← All posts

How LLMs Reduce Options Before Recommending Businesses

Analysis of the filtering process AI systems use to narrow potential recommendations to the businesses they actually suggest.

By SEEN Research
  • ai-discovery

When a user asks an AI assistant for a local business recommendation, the system does not evaluate every business in the category. Instead, it applies a series of filters that progressively narrow the option set before final recommendation. Understanding this filtering process reveals why many businesses never reach the recommendation stage.

The Filtering Funnel

AI recommendation appears as a single output, but it likely involves multiple filtering stages:

Stage 1: Information Access

The first filter determines which businesses AI can find information about. Businesses must be:

  • Accessible to AI crawlers (not blocked by robots.txt)
  • Represented in AI training data or real-time search
  • Present on platforms AI can access

Businesses invisible at this stage never enter consideration.

Stage 2: Entity Recognition

AI must recognize the business as a distinct, identifiable entity:

  • Clear business name and identity
  • Consistent information confirming entity
  • Sufficient information to form entity model

Fragmented or unclear entities may not be recognized as recommendation candidates.

Stage 3: Category Matching

AI must match the business to the user’s need:

  • Service type matches query intent
  • Business type is appropriate for the request
  • Capabilities align with specific need (if stated)

Businesses outside the matching category are excluded.

Stage 4: Geographic Matching

For local queries, AI must confirm geographic relevance:

  • Service area includes user’s location
  • Business operates in relevant geographic scope
  • Coverage is explicit, not inferred

Businesses with unclear or non-matching geography are excluded.

Stage 5: Trust Assessment

AI evaluates trust signals for remaining candidates:

  • Review presence and sentiment
  • Credential verification
  • Consistency signals
  • Authority indicators

Businesses with insufficient trust signals are excluded.

Stage 6: Final Selection

From remaining candidates, AI selects businesses to recommend based on:

  • Relative strength of trust signals
  • Specificity of match to query
  • Recency and currency of information
  • Differentiating factors relevant to query

Only businesses passing all stages reach recommendation.

Filtering at Each Stage

StageFiltering CriteriaTypical Failure Rate
Information AccessCrawlability, presence20-30% of local businesses
Entity RecognitionClarity, consistency15-25% of accessible businesses
Category MatchingService alignmentVariable by query
Geographic MatchingCoverage confirmation20-40% for specific locations
Trust AssessmentSufficient signals30-50% of geographically matched
Final SelectionRelative strengthVariable by competition

These estimates suggest that a small minority of businesses in any category reach the final selection stage.

Why Filtering Matters

The filtering funnel has important implications:

Early Elimination

Businesses eliminated in early stages never compete. A business with excellent service but blocked AI crawlers never reaches trust assessment.

Cumulative Effect

Each stage eliminates a percentage. If 70% pass each of six stages, only ~12% reach final selection (0.7^6 ≈ 0.12).

Optimization Efficiency

Addressing early-stage issues has outsized impact. Fixing crawl access enables competition; improving final-stage factors faces more competition.

Competitive Context

Final selection depends on alternatives. A business might pass all filters but still lose to competitors with stronger signals.

What This Means for Local Service Businesses

Understanding the funnel helps prioritize optimization:

HVAC Industry

HVAC businesses should ensure:

  • Crawl accessibility (robots.txt updated for AI crawlers)
  • Clear entity definition (consistent NAP, schema markup)
  • Category accuracy (HVAC-specific categorization)
  • Service area clarity (explicit geographic coverage)
  • Trust documentation (certifications, reviews, credentials)

Restoration Services

Restoration businesses should verify:

  • Crawler access (particularly for emergency services content)
  • Entity clarity across restoration types (water, fire, mold)
  • Category accuracy (restoration vs. general contracting)
  • Coverage area documentation (emergency response geography)
  • Certification documentation (IICRC credentials)

Mold Remediation

Mold remediation businesses should confirm:

  • Crawl accessibility
  • Clear entity as mold remediation specialist
  • Category distinction from general restoration
  • Coverage area definition
  • Certification prominence (AMRT, related credentials)

Plumbing Services

Plumbing businesses should ensure:

  • Crawl accessibility
  • Clear plumbing entity definition
  • Category specificity (plumber vs. general contractor)
  • Service area clarity (routine and emergency)
  • License and credential documentation

Electrical Contractors

Electrical contractors should verify:

  • Crawl accessibility
  • Clear electrical entity definition
  • Category accuracy (electrician vs. general contractor)
  • Geographic coverage (licensing jurisdiction alignment)
  • License level documentation (master electrician credentials)

Filtering analysis suggests most businesses fail at multiple stages:

  • Stage 1 (Access): Outdated robots.txt blocks AI crawlers
  • Stage 2 (Entity): Inconsistent NAP fragments entity
  • Stage 3 (Category): Generic categorization loses specific matches
  • Stage 4 (Geography): Ambiguous service area excludes location queries
  • Stage 5 (Trust): Insufficient or inaccessible trust signals
  • Stage 6 (Selection): Weaker profile than competition

Addressing only later stages while failing earlier stages produces no improvement.

Structuring a Business for AI Visibility

Optimization should proceed in filter order:

Stage 1 (Access): Ensure AI crawlers can access website content (robots.txt, technical accessibility).

Stage 2 (Entity): Implement schema markup, ensure NAP consistency across all platforms.

Stage 3 (Category): Use specific business categorization everywhere; create category-specific content.

Stage 4 (Geography): Define service area explicitly in schema, Google Business Profile, and website content.

Stage 5 (Trust): Document credentials, build multi-platform reviews, ensure citation consistency.

Stage 6 (Selection): Develop differentiating content, build unique authority signals.

Platforms like NowSeen.ai assess businesses across filter stages to identify where filtering failure occurs.

Where AI-Driven Discovery Is Headed

Filtering dynamics will likely intensify:

Stricter Early Filters

As AI improves, early-stage filters may become more stringent, requiring better technical implementation.

Deeper Category Matching

AI may develop more nuanced category matching, favoring businesses with specific rather than general positioning.

Real-Time Assessment

Filtering may become more dynamic, with businesses entering or leaving consideration based on recent signals.

Reduced Final Sets

As AI confidence grows, final selection may narrow to fewer recommendations, intensifying competition among businesses that pass earlier filters.

Conclusion

AI recommendation involves a filtering funnel that progressively narrows potential candidates. Businesses must pass multiple stages—from basic accessibility to final competitive selection—to be recommended.

Understanding where filtering failure occurs helps prioritize optimization. Addressing early-stage issues (crawl access, entity clarity) enables competition. Addressing later-stage issues (trust signals, differentiation) improves competitive position among businesses that reach final stages.

Most businesses fail at multiple stages. Comprehensive AI visibility optimization addresses the entire funnel, not just the final selection stage.