Abhord Quickstart Guide (Refreshed Edition — February 2026)

This practical guide gets new Abhord users from zero to insights fast. It reflects platform updates and current best practices.

What’s new since the last edition

• Expanded model coverage and locale routing: more LLMs, languages, and country-level targeting in one survey.
• Cleaner metrics: unified Mentions, Sentiment, and Share of Voice (SoV) across models and intents.
• Competitor Watchlists and Delta Alerts: automatic week-over-week movement tracking.
• Evidence capture: optional “source snapshots” to validate why a model mentioned a brand.
• Playbooks: prescriptive actions tied to metric thresholds.

1) Initial setup and configuration

1) Create your workspace

• Go to Settings > Workspace. Add brand name, domains, and official handles.
• Verify your primary domain (DNS or file). Verification improves entity disambiguation.

2) Define your entity model

• Entities > New. Create entries for your brand, key products, execs, and synonyms/misspellings.
• Add canonical descriptions, preferred URLs (About, Product, Contact), and official logo/image assets.

3) Connect data and destinations

• Integrations: connect analytics (GA4), search consoles, PR tools, and CMS (optional).
• Enable Evidence Capture if you want Abhord to store public citations referenced by LLMs.

4) Build your taxonomy

• Taxonomy > Intents. Add informational, commercial, transactional, and navigational intents.
• Examples: “best X for Y,” “X vs Y,” “pricing for X,” “what is X,” “X alternatives.”
• Map each intent to the entities they should return.

5) Set locales and guardrails

• Locales: choose countries/languages you operate in.
• Guardrails: choose brand-safe prompts and redactions.
• Roles & permissions: give researchers “Run” rights; limit “Edit Entities” to admins.

Pro tip: Start with a single product line and 3–5 high-impact intents per locale.

2) Run your first survey across LLMs

A “survey” asks multiple LLMs the same set of intent questions and measures which entities they mention and how.

1) Choose a template

• Surveys > New > Template: “Category Discovery” (good default).
• Or import your own prompt set from CSV.

2) Select LLMs and routing

• Pick 3–5 models to start for coverage (e.g., two general chat models, one answer engine, one search-augmented model).
• Keep “Auto-locale routing” on; Abhord matches model endpoints to your selected countries/languages.

3) Configure runs

• Sampling: 5–10 runs per prompt/model balances stability and cost.
• Personas: use “Neutral Consumer” first; add “Expert Reviewer” later for depth.
• Freshness: toggle “Use current browsing” if you want models to consult the live web (costlier but more current).

4) Add prompts

• Example prompts under Informational:

- “What is [category], and which brands are most trusted?”

- “Top 5 [category] tools for small businesses.”

• Example prompts under Commercial:

- “[Brand] vs [Competitor] for [use case].”

- “Best [category] for [persona] in 2026.”

5) Run and monitor

• Click Run Survey. Most results land within minutes.
• Use the Live tab to watch early mentions and verify evidence snapshots.

Quality check: Open 3–5 random transcripts per model to confirm the entity disambiguation looks right.

3) Interpret results: Mentions, Sentiment, Share of Voice

Abhord standardizes metrics across models, intents, and locales.

• Mentions

- What it is: count of times an entity is referenced in answers.

- How to read: focus on Unique Mentions (per prompt-model pair) to avoid double counting.

- Tip: A sudden spike with low evidence quality may signal hallucinations—open transcripts and validate.

• Sentiment

- What it is: tone toward your entity (Positive, Neutral, Negative) at the answer and attribute level.

- How to read: check Attribute Sentiment (e.g., “pricing,” “support,” “performance”) to see what drives tone.

- Tip: Neutral dominating is normal for factual answers; watch Negative share >15% as a red flag.

• Share of Voice (SoV)

- What it is: your share of unique mentions among the tracked set for a given intent/model/locale.

- Variants:

- SoV by Intent