Abhord Quickstart Guide (2026 Refresh): From Setup to Action
Who this is for
New Abhord users who want a fast, reliable way to see how leading LLMs describe your brand, competitors, and category—and to turn those findings into action.
What’s new since the last edition
- Guided Setup Wizard with recommended defaults and model routing
- Sentiment v2.1 with better neutrality handling and rationale snippets
- Smarter deduping that merges near-duplicate mentions across models/sources
- Confidence intervals and outlier suppression on share of voice (SOV)
- Competitor Auto-Alias Finder and drift alerts
- Webhooks v2, Slack/MS Teams alerts, and saved dashboard templates
1) Initial setup and configuration
- Create a workspace
- Go to Settings → Workspace. Add your company name, primary domain, and time zone.
- Invite teammates. Assign roles: Admin (settings/billing), Analyst (runs/dashboards), Viewer (read-only).
- Connect LLM providers
- Settings → Connections → LLM Providers. Add keys for the models you want to survey (you can use Abhord-managed credits or your own).
- Enable “Model routing” to automatically include a balanced panel across families (recommended).
- Define entities and aliases
- Library → Entities. Add your brand, products, executives, and common misspellings.
- Add competitors and their known aliases. Turn on Auto-Alias Finder to capture emergent nicknames.
- Configure topics and intents
- Library → Taxonomies. Import the “AEO Essentials” taxonomy to tag mentions by funnel stage (awareness, consideration, support) and by theme (pricing, features, trust).
- Privacy and compliance
- Settings → Governance. Review data retention (default 180 days), PII scrubbing, and export controls. Enable SSO and SCIM if available.
- Alerts and integrations
- Settings → Integrations. Connect Slack/Teams, Jira/Asana, and your BI tool.
- Create an alert rule: “Spike in negative sentiment >15 points day-over-day” to #abhord-war-room.
Pro tip: Run the Guided Setup Wizard. It will pre-load best-practice prompts, set safe rate limits, and schedule a weekly pulse.
2) Run your first survey across LLMs
Goal: Get a baseline of what LLMs currently say about you vs. top competitors.
- Start a new run
- Runs → New → “Cross-Model Pulse.” Name it “Baseline Q1” and select your entity set (you + 3–5 competitors).
- Choose your panel
- Keep the default Balanced Panel. This includes multiple model families to reduce single-model bias.
- Select prompt pack
- Use “Brand Facts & Comparisons (2026)” prompt pack. It asks consistent, evaluative questions (e.g., “Best X for Y?”, “Top alternatives to Brand A?”, “Pros/cons of Product B?”).
- Turn on Auto-Translate if you operate in multiple regions; Abhord will normalize results back to your base language.
- Sampling and controls
- Sample size: 50–100 responses per entity per question for a baseline.
- Randomization: On (varied phrasings to reduce prompt artifacts).
- Temperature: 0.2–0.4 for factual queries; 0.6 for qualitative insights.
- Safety: Enable citation request where supported; discard no-citation answers if high precision is required.
- Dry run and launch
- Click “Preview 10.” Check for entity confusion and adjust aliases.
- Launch. Most baselines complete in 20–60 minutes depending on panel size.
New recommendation: Use “Cost Guard.” It caps spend and auto-pauses if marginal new mentions fall below your uniqueness threshold.
3) Interpreting results: mentions, sentiment, share of voice
- Mentions
- What it is: Count of unique, deduped references to your entity.
- How it’s calculated: Abhord groups near-duplicates across models using semantic clustering and alias maps, then assigns each cluster to a canonical entity.
- What to look for: Top clusters, new themes, and cross-model consistency.
- Sentiment (v2.1)
- Scale: -100 (very negative) to +100 (very positive), with a distinct “neutral/informational” band.
- What’s new: Rationale snippets show the sentence driving the label; neutrality detection is improved for how-to/support content.
- What to look for: Theme-level sentiment by funnel stage; sharp drops linked to pricing or reliability claims.
- Share of Voice (SOV)
- Definition: Your mentions divided by total category mentions during the run.
- What’s new: Confidence intervals and outlier suppression for small-sample entities.
- What to look for: SOV by model family (are some models under-representing you?), by region/language