Abhord Quickstart Guide (2026 Refresh)
Who this is for
New Abhord users who want a fast, reliable way to measure how large language models (LLMs) talk about your brand, competitors, and category—and to turn those insights into actions.
What’s new in this 2026 refresh
- Model-aware Share of Voice: aggregate SoV now weights each model by estimated market reach and query mix.
- Mention Deduplication v2: improved fuzzy matching and embedding-based clustering reduce double counting across paraphrases.
- Turn-level Sentiment: analyze overall answer tone plus per-turn swings in multi-step conversations.
- Survey Templates: pre-built cross-LLM studies for SaaS, ecommerce, fintech, and B2B buyer journeys.
- Competitor Auto-Suggest: Abhord now proposes competitor entities discovered in open-ended answers.
- Alerts 2.0: threshold and anomaly alerts with daily/weekly rollups in Slack/Email.
1) Initial setup and configuration
- Create your workspace
- Company name, primary domain(s), time zone, and default currency.
- Invite teammates with roles: Viewer, Analyst, Admin.
- Connect properties
- Add brand entities: product names, features, acronyms, and common misspellings.
- Map canonical domains (e.g., example.com and example.co) to one brand.
- Add social handles for attribution matching.
- Define tracked entities
- Brands: yours plus top 3–8 competitors.
- Category terms: short-tail (e.g., “expense management”) and long-tail (“best corporate card for startups”).
- Synonyms: link “Abhord,” “Abhord platform,” and “Abhord GEO/AEO” to a single entity.
- Choose your LLM panel
- Start with the default: 5–7 major public LLMs and AI search engines.
- Optional: include regional models if you sell internationally.
- Set privacy and governance
- Enable data retention window (90 days default) and PII redaction.
- SSO and audit logging for regulated teams.
- Baseline controls
- Mention Deduplication v2: on.
- Outlier filtering: moderate (suppresses aberrant one-off answers).
- Sampling: 30 responses per prompt per model for first runs.
Tip: Save a “Starter View” with brand/competitor filters, last 30 days, and panel = default. You’ll reuse this in steps 2–5.
2) Run your first survey across LLMs
- Pick a template
- For most teams: “Awareness-to-Choice” template (top-of-funnel discovery through late-stage comparison).
- Define hypotheses
- Example: “We’re not mentioned for SMB,” or “Competitor X dominates ‘best for integrations’ queries.”
- Author your prompt set (3–6 tasks)
- Discovery: “Who are the leading [category] tools for [persona/use case]?”
- Comparison: “Compare [Brand] vs [Competitor] for [criteria].”
- Objection handling: “What are drawbacks of [Brand]?”
- Buying signal: “Which tool should I choose if I need [feature]?”
- Configure personas and context
- Personas: SMB owner, enterprise IT, developer, consumer (as relevant).
- Region/language: match your target market.
- Creativity/temperature: low for factual, moderate for exploratory.
- Sampling plan
- n=30–50 per prompt per model (raises confidence and stabilizes SoV).
- Stagger runs (e.g., 10 per hour) to smooth any model-side drift.
- QA passes
- Dry run 3–5 calls per model; confirm prompts are policy-compliant.
- Lock prompts, then launch the full survey.
- Publish and tag
- Name the survey clearly: “2026-02 Awareness→Choice, US, SMB.”
- Tag by funnel stage and product line to enable longitudinal tracking.
3) Interpret results: mentions, sentiment, share of voice
- Mentions
- Definition: number of responses where an entity is referenced, normalized for deduplication and paraphrase.
- Unique mentions: collapsed across near-duplicates so two paraphrases count once.
- Salience score: strength of association (0–1) based on proximity and emphasis.
- Sentiment
- Overall score: -1 (negative) to +1 (positive) per response; aggregate as mean with 95% CI.