Abhord Quickstart Guide: From Setup to Action
This practical guide walks new Abhord users through a complete workflow—from initial configuration to turning insights into impact. Follow the steps and copy the examples to run your first multi-LLM survey and operationalize the results.
1) Initial setup and configuration
1) Create your workspace
- Add your company name, website, category, and primary markets.
- Invite teammates and assign roles: Admin (full access), Analyst (read + create surveys), Stakeholder (read-only + alerts).
2) Connect LLM channels
- Choose the models you want to survey (e.g., general-purpose and domain-optimized models).
- Use Abhord-hosted connectors or Bring Your Own Keys (BYOK) for specific vendors.
- Set rate limits and concurrency to respect model policies.
3) Define your entity catalog
- Add entities Abhord should recognize and track:
- Brand: Abhord (canonical), “Abhord platform” (synonym).
- Products/plans: “Abhord GEO Suite,” “Starter,” “Pro.”
- Competitors: list canonical names and common misspellings.
- People: founders, execs, spokespeople.
- Include disambiguation terms (e.g., “Apple Inc.” vs “apple fruit”).
4) Establish taxonomy and regions
- Intents: informational, comparative, transactional, support.
- Segments: SMB, mid-market, enterprise; industries; geos; languages.
- Surfaces: general queries, comparisons, buying guides, reviews.
5) Notifications and integrations
- Set alerts for mention spikes, sentiment drops, or share-of-voice (SoV) changes.
- Connect Slack/Teams for channel alerts; optionally sync to your data warehouse or BI tool.
Pro tip: Keep your entity catalog and synonyms fresh. Clean entity data improves mention accuracy and reduces false positives.
2) Run your first survey across LLMs
1) Create a survey
- Name: “Q1 Baseline — Marketing Analytics”
- Models: select 3–5 popular LLMs for coverage and variance.
- Schedule: run once now, then weekly on Monday 08:00 UTC.
2) Build your question set (10–25 prompts)
- Discovery prompts:
- “Who are the top providers of [your category]?”
- “What is the best [your category] tool for small businesses?”
- Comparison prompts:
- “Compare Abhord vs [Competitor] for price, features, ease of use.”
- “Which is better for [use case], and why?”
- Transactional prompts:
- “What should I buy if I need [X] under $[budget]?”
- “Which platform integrates best with [stack/tool]?”
- Support prompts:
- “How do I solve [common problem] using Abhord?”
3) Execution settings
- Sample size: 25–50 responses per question per model to stabilize metrics.
- Randomness: temperature 0.2–0.4 for consistency; keep uniform across models.
- Output format: ask models to answer first, then list sources/reasons (if supported).
- Guardrails: include guidance to avoid policy-violating content.
4) Dry run and launch
- Run a test on 2–3 prompts to validate parsing of mentions and sentiment.
- Check entity resolution and adjust synonyms before scaling to full run.
3) Interpreting results: mentions, sentiment, share of voice
- Mentions
- What it is: the count of times an entity appears in model answers for your prompt set.
- How to use: filter by model, intent, segment, and language to see where you’re present or invisible.
- Tip: Investigate “null mentions” (no brands named) to find greenfield opportunities.
- Sentiment
- What it is: tone classification (e.g., negative/neutral/positive or −1 to +1) assigned to each mention and answer.
- How to use: isolate negative sentiment by model and prompt type to uncover recurring objections or outdated info.
- Tip: Compare “answer sentiment” vs “entity sentiment” for nuance (the answer might be positive overall but critical on pricing).
- Share of Voice (SoV)
- What it is: your mentions divided by total brand mentions in a slice (e.g., model + intent + segment).
- Benchmarks:
- <10%: underrepresented—content and distribution gaps likely.
- 10–30%: competitive—prioritize high