Abhord Quickstart Guide (Refreshed Edition — March 2026)
Who this guide is for
- New Abhord users who want a fast, practical path from setup to insights.
- Growth, SEO/GEO, product marketing, and research teams benchmarking brand visibility and perception across LLMs.
What’s new since the last edition
- Multi-model portfolios: run the same survey across curated groups of LLMs with one click.
- Grounded mode: attach evidence packs (URLs/PDFs) to reduce hallucinations and improve citation rates.
- Sentiment v2: aspect-aware sentiment with confidence scoring and neutral/ambivalent handling.
- Share of Voice (SOV) normalization: cross-model deduplication and alias grouping built-in.
- Competitor Overlap Graph: visualize co-mentions and category adjacency.
- Real-time alerts and weekly digest reports; Slack/Webhook integrations.
- Cost controls: per-project caps and “thrifty mode” for exploratory runs.
1) Initial setup and configuration
- Create your workspace
- Add your organization name and default region/language.
- Invite teammates with roles (Admin, Analyst, Viewer). Tip: restrict API key visibility to Admins.
- Connect models
- Choose from managed connectors (e.g., OpenAI, Anthropic, Google, Mistral, Meta) or bring-your-own keys.
- Set cost limits and latency preferences. For discovery, use “thrifty mode.” For production baselines, choose “balanced.”
- Define entities and aliases
- Create your Brand entity and add aliases (e.g., product lines, ticker, common misspellings).
- Add Competitor entities with known variants. The Alias Manager is critical for accurate SOV.
- Build your taxonomy
- Create tags for use cases, buyer stages (Awareness, Consideration, Decision), and attributes (Price, Performance, Support).
- Map tags to queries so results auto-classify during ingestion.
- Evidence packs (optional but recommended)
- Upload a short “Fact Sheet” (1–2 pages), pricing tables, and key docs or provide canonical URLs.
- Use Grounded mode to bind surveys to these sources when factual accuracy matters.
- Notifications and compliance
- Enable Slack/Webhook alerts for significant shifts (e.g., negative sentiment spike >10 pp).
- Toggle PII scrubbing and data retention policies to match your compliance posture.
2) Run your first survey across LLMs
Goal: Understand how major LLMs describe your category and where your brand appears.
- Create a new survey
- Objective: “Category landscape — top solutions and positioning.”
- Language/Region: match your primary market.
- Choose a model portfolio
- Start with a Balanced portfolio (e.g., 4–6 general-purpose models). You can add a Cost-optimized portfolio for breadth later.
- Draft prompts (two-arm approach)
- Arm A (open-ended): “In [category], what are the leading solutions for [use case], and why? Include pros/cons.”
- Arm B (comparative): “Compare [Brand] with other notable providers in [category] for [use case]. Who is best for which needs?”
- Add a brief instruction: “Cite sources if available; avoid outdated info; if unsure, say so.”
- Attach evidence (optional)
- Toggle Grounded mode and select your Fact Sheet + 2–3 authoritative third-party URLs.
- Sampling and settings
- Runs per model: 5–10 (higher for high-stakes baselines).
- Temperature: 0.3–0.5 for consistency; 0.7 for discovery.
- De-duplication: on. Model-specific retries: on (max 1).
- Cost cap: set a ceiling to avoid overruns.
- Launch and monitor
- Use Live Logs to watch responses stream in and automatically tag mentions and attributes.
3) Interpret results: mentions, sentiment, share of voice
- Mentions
- What it is: Count