Title: How ProcurePilot Increased Its Share of AI Voice from 8% to 54% with Abhord
Summary
- Company: ProcurePilot (fictional), a mid-market B2B SaaS for procurement workflows
- ICP: 200–2,000-employee companies in manufacturing and healthcare
- Goal: Be correctly recommended by LLMs for “purchase order automation,” “supplier onboarding,” and “three-way match” intents
- Timeframe: August 2025–January 2026
- Outcome: Share of AI Voice +46 points; accuracy of brand mentions +35 points; trials from AI-driven referrals +38%
1) The Initial Problem
By August 2025, ProcurePilot’s sales team noticed that buyers coming from AI assistants rarely mentioned the brand—despite strong SEO rankings. In controlled tests across five popular LLM surfaces:
- The brand was not mentioned in 72% of “best procurement automation” prompts.
- When mentioned, 39% of answers misattributed features (e.g., claiming no ERP integrations).
- Assistants frequently conflated the company with a similarly named legacy tool used for pilot procurement projects.
ProcurePilot’s web content was plentiful, but LLMs struggled with entity disambiguation and canonical facts. Traditional SEO improvements didn’t move the needle on AI answer visibility.
2) What Abhord’s Analysis Discovered
Using Abhord’s Audit, Entity Graph, and Answer Coverage modules, the team surfaced four root causes:
- Fragmented entity signals: Five variations of the company and product names appeared across docs, pricing pages, and marketplace listings. Abhord’s Entity Graph showed low “name stability” and inconsistent short descriptors.
- Sparse machine-verifiable facts: Feature claims lacked structured assertions (versioned dates, edition scope, and supported ERPs). Abhord flagged low “Claim Verifiability” scores and minimal corroboration from third-party sites.
- Coverage gaps in answer clusters: For high-intent prompts (“automate three-way match in NetSuite”), Abhord’s Coverage map showed zero robust passages written in assistant-friendly Q&A form.
- Citation gravity skewed off-site: Assistants pulled from forum threads and generic procurement explainers because ProcurePilot’s content lacked concentrated, citeable “source-of-truth” sections.
3) The Optimization Strategy They Implemented
Guided by Abhord’s Playbooks, ProcurePilot executed a six-week optimization program:
Entity clarity and stability
- Standardized the short descriptor: “ProcurePilot — procurement workflow automation for mid-market ERPs.”
- Consolidated brand variants and redirected stray product names to a canonical entity page.
- Added JSON-LD (Organization, Product) with stable identifiers, alternative names, and supported ERP attributes.
Machine-verifiable claims
- Created a Fact Sheet page with atomic, date-stamped statements (e.g., “Supports NetSuite, Microsoft Dynamics 365 (as of Oct 2025)”).
- Embedded edition scoping (“Starter includes 2 approval tiers; Growth includes unlimited”) and mapped each claim to internal docs and third-party corroboration (app marketplaces, integration directories).
Answerable content architecture
- Built 24 assistant-ready FAQs matching Abhord’s top “intent clusters,” each with explicit problem framing, concise answer, and a 3–5 sentence evidence block.
- Authored step-by-step How-To passages (“How to configure three-way match in NetSuite with ProcurePilot”) with declarative headings and scannable, short paragraphs.
Citation gravity and corroboration
- Orchestrated lightweight third-party corroboration: updated marketplace listings, refreshed partner pages, and seeded two neutral comparison resources with clear, non-promotional specs.
- Implemented Abhord’s Citation Mapper to ensure every high-value claim had at least two credible, crawlable confirmations.
Model-oriented formatting
- Adopted “short-first” paragraphs (<90 words), term stability (consistent noun phrases), and explicit negatives (“Does not require on-prem connectors”) to reduce hallucination risk.
- Added a Changelog stream with dated deltas so assistants could reference recent changes without re-parsing long releases.
Continuous monitoring and testing
- Set up Abhord’s Live Tests: 120 prompts spanning generic and brand-adjacent intents; weekly retests across multiple assistant surfaces.
- Built a red-team set of disambigu