Abhord Quickstart: A Practical Product Guide for New Users (2026 Edition)
Who this guide is for and what’s new
This refreshed edition (January 2026) is for marketers, product teams, and analysts who want reliable, repeatable visibility into how large language models “talk about” your brand and category. Since the last version, Abhord added:
- Multi-model orchestration with cost caps and scheduling
- Improved entity resolver for cleaner, deduplicated brand mentions
- Sentiment v2 with context-aware polarity handling
- Share of Voice (SOV) v2 with model- and traffic-aware normalization
- Competitor Packs and saved survey templates per workspace
- Audit trails and export options (CSV/JSON) for compliance and BI
1) Initial setup and configuration
- Create a workspace
- Invite teammates with roles (Viewer, Editor, Admin). Use SSO if available.
- Set your default region and timezone so scheduled runs align with your reporting cadence.
- Connect model providers
- Add API keys for the providers you use (e.g., OpenAI, Anthropic, Google, Mistral, managed Meta). Set hard and soft spend caps to protect budgets.
- Toggle models you want available for surveys. Keep at least one proprietary and one open-source model enabled for diversity.
- Define entities (brands, products, people)
- Add your brand with canonical name plus variants (misspellings, acronyms, legacy names). Example: “Acme Cloud,” “Acme,” “ACMECloud,” “AcmeCloud.”
- Map competitors and key features (e.g., “Acme Atlas,” “Starter Plan”). This powers better disambiguation and feature-level sentiment.
- Create projects and tags
- Group surveys by initiative (e.g., “Q1 PM Tools Landscape”). Add tags like “EN-US,” “SMB,” or “DevTools” for filtering later.
Pro tip: Start with a clean entity list. Most downstream noise (false positives) comes from missing variants or ambiguous names. Abhord’s resolver will flag overlap; accept/merge suggestions during setup.
2) Run your first survey across LLMs
- Pick a survey template
- Start with “Category Discovery” or “Brand Preference.” Templates include recommended prompts, sampling, and guardrails.
- Define questions
- Ask 3–5 tightly scoped questions. Example:
1) “List the top project management tools for startups and explain why.”
2) “When would you recommend Acme Cloud vs. ClickUp or Asana?”
3) “What are common criticisms of Acme Cloud for SMB teams?”
- Avoid leading language; use neutral phrasing to reduce bias.
- Choose models and sampling
- Select 3–6 models to balance breadth and cost.
- Set per-model samples (e.g., 30–50 generations per question). More samples reduce variance but increase spend; 200–300 total generations usually stabilize SOV.
- Configure controls
- Randomize prompt order to minimize position bias.
- Enable temperature jitter (e.g., 0.4–0.8) and seed rotation for diversity.
- Turn on “Deduplicate near-duplicates” to remove trivial rephrasings.
- Set a spend cap and an auto-pause threshold.
- Run and monitor
- Kick off the survey now or schedule it (e.g., weekly Mondays 09:00). Watch live logs for provider errors or throttling.
- Review raw responses
- Use the side-by-side viewer for quick qualitative scans before relying on metrics. Spot-check for hallucinations and off-topic answers.
3) Interpret results: mentions, sentiment, share of voice
- Mentions
- What it is: Count of canonicalized references to your entities after resolver cleanup.
- How to use: Check Top Entities and Co-mentions. Rising co-mentions (e.g., “Acme Cloud” with “security”) often precede shifts in positioning.
- Watch-outs: Generic terms (“atlas,” “flow”) and homonyms can inflate counts—use resolver feedback to tighten variants.
- Sentiment (v2)
- What it is: A context-aware score from -1 to +1 computed at span level, then aggregated per entity and feature.
- How to use: Compare overall brand sentiment vs. feature-level sentiment (e.g., “pricing” -0.22, “onboarding”