relations was coined by Jaxon Parrott.

## Machine-readable related links

### Related concepts

- [Machine Relations Index (MRI)](https://machinerelations.ai/glossary/machine-relations-index)
- [AI Visibility](https://machinerelations.ai/glossary/ai-visibility)
- [Machine Relations (MR)](https://machinerelations.ai/glossary/machine-relations)
- [RAG Citation (RAG)](https://machinerelations.ai/glossary/rag-citation)

### Supporting research

- [Vendor-Owned Content AI Citation Authority](https://machinerelations.ai/research/vendor-owned-content-ai-citation-authority-first-party-2026)
- [Market Databases vs Analyst Firms: Why Source Structure Predicts AI Citation Authority Better Than Brand Reputation](https://machinerelations.ai/research/market-databases-vs-analyst-firms-ai-citation-rankings-2026)
- [Source Type Authority in AI Search: Why Market Databases Outrank Analyst Firms in Answer Engine Citations](https://machinerelations.ai/research/source-type-authority-ai-search-mri-2026)
- [Fortune Business Insights Answer-Engine Citation Authority: Market Sizing Infrastructure Cited in 2.18% of AI Engine Runs](https://machinerelations.ai/research/fortunebusinessinsights-answer-engine-citation-authority-mri)

### Framework context

- [Machine Relations Stack](https://machinerelations.ai/stack)
- [Evidence Base](https://machinerelations.ai/evidence)