Title: Abhord Quickstart Guide (Refreshed Edition)
What’s new in this refresh
- Cross-LLM orchestration templates: prebuilt projects that query multiple models with consistent settings.
- Entity Normalization v2: better alias handling and near-duplicate collapsing across models and time windows.
- Share-of-Voice (SOV) by segment: breakouts by model family, query type (Q&A vs long-form), and time period.
- Mentions Quality Score: evidence-backed scoring so you can weight high-confidence mentions.
- Automation upgrades: webhooks, Slack/Teams alerts, and scheduled CSV exports from any saved view.
1) Initial setup and configuration
- Create a workspace
- Add your brand, product lines, and known aliases (YourBrand, Your Brand, YB). Include common misspellings.
- Invite teammates and assign roles (Viewer, Analyst, Admin). Limit edit rights to maintain metric consistency.
- Connect destinations (optional)
- Slack/Teams: pick a channel for alerts.
- Webhook: paste an endpoint for downstream dashboards or a CDP.
- Email digests: choose daily or weekly rollups.
- Configure projects
- Choose a template: “Brand Baseline,” “Competitor Tracking,” or “Launch Watch.”
- Select LLMs to sample. Recommendation: include at least three distinct families to reduce model bias.
- Fix global settings: temperature, tool/browsing use, region/locale. Keep these constant across runs for valid comparisons.
- Build your dictionary
- Entities: brand, product, executive names, and competitors.
- Topics: your most important use cases (“privacy,” “pricing,” “integration with X”).
- Exclusions: non-relevant homonyms (e.g., “Acorn (finance)” vs “Acorn (tree)”).
Tip (new): Turn on Near-Duplicate Collapsing at 85–90% similarity to prevent overcounting repeated or templated answers across models.
2) Run your first survey across LLMs
- Draft 8–12 high-intent queries users actually ask, for example:
- “Best [category] for [audience/use case]”
- “Top alternatives to YourBrand”
- “YourBrand vs CompetitorA for [scenario]”
- “Is YourBrand good for [constraint: budget/region/compliance]?”
- Use the Prompt Library
- Apply a prebuilt “Cross-LLM Q&A” template. It standardizes instructions (answer style, length, justification).
- Add neutral framing to reduce bias: “Evaluate options impartially; list reasons; cite known facts if available.”
- Choose models and cadence
- Start with 3–5 models. Run once now (baseline), then schedule daily or weekly.
- Lock parameters (temperature, tools)