Seekde Research Methodology

What Seekde Research Methodology Means

Seekde is an independent research publication and technical exploration project analyzing artificial intelligence search engines, answer engines, and generative discovery systems.

In an emerging field often marked by speculative ranking formulas, research integrity requires methodological transparency. This methodology defines Seekde’s editorial standard, establishing explicit rules for gathering, classifying, and publishing evidence across our corpus.

To maintain analytical clarity, Seekde enforces an operational separation across nine categories of statements:

• DOCUMENTED_FACT: Technical specifications established by official platform documentation or recognized standards bodies.
• DIRECT_OBSERVATION: Direct interface observations recorded at a specific timestamp under disclosed test conditions.
• TEST_RESULT: Quantitative metrics derived from controlled test sets using disclosed calculation formulas.
• INFERENCE: Logical deductions linking documented facts or observations together, explicitly labeled as analytical interpretations.
• HYPOTHESIS: Plausible explanations or emerging models requiring further controlled testing before being accepted as factual.
• RECOMMENDATION: Practical guidance provided to search practitioners, framed around risk profiles and current evidence.
• OPINION: Editorial value assessments or subjective viewpoints, clearly distinguished from empirical findings.
• ILLUSTRATIVE_EXAMPLE: Sample datasets, demonstration outputs, or mock explanatory values used strictly for illustration and never presented as measured evidence.
• PLANNED_NOT_LIVE: Future product concepts or roadmap tools explicitly distinguished from current preview capabilities.

This methodology governs our research investigations, controls for output volatility, guides tool evaluations, and ensures data integrity. It operates alongside our institutional Source & Citation Policy and Editorial & Corrections Policy, with technical architectural foundations established in our monograph on Generative Answer Engine Architecture.

---

Evidence Types We Publish

Seekde categorizes editorial coverage into four distinct evidence classes:

Evidence Class	Core Scope	Required Evidence Base	Labeling Requirement
Desk Research	Conceptual frameworks and educational overviews of AI search.	Synthesis of credible industry sources and foundational retrieval concepts.	Standard editorial format without experimental claims.
Documentation-Based Analysis	Technical evaluations of search crawlers, protocols, and platform policies.	Direct citations to official developer documentation and protocol specifications.	Explicit source attribution to Tier 1 documentation.
Hands-On Testing	Evaluations of commercial SEO toolkits, generative interfaces, or software workflows.	Direct practitioner interaction under an active account, recorded inputs, observed outputs, and limitation logging.	Labeled as Hands-On Review or Direct Workflow Test.
Original Research	Empirical studies evaluating prompt volatility, citation patterns, or model visibility.	Structured sample design, controlled execution protocols, complete datasets, and calculation methods.	Labeled as Original Research Study with methodology appendix.

---

The Four-Tier Editorial Source Hierarchy

To ensure consistency across our publications, Seekde evaluates cited references against a four-tier source hierarchy defined in our Source Policy:

• Tier 1 (First-Party Documentation & Official Standards): Official developer documentation and protocol standards (e.g., Google Search Central, OpenAI Developer Documentation, Schema.org, W3C) are authoritative for crawler user-agents, directive support, and declared platform features.
• Tier 2 (Academic & Benchmark Research): Peer-reviewed papers and formal benchmark preprints (e.g., arXiv, ACM Digital Library) provide theoretical and empirical grounding for information retrieval mechanics, neural reranking, and generative engine optimization models.
• Tier 3 (Credible Independent Industry Reporting): Established journalism and specialist search publications (e.g., Search Engine Land, Search Engine Journal, Digiday) are used for industry reporting, executive interviews, and market trends.
• Tier 4 (Practitioner Experiments & Community Observations): Practitioner case studies and community discussions help identify emerging anomalies and generate hypotheses.

Source Governance Rule: Official documentation remains authoritative for declared platform behavior. Credible empirical observations may document discrepancies, but unverified anecdotes cannot overturn primary or academic evidence.

---

Platform and AI-Search Testing Protocol

A central methodological challenge in generative search research is output volatility. The GEO study (Aggarwal et al., 2023) runs experiments across five random seeds and reports averages to reduce variance in results. Seekde therefore uses repeated observations rather than treating one generated response as an enduring ranking fact.

The Single Observation Rule

A single AI search response is an isolated observation, not a stable ranking fact.

Seekde forbids generalizing sweeping conclusions from a single prompt execution. Research studies must evaluate patterns across structured observation windows.

Test Record Parameters

When recording observations from conversational AI search platforms, researchers document available observable conditions:

1. Timestamp (UTC): Date and time of submission.
2. Market & Locale: Country, language, regional settings.
3. Session Context: Logged-in or private session state.
4. Prompt Text: Literal query characters submitted.
5. Prompt Variant: Query intent classification.
6. Sample Distribution: Number of runs within the window.
7. Platform Surface: Engine interface evaluated.
8. Model Identifier: Version tag displayed in interface.
9. Observed Output: Response text, truncation, or errors.
10. Attribution: Mentions, citations, links, recommendations.

Seekde records observable interface conditions and does not assert undisclosed internal model parameters as factual certainties without first-party documentation.

---

Controlling Comparative Evaluations

When conducting comparative evaluations, researchers enforce structured test controls:

• Identical Query Sets: Competing entities are evaluated against the exact same prompt set.
• Bounded Observation Windows: Tests use bounded observation windows to reduce time-related confounding across comparative runs.
• Disclosed Regional Settings: Queries use consistent and disclosed market, language, and regional settings.
• Session Contamination Precautions: Tests are conducted in clean, isolated browser profiles as an experimental precaution to reduce potential session or query history carry-over.

---

First-Party Evidence Types and Visual Governance

Seekde recognizes ten acceptable first-party technical evidence types:

1. Genuine product and platform screenshots.
2. Screen recordings of verified workflows.
3. Sanitized production server access logs.
4. Captured API responses where permitted.
5. Structured query output records.
6. Structured CSV and JSON datasets.
7. Verified repository source-code inspection.
8. Verified live product implementation state.
9. Original analytical charts derived from real datasets.
10. Documented researcher test notes.

Screenshot and Server Log Sanitization

Screenshots must depict genuine interfaces; security challenges, interstitials, blank windows, or errors are rejected as invalid evidence. All screenshots and logs are sanitized to redact passwords, API keys, tokens, PII, and account identifiers.

Crawler server log analyses record observation windows, verified user-agent tokens, requested paths, HTTP status codes, and relevant headers, while removing unnecessary IP addresses and sensitive query strings.

Data, Charts, and Illustrative Visuals

Original charts and tables must remain traceable to an underlying dataset, calculation logic, date range, units, and aggregation method. Non-empirical or mock visualizations must be visibly labeled as an Illustrative example.

AI-Generated Image Policy

In accordance with Seekde’s standing visual policy, any materially AI-generated public image must display the reader-visible caption: Image generated by AI. Alt text describes visual content and does not substitute for visible disclosure. AI-generated imagery is strictly illustrative and must never serve as evidence that a platform produced a result, a tool capability exists, or a benchmark ran.

---

AI-Assisted Text and Editorial Responsibility

Seekde may utilize artificial intelligence tools for drafting, text organization, summarizing notes, and planning. AI output is never treated as factual evidence. Factual statements require verification against documentation, tests, or datasets. Editors maintain responsibility for truthfulness, verification, and publication approval.

---

Source-to-Claim Matching and Conflicting Evidence

Material factual statements in Seekde publications are evaluated against three verification statuses:

• SUPPORTED: The cited source directly substantiates the specific factual claim made in the text.
• PARTIALLY_SUPPORTED: The source provides related context, but does not fully verify all aspects of the claim without additional analysis.
• NOT_SUPPORTED: The source does not substantiate the claim. Material factual claims in this state must be corrected, reclassified as inference or opinion, or removed prior to publication.

Seekde prohibits circular self-citations: target URLs or internal draft assertions are never cited as proof of external platform facts.

Handling Conflicting Evidence

When credible sources or real observations disagree:

1. Researchers record divergent statements or observations from each source.
2. Contextual factors, collection dates, and software versions are documented.
3. Plausible technical reasons for discrepancies are identified where evidence supports them.
4. Seekde transparently states which interpretation is adopted while preserving unresolved uncertainty.

---

Tool-Review and Original-Research Standards

Seekde enforces rigorous protocols before publishing commercial tool evaluations or empirical research.

Tool Review Standard

A publication is labeled a Hands-On Review only after direct testing under an active account. Reviews document subscription tiers, test dates, inputs, observed outputs, workflow captures, limitations, pricing snapshots, and unreviewed features. Guides lacking hands-on verification (Maps 49–55) remain held under HOLD_FOR_HANDS_ON status or publish strictly as documentation overviews.

Original Research Standard

Empirical studies labeled Original Research require a defined research question, sample construction, collection procedure, datasets, failure handling, calculation formulas, limitation disclosures, and reproducibility notes. Pipeline studies (Maps 57–60) remain held under HOLD_FOR_DATA status until empirical collection and calculations are complete.

---

Procedural Reproducibility and Reader Checklist

Seekde does not use exact word-for-word output repetition as its reproducibility criterion. Instead, we use procedural reproducibility: disclosed test conditions, exact prompts, run counts, observation windows, preserved outputs, and documented exclusions.

Seekde provides complete protocol transparency so independent researchers can replicate the methodology:

Raw observation outputs, structured data files, query logs, and test notes are preserved to support editorial audit, verification, and refresh workflows, consistent with legal and privacy requirements.

---

Commercial Independence and Editorial Separation

Seekde may include advertising, sponsored content, affiliate relationships, or commercial partnerships where permitted by site policy.

Commercial relationships never lower the evidentiary standards applied to factual editorial claims. Payment or sponsorship is never treated as evidence of platform performance or product quality. All commercial and sponsored materials must be clearly disclosed in accordance with our Sponsored Content Policy and Editorial Disclosure.

---

Product Claims vs. External Editorial Claims

Seekde maintains a strict distinction between internal product claims and external ecosystem claims:

• Seekde Product Claims: Statements regarding Seekde capabilities must strictly reflect verified repository code and accepted product truth documentation.
• External Ecosystem Claims: Statements regarding third-party search engines, AI models, or crawlers require Tier 1 documentation, controlled empirical testing, or clearly labeled analytical inference.

---

Material Limitations of AI Search Research

All research published by Seekde is subject to inherent technical limitations:

1. Single-Run Limitation: Seekde treats one generated response as one observation. Claims about recurring patterns require repeated observations under disclosed conditions.
2. Contextual & Market Variance: Seekde records locale, account/session state, query context, platform surface, and observation time as test conditions. Differences observed between runs are recorded without attributing them to a particular hidden platform mechanism unless supporting evidence exists.
3. Undocumented Platform Shifts: When an observed output changes and the technical cause is not documented by the platform, Seekde records the difference without attributing it to proprietary ranking, retrieval, or generation parameters.
4. Sample Size Boundaries: Studies reflect finite sample sizes and specific observation windows; they should not be extrapolated into universal algorithmic laws.
5. Correlation vs. Causation: Technical commonalities among cited sources demonstrate an association, not proof of a direct ranking factor.

---

Editorial Freshness, Updates, and Corrections

To keep research accurate, Seekde operates defined review cycles:

• Class A (30–60 days): Rapidly changing platform features, crawler user-agent tokens, and commercial tool interfaces.
• Class B (3–6 months): Foundational research guides, methodology specifications, and architectural overviews.
• Class C (6–12 months): Evergreen conceptual analyses and historical overviews.

Material changes to official platform documentation, crawler tokens, or core protocols prioritize immediate editorial review.

Corrections Workflow

When a material factual error is identified, editors re-verify the claim against primary sources and test records, apply the correction, and document revision context under our Corrections Policy.

---

What This Methodology Does Not Claim

Seekde maintains full transparency regarding our operational status:

• Seekde is NOT an automated SaaS tracking dashboard: We do not operate an enterprise backend running continuous automated rank tracking or daily multi-engine scraping.
• Seekde does NOT calculate proprietary ranking scores: We do not market arbitrary authority scores or synthetic AI visibility metrics.
• The Seekde Homepage is an Interactive Preview: The explorer interface on the Seekde homepage is a demonstration preview using curated sample datasets to illustrate discovery concepts. It is not an active multi-engine search API.

---

Research-Method References

Seekde grounds its research methodology in primary technical specifications and published literature:

• Aggarwal, P., et al. (2023). GEO: Generative Engine Optimization. arXiv:2311.09735: Benchmark study evaluating source attribution under generative optimization, reporting averages across five random seeds to reduce variance.
• Google Search Central. AI Features and Your Website: Official documentation establishing that AI Overviews and AI Mode are Google Search features, specifying that link eligibility requires standard Search indexing and snippet eligibility, with crawling governed by Googlebot and robots.txt.
• OpenAI Developers. Overview of OpenAI Crawlers: Official developer documentation establishing that OAI-SearchBot is used to surface websites in search results in ChatGPT’s search features, while GPTBot is used for content that may be used in training foundation models.

---

Where to Read Next

• Explore our organization in What Is Seekde? and About Seekde.
• Review our publication standards in the Editorial Policy and Source Policy.
• Learn about error handling in the Corrections Policy.