Vellum Organic Growth Opportunities

A massive programmatic expansion of model-to-model comparisons that go beyond head terms to include specific task-based performance and production readiness. These pages provide developers with a definitive guide on which model to choose for specific technical constraints.

Example Keywords

• "gpt-4o vs claude-3-5-sonnet for extraction"
• "llama-3-70b vs gemini-1-5-pro latency comparison"
• "best model for tool calling vs gpt-4 turbo"
• "structured output reliability claude vs openai"

Rationale

Vellum's current organic traffic is heavily concentrated in model selection behavior, with the leaderboard driving over 54% of traffic. Expanding into the long-tail 'vs' space captures high-intent developers at the exact moment they are making architectural decisions.

Topical Authority

Vellum already possesses significant authority in the 'model ranking' space; this play leverages that existing trust to dominate the more specific, technical comparison queries that competitors like LangChain currently capture.

Internal Data Sources

Use Vellum’s proprietary LLM Leaderboard dataset, historical model incident/status data, and SDK documentation for tool-calling and streaming implementation details.

Estimated Number of Pages

10,000+ (Covering thousands of model pairs across various modalities and task-specific modifiers)

A comprehensive directory of JSON schemas and validation rules for real-world business document extraction. These pages serve as a technical reference for developers building data extraction pipelines using LLMs.

Example Keywords

• "invoice json schema for llm extraction"
• "medical claim form data extraction schema"
• "bill of lading parser json format"
• "certificate of insurance extraction fields"

Rationale

Data extraction is a primary use case for Vellum's enterprise customers. By providing ready-to-use schemas, Vellum attracts developers looking for implementation-ready artifacts rather than just high-level advice.

Topical Authority

Vellum’s existing use-case tags for 'data-extraction' and 'data-transformation' provide a foundation for this technical authority, positioning the brand as an expert in structured LLM outputs.

Internal Data Sources

Leverage Vellum’s template gallery, existing extraction use-cases, and product documentation regarding JSON mode and tool-calling validations.

Estimated Number of Pages

25,000+ (Covering hundreds of document types across multiple industries and output standards)

A programmatic registry of Model Context Protocol (MCP) servers and tool manifests that allow agents to interact with third-party SaaS platforms. This directory provides the 'how-to' for connecting agents to the enterprise software stack.

Example Keywords

• "mcp server for jira integration"
• "model context protocol slack manifest"
• "mcp tool schema for salesforce"
• "how to connect ai agent to servicenow mcp"

Rationale

MCP is a rapidly emerging standard for agentic workflows. By owning the directory of these manifests, Vellum can capture the 'builder' audience early in the adoption cycle of agentic architectures.

Topical Authority

Vellum has already begun publishing content on MCP and agentic commerce; this play scales that early signal into a massive technical moat that competes with LangChain’s integration footprint.

Internal Data Sources

Utilize Vellum’s Workflows SDK documentation, existing partner integration lists, and community-sourced troubleshooting data for tool-calling.

Estimated Number of Pages

6,000+ (Covering hundreds of SaaS tools and their specific capabilities/actions)

A public library of task-specific evaluation datasets, regression test cases, and scoring rubrics. These pages provide the 'golden sets' that teams need to prove their LLM applications are production-ready.

Example Keywords

• "llm test cases for support ticket triage"
• "prompt regression dataset for summarization"
• "scoring rubric for rag chatbot accuracy"
• "golden dataset for invoice extraction"

Rationale

Reliability is the single biggest hurdle for enterprise AI adoption. Providing the testing infrastructure (test cases and rubrics) positions Vellum as the essential platform for moving from prototype to production.

Topical Authority

Vellum’s core product value is built on evaluation and monitoring; providing the datasets themselves is a natural extension of the brand's authority in AI reliability.

Internal Data Sources

Use Vellum’s evaluation product documentation, sample test suites, and insights from webinar transcripts regarding practitioner 'acceptance criteria'.

Estimated Number of Pages

15,000+ (Covering hundreds of tasks across various industries and difficulty levels)

Highly specific guides that recommend the optimal LLM for niche business tasks based on cost, latency, and accuracy constraints. These pages act as a 'buyer's guide' for technical product managers.

Example Keywords

• "best model for pii redaction in healthcare"
• "best llm for contract clause extraction"
• "best model for high-volume invoice processing"
• "most cost-effective model for customer support triage"

Rationale

While the general leaderboard attracts broad interest, use-case specific queries attract buyers with an active project. These pages convert searchers into Vellum users by offering a direct path to evaluation.

Topical Authority

Vellum’s existing benchmark analysis posts and industry-specific landing pages provide the necessary context to rank for these 'best model for' long-tail queries.

Internal Data Sources

Incorporate data from Vellum’s case studies, industry-specific templates, and internal benchmark data for specific tasks like summarization or extraction.

Estimated Number of Pages

2,000+ (Covering a matrix of business workflows and industry-specific constraints)

Improvements Summary

Restructure the LLM leaderboard cluster to answer “best LLM right now” queries immediately, with an indexable rankings table, visible “Last updated,” and a clear methodology section. Expand the coding and open-source leaderboards with benchmark-led tables, decision filters (cost, latency, license, VRAM), and tighter hub-and-spoke internal linking from /llm-leaderboard to key spokes.

Improvements Details

Add an “Answer First” block (40–70 words) to /llm-leaderboard, followed by a server-rendered HTML table and a changelog; build H2 sections for “Best LLM right now,” “LLM benchmarks leaderboard,” “Best LLM for coding,” “Best open source LLMs,” and “Cost vs quality.” Update /best-llm-for-coding with a true “llm coding leaderboard” table (HumanEval/MBPP/SWE-bench columns + overall score) plus “Best for debugging/review/refactoring/agentic SWE” callouts. Update /open-llm-leaderboard with license/commercial-use columns, hardware (VRAM) guidance, quantization, and fine-tuning support; add FAQPage + ItemList schema, and improve titles/meta around “llm leaderboard,” “llm rankings,” “best llm for coding,” and “best open source llm.”

Improvements Rationale

The target pages already align with high-volume, low-competition terms like “llm leaderboard” and “llm rankings,” but need clearer snippet-friendly formatting, stronger recency cues, and transparent ranking methods to win clicks and rankings. Indexable tables, FAQs, and schema help Google extract answers for “right now/latest/2025” variants, while deeper task-based sections and internal links strengthen topical authority across the cluster.