Abhord Product Guide (Refreshed, January 2026)

This practical guide walks new Abhord users through setup, running your first cross-LLM survey, reading results, tracking competitors, and acting on insights. It includes 2026 updates and recommendations based on the latest model landscape and GEO/AEO best practices.

New in this edition:

• Cross-model calibration for sentiment and entity detection to reduce model bias.
• Alias Manager to normalize brand/URL/name variants (and common misspellings).
• SoV weighting by estimated model reach and query type.
• Search-augmented vs. recall-only mode selection per survey block.
• Cost guardrails: per-run caps, auto-downshifting to smaller models, and caching.
• Alerts and exports: Slack/email/webhook alerts and CSV/JSON export for BI tools.

1) Initial setup and configuration

1) Create your workspace

• Add brand entities: official brand name, product lines, URLs, social handles.
• Add aliases: common abbreviations, misspellings, legacy names (use Alias Manager).
• Define categories: e.g., “GEO platforms,” “LLM analytics,” “Answer engines.”

2) Connect data and compliance

• Upload reference pack: homepage, docs, pricing, comparison pages, PDFs.
• Set data handling: choose retention window, redact PII, and enable safe-mode logging if required by your org’s policy.

3) Choose models and budgets

• Enable a balanced panel: at minimum include one proprietary model (e.g., GPT/Claude/Gemini), one open-weight (e.g., Llama/Mistral), and one search-augmented model.
• Set cost guardrails: daily and per-run caps; opt into response caching for repeating prompts.

4) Team access and notifications

• Invite collaborators with roles (viewer, analyst, admin).
• Connect notifications (Slack/email) and webhooks for automation.

Pro tip: Spend 10 minutes on aliases and references. Clean entities and canonical sources improve recognition accuracy and reduce false mentions.

2) Run your first survey across LLMs

Goal: Learn how LLMs currently answer category-defining queries and where your brand appears.

1) Create a survey

• Name: “Q1 Baseline—GEO Platforms.”
• Audience: “Global, English.”
• Cadence: One-time (baseline), then schedule weekly.

2) Add question blocks

• Discovery queries (top-of-category): “What are the leading GEO platforms?” “Which tools help optimize answers from LLMs?”
• Intent queries (brand + task): “Best way to run multi-LLM surveys?” “How to track share of voice across LLMs?”
• Comparison queries: “Abhord vs [Competitor] for LLM insights.”

3) Pick model panel and modes

• Panel example: GPT-4.x, Claude 3.5, Gemini 2.x, Llama 3.x 70B, Mistral Large.
• Mode: Run discovery blocks in search-augmented mode; run comparison blocks in recall-only to test latent knowledge.
• Samples: 3–5 runs per prompt per model to reduce variance.

4) Add context

• Attach the reference pack to intent and comparison blocks to test how well models use your canonical sources.

5) Launch and monitor

• Confirm budget preview. Enable auto-retry for transient API errors.
• Tag this run “Baseline-Jan-2026” for trend comparisons.

Pro tip: Keep prompts neutral. Avoid leading phrasing (e.g., “Why Abhord is best”) to get a reliable market read.

3) Interpreting results: mentions, sentiment, share of voice

Mentions

• What it is: Count of times your brand (or alias) appears across answers.
• What to watch: Duplicates and partial matches are auto-deduped; inspect “Ambiguous” tab for edge cases like generic words in your name.

Sentiment

• How it’s scored: Per-mention sentiment, calibrated across models using the cross-model baseline to reduce polarity drift.
• Tip: Read verbatims for