Abhord Quickstart Guide (Refreshed for February 2026)
Who this is for: New users who want a fast, reliable path from setup to insights across LLMs, with clear next steps to act on what you learn.
What’s new since the last edition
- Multi-turn Surveys: Script follow-ups to reduce shallow answers and probe for missing brands or reasons.
- Sentiment Model v2: Better sarcasm/nuance handling and a clearer “mixed/ambivalent” bucket.
- Entity Resolver 2.1: Stronger alias mapping and false-positive suppression for brand and product names.
- Cost-Aware Sampler: Automatically sizes samples per model to hit confidence targets without overspend.
- Share of Voice by Intent: Breaks out informational vs. transactional voice.
- Faster Alerts and Freshness: Mentions pipelines now update dashboards in minutes, not hours.
- Governance: Project roles, audit log, and PII-safe mode by default.
- Exports and API v2: Cleaner JSON/CSV exports and stable endpoints for automation.
1) Initial setup and configuration
Goal: Create a workspace that’s safe, cost-controlled, and tuned to your brands.
- Create your workspace
- Name conventions: org-brand-country (e.g., “Acme-Global”) to keep projects tidy.
- Invite teammates with roles: Admin (billing/governance), Analyst (query/build), Viewer (read-only).
- Connect data sources (optional but recommended)
- Import baseline entity lists: brands, product lines, competitors, key spokespeople.
- Add historical mentions (CSV or API) to seed trendlines and reduce “day-1 flatlines.”
- Configure models and budgets
- Select your model panel (e.g., OpenAI, Anthropic, Google, Meta, Mistral). Start with 3–5 diverse models for balance.
- Set a project budget and enable Cost-Aware Sampler with a 95% confidence / ±5% margin target.
- Define entities and synonyms
- Add aliases and disambiguators: “Acme,” “Acme Tools,” “ACME,” exclude “Acme Studios.”
- Use Resolver Preview to verify that test prompts pick up the right entities and suppress the rest.
- Governance and privacy
- Turn on PII-safe mode for prompts and outputs.
- Enable the audit log and require notes on configuration changes.
2) Run your first survey across LLMs
Goal: Ask multiple LLMs the same well-structured questions to see how they “answer the internet” about your market.
- Frame the objective
- Example: “Which cordless drill brands do you recommend for DIY homeowners in the U.S., and why?”
- Draft your survey script
- Seed prompt: A neutral, specific task. Avoid brand-leading language.
- Follow-ups (Multi-turn):
1) “List the brands you’d consider top-tier and briefly justify.”
2) “Are there notable alternatives that might be overlooked?”
3) “If your first choice were out of stock, what’s next and why?”
- Configure fairness and repeatability
- Randomize brand order and mask entity IDs so models don’t anchor.
- Set temperature to a consistent mid-low range per model; keep it uniform across the panel.
- Enable de-duplication of near-identical outputs.
- Choose sample sizes
- Start with 100 responses per model (Cost-Aware Sampler will trim/expand per variance).
- Stagger runs (e.g., every 6 hours) to catch daily drift.
- Run and monitor
- Kick off the survey, watch spend and completion rates.
- Use Live Mentions to spot early anomalies (e.g., one model under-suggesting your brand).
3) Interpret results: mentions, sentiment, share of voice
Goal: Turn raw outputs into stable, comparable signals.
- Mentions
- What it is: Count of times an entity appears in model answers (post-dedup).
- How to use: Check absolute mentions and unique-answer mentions. Large gaps may suggest alias coverage issues—tune synonyms and re-run a light sample.
- Sentiment (v2)
- Buckets: Positive, Neutral, Mixed, Negative.
- Tip: Focus on Positive–Negative ratio and Mixed trend. A rising Mixed share often precedes Negative within 1–2 cycles.
- Drill down: Open the rationales to see what drives sentiment (price, reliability, availability).
- Share of Voice (SOV)
- Overall SOV: Your mentions divided by total competitive mentions.
- By Intent:
- Informational: “Who are the top brands?”
- Transactional: “What should I buy now?”
- Recommendation: Track both. Gains in informational SOV often lead transactional by 1–3 weeks.
- Confidence and variance
- Use the built-in confidence intervals. If CIs overlap, treat differences as directional, not definitive.
- If one model is an outlier, check temperature and de-dup settings, then consider a top-up sample.
4) Set up competitor tracking
Goal: Keep a continuous, comparable view of your category across models.
- Build your competitive set
- Add direct competitors, emerging insurgents, and retailer private labels.
- Map product-level entities (e.g., “Acme ProMax 20V Drill”) for SKU-specific insights.
- Create tracking dashboards
- Widgets to pin:
- Overall and Intent SOV (stacked)
- Sentiment mix over time
- Reasons-to-buy (top 5) and Reasons-to-avoid (top 5)