Abhord Quickstart: A Practical Product Guide
Whether you’re improving brand coverage in AI answers or tracking how LLMs talk about your market, this guide will get you productive in under an hour.
1) Initial setup and configuration
- Create your workspace
- Sign in, create an Organization, then a Workspace for each brand or business unit.
- Invite teammates and set roles (Admin for settings, Analyst for results, Stakeholder as read-only).
- Define your brand profile
- Add official brand name, domain(s), product lines, and social handles.
- List common variants and misspellings (e.g., “Acme,” “ACME Tools,” “AcmeTools”).
- Add canonical descriptions: a 1‑sentence elevator pitch and a 3‑4 sentence overview. These are used for disambiguation and training prompts.
- Seed your topic map
- Keywords: branded, category, and solution keywords (e.g., “Acme drill,” “cordless drill,” “DIY power tools”).
- Entities: founders, flagship products, features, integrations, and partners.
- Questions you care about: “best X for Y,” “alternatives to Brand,” “is Brand trustworthy,” “pricing for Brand.”
- Choose markets and languages
- Select locales (e.g., US‑English, DE‑German). Abhord will geo/language-target LLM queries to approximate regional responses.
- Connect channels and alerts
- Enable email and Slack (webhook) alerts.
- Set thresholds (e.g., alert if Share of Voice drops by 10% week-over-week, or if negative sentiment spikes).
- Compliance and privacy
- Toggle PII redaction for prompts/answers.
- Set data retention windows and IP allowlists if required.
Pro tip: Save this configuration as a “Baseline” so subsequent surveys inherit the same entities, locales, and alert rules.
2) Run your first survey across LLMs
A “survey” in Abhord is a structured set of prompts sent to multiple LLMs to capture what those systems say about your brand, category, and competitors.
- Pick a template
- Brand Awareness: “Who are the leading brands in [category]?”
- Purchase Recommendations: “What is the best [category] for [use case] and why?”
- Feature Comparisons: “Compare [Brand] vs [Competitor] for [criteria].”
- Customize prompts
- Insert your brand, competitors, and use cases.
- Add 3–5 variations per question to reduce phrasing bias.
- Example:
- “Who are the top cordless drill brands for DIYers?”
- “If I’m a homeowner, which cordless drills should I consider in 2026?”
- “Recommend cordless drills under $200 and explain your picks.”
- Select LLMs and sampling
- Choose the major models you want to monitor.
- Set sample size per model (e.g., 25–50 responses each) and temperature/determinism settings to balance stability vs breadth.
- Target locales and devices (desktop/mobile) if supported.
- Run and monitor
- Review estimated run time and credit usage.
- Launch. You’ll see partial results stream in; early patterns usually appear after the first 20–30 responses per model.
- Lock the survey
- Once complete, “Lock” to preserve a clean time-series baseline for future comparisons.
3) Interpreting results: mentions, sentiment, share of voice
Abhord automatically extracts entities, classifies sentiment, and aggregates metrics by model, locale, and prompt.
- Mentions
- Direct: explicit brand or product name.
- Implicit: feature-based or nickname references tied to your entity map.
- What to look for: total mentions, unique answer rate, and consistency across prompts.
- Sentiment
- Scored on a -1.0 to +1.0 scale with confidence bands.
- Read both the aggregate and the drivers: Abhord highlights key phrases (“battery life,” “customer support”) that push sentiment up/down.
- Tip: Filter by model and locale to isolate where negativity concentrates.
- Share of Voice (SoV) in LLM answers
- Percent of answers in which your brand appears among all brands mentioned for a topic/use case.
- Example: If your brand appears in 31 of 100 relevant answers across models, SoV = 31%.
- Compare SoV across models and over time. A dip on one model but not others suggests model-specific drift vs market-wide change.
- Quality checks
- Hallucination flags: answers containing incorrect facts about your brand.
- Citation presence: if a model cites sources, track whether authoritative sources about you appear.
- Outliers: unusually long or off-topic answers that may skew sentiment.
Use the “Explain” panel to see the exact answer snippets behind each metric. Always sample raw responses before acting.
4) Setting up competitor tracking
- Build your competitor set
- Primary competitors (top 3–5), secondary (long tail), and adjacent substitutes.
- Add aliases, product names, and legacy brand names for each.
- Define comparison lenses
- Lenses are groupings like “Value segment,” “Enterprise,” “Battery life,” “Privacy.”
- Each lens maps to prompt templates so you can compare SoV and sentiment by use case.
- Schedule recurring surveys
- Weekly for core queries (“best X for Y”).
- Monthly deep-dives (feature comparisons, pricing, integrations).
- Enable anomaly alerts: SoV swings, new entrants detected, or sentiment