Docs
Monitoring

How prompts work

Understanding Heading's prompt generation and quality scoring

Prompts are the foundation of Heading's monitoring. Each prompt represents a question that real people might ask an AI assistant about your business category.

Automatic prompt generation

When you add a property, Heading generates prompts automatically based on:

  • Your domain: what your website is about
  • Business vertical: your industry category (legal, plumbing, SaaS, etc.)
  • Location: where your business operates
  • Competitor landscape: what people ask when comparing options

Generation covers informational, transactional, comparison, and recommendation intents. A free property starts with a 20-prompt snapshot portfolio. Paid plan quotas are 50 (Starter), 100 (Growth), or 250 (Scale). See Plans.

Prompt quality scoring

Not all prompts are equally useful. Heading scores each prompt on three factors:

FactorWhat it measuresScore range
SpecificityDoes the prompt mention category + location + intent?1-5
Citation likelihoodDoes this type of query typically generate source links?1-5
Competitive densityHow many brands compete for this prompt?1-5

Combined score out of 15. Prompts scoring below 8 are flagged for review.

High-scoring prompts are specific and likely to generate citations. "Best personal injury lawyer in Manchester" scores higher than "what is a lawyer" because it's more likely to cite a specific firm.

Prompt portfolio score

Your prompt portfolio score (0 to 100) measures how well-balanced and effective your set of active prompts is. A higher score means your prompts cover a wider range of the questions real people ask about your industry, which gives you better visibility data.

The score is calculated across six dimensions, each weighted by importance:

DimensionWeightWhat it measures
Tier balance25%Distribution across prompt tiers: brand, category, problem/solution, and adjacent queries. A balanced portfolio covers all four.
Specificity20%Ratio of specific prompts (mentioning location, brand, or intent) to broad ones. Target: 60% specific.
Buying stage20%Coverage across awareness, consideration, and decision stages. All three stages should be represented.
Conversational tone15%How natural your prompts sound: word count, question marks, and phrasing style. AI assistants respond better to conversational queries.
Question variety10%Mix of question types (what, how, why, which, other). Higher entropy means better coverage of how people actually ask questions.
Competitor mentions10%Percentage of prompts that reference competitors. Target: around 20%, enough to track comparisons without over-indexing.

Each dimension is scored 0 to 100, then combined using the weights above to produce your overall portfolio score.

A score of 70+ is strong. Below 40 means your prompts are clustered in a narrow range and you're likely missing important queries. Use the radar chart on the Prompts page to see which dimensions need attention.

Improving your score

  • Suggest more per tag in underrepresented segments. Heading generates prompts to fill the gaps (see Tags & segmentation)
  • Track suggested prompts that complement your existing set
  • Add custom prompts for competitor comparisons or specific buying stages you're missing

Managing prompts

You can:

  • Track: include in monitoring (default for high-scoring prompts)
  • Reject: exclude from monitoring (useful for irrelevant prompts)
  • Add custom: write your own prompts for specific monitoring needs
  • Bulk actions: track, reject, or delete multiple prompts at once
  • Tag: assign tags for segmentation and filtering (see Tags & segmentation)