Abhord Quickstart Guide (Refreshed Edition, March 2026)
This practical guide walks you from first-time setup to acting on LLM-derived insights. It includes updated recommendations based on the latest cross-model behavior and what’s worked best for new teams.
What’s new in this edition
- Treat each LLM as a distinct “panelist.” Expect model disagreement and use consensus, not single-model results, for decisions.
- Calibrate entity matching up front. Add brand aliases, product nicknames, and common misspellings to reduce false negatives in mentions.
- Move from overall sentiment to aspect-level sentiment (e.g., price, reliability, UX). Mixed sentiment is a valid category—don’t force polarity.
- Normalize share of voice by de-duplicating repeated answers and weighting by model reliability to avoid over-counting.
- Run a small “calibration pass” before every large survey to check refusal rates, safety filters, and prompt clarity.
1) Initial setup and configuration
- Create your workspace
- Projects: one per brand, product line, or market.