Abhord Quickstart Guide (2026 Refresh)

Welcome to Abhord, the GEO/AEO platform built to measure and improve how large language models (LLMs) talk about your brand. This refreshed edition reflects 2026 updates: broader model coverage, improved entity resolution, aspect-level sentiment, smarter share‑of‑voice math, and faster survey pipelines.

What’s new since the last edition

• Broader model fleet: latest GPT‑4.x family, Claude 3.5+, Gemini 1.5, and top open‑weights (e.g., Llama 3.x) via bring‑your‑own‑key or Abhord-managed credits.
• Deterministic “Benchmark Mode”: per‑model temperature locks, seeds, and retry logic for replicable results.
• Upgraded Entity Resolver: synonym lists, misspelling handling, and false‑positive suppression for brand names with common words.
• Aspect Sentiment: sentiment at the feature/topic level (e.g., “pricing,” “privacy,” “support”), plus confidence scores.
• Normalized Share of Voice (SOV): weighting by answer prominence and salience to prevent overcounting long responses.
• Narrative Detector: detects emergent phrases and claims trending across models (“X is the budget leader,” “Y is HIPAA‑compliant,” etc.).
• Faster runs: rate‑limit‑aware parallelization, with safe‑mode throttling for fragile APIs.
• Workflow add‑ons: Slack/Email alerts, CSV/JSON exports, and Webhooks for downstream BI.

1) Initial setup and configuration

1) Create a workspace

• Name it after your brand or product line.
• Invite teammates with roles: Admin (billing + settings), Analyst (create/run surveys), Viewer (read‑only).

2) Connect models

• Choose either:

- Bring your own API keys (OpenAI, Anthropic, Google, etc.).

- Use Abhord credits for one‑click access to curated models.

• Tip: enable Benchmark Mode for surveys you’ll repeat monthly.

3) Define entities

• Add your brand, products, and competitors.
• Provide: canonical name, common abbreviations, known misspellings, and exclusions (e.g., “Acme” ≠ “Acme Tools Inc.” if irrelevant).
• Attach reference URLs or canonical descriptions; these power entity disambiguation and help detect hallucinations.

4) Set locales and languages

• Select priority markets (e.g., en‑US, en‑GB, de‑DE).
• If you serve multi‑region pricing or features, turn on Region‑Aware Prompts.

5) Configure governance

• Enable PII scrubber by default.
• Set data retention (recommended: 90 days for raw answers; unlimited for aggregated metrics).
• Add Slack or Email alerts for threshold changes.

Recommended defaults for new teams:

• 4–6 models, 10–20 samples/model/intent, temperature ≤ 0.4 in Benchmark Mode, daily or weekly cadence.

2) Run your first survey across LLMs

Objective example: “Understand how LLMs recommend CRM tools for startups and where our brand appears.”

1) Choose a template

• Brand Perception, Buying Guide, How‑To, or Top‑Answer (one‑shot summary).
• Start with Buying Guide if you care about conversion‑adjacent queries.

2) Define intents (queries)

• Draft 8–15 intents that mirror real user language:

- “best CRM for startups 2026,” “affordable CRM with email automation,” “HubSpot alternatives,” etc.

• Cluster by theme; Abhord will rotate paraphrases to reduce sampling bias.

3) Select models and parameters

• Pick 4+ models spanning proprietary and open‑weights.
• Set sample size (n) per model (