Abhord Quickstart Guide (Refreshed — March 2026)
Who this is for
New Abhord users who want a fast, reliable way to measure how large language models (LLMs) talk about your brand and competitors—and to turn those insights into action.
What’s new since the last edition
- Faster survey cycles and drift-aware scheduling: LLM outputs change more frequently; we now recommend weekly or event-triggered runs.
- Expanded model coverage guidance: Include at least one frontier model, one assistant-style model, one search-augmented model, and one open-weight model for balance.
- Better entity hygiene practices: Standardize brand/competitor aliases to improve mention detection and share-of-voice accuracy.
- Citations-first analysis: Track and tag sources LLMs cite (docs, reviews, Wikipedia/Wikidata, GitHub, app stores) to guide content fixes.
- Scenario sets over single prompts: Run grouped prompts across the funnel (informational, comparative, transactional) to reduce variance and surface consistent gaps.
1) Initial setup and configuration
Goal: Create a clean foundation so Abhord recognizes your brand and competitors consistently.
- Create a workspace and project
- Name it after the market/use case (e.g., “US — Project Management — SMB”).
- Set default locale(s) and time zone; start with your primary market.
- Define your brand entity
- Canonical name: Your official brand and product names.
- Aliases: Common misspellings, acronyms, legacy names (e.g., “Acme PM,” “AcmeProject,” “Acme Project Manager”).
- Exclusions: Words that should not count as mentions if they’re generic.
- Add competitors
- Include canonical names and aliases for each (e.g., “Basecamp,” “37signals Basecamp,” “Base Camp”).
- Group by tier (direct, adjacent, aspirational) for cleaner reporting.
- Select models to survey
- Suggested balanced matrix:
- Frontier/generalist: at least 1 (e.g., top-tier chat model).
- Assistant-style with tools: at least 1 (often used by end users).
- Search-augmented answer engine: at least 1 (influenced by the open web).
- Open-weight/local family: at least 1 (for transparency and trend contrasts).
- If available, enable geo-routing so models that localize answers are tested in your target country.
- Configure schedules and alerts
- Cadence: weekly baseline + ad hoc runs after major releases, pricing changes, or big press.
- Alerts: set thresholds for share-of-voice drops, sentiment dips, or competitor surges.