How to Improve AI Visibility: Why Earned Media Outranks Your Website in ChatGPT Search

Most UK professional services firms direct their AI visibility budgets at their own websites. ChatGPT retrieves from earned media sources 95% of the time and from brand websites approximately 1% of the time (AuthorityTech, 2025; Averi, 2025). Only 11% of UK companies have 75–100% AI-discovery-ready content (WhiteHat SEO, 2026). Both statistics share the same cause.

This pattern appears in every assessment we run. A firm carries a strong domain, detailed service pages, clean schema markup, and no AI recommendations. Firms in this position ask why their competitors appear in ChatGPT while they do not. The gap is source type. AI retrieval systems apply a source type filter before evaluating content quality, and firms without earned media presence fail that filter before AI reads a single word of their website (AuthorityTech, 2025).

How AI Constructs Its Source Pool Before Reading Your Content

When ChatGPT or Perplexity processes a professional services query, the model constructs a candidate retrieval pool before ranking individual results (OpenAI, 2024). The model applies source type as one of its first retrieval criteria. Third-party editorial coverage, review platforms, professional forums, and industry directories sit in a higher-weighted source pool than brand-owned pages (AuthorityTech, 2025; Semactic, 2025). The model then runs a corroboration check: does external evidence confirm this business exists and operates as described?

The earned media source filter is the retrieval weighting AI systems assign to source types before assessing individual content quality (Semactic, 2025). It runs before AI measures authority. A firm with strong domain authority but thin earned media presence scores well on traditional SEO metrics and still fails AI retrieval for recommendation queries (Averi, 2025). AI models treat corroboration as a hallucination risk signal: recommending a business endorsed by multiple independent sources carries lower error risk than recommending one that appears only on its own site (Anthropic, 2024).

Why Am I Not Appearing in ChatGPT Recommendations?

Four patterns repeat across UK firms with no AI citations despite credible websites.

Website-only corroboration. The firm has a detailed site and Google Business Profile but no editorial coverage in sector press or regional business media. AI retrieval finds self-declaration and no independent confirmation (AuthorityTech, 2025).

Single-platform review presence. Businesses on four or more review platforms receive between 4.6 and 6.3 ChatGPT citations; single-platform businesses receive 1.8 (AuthorityTech, 2025). Most UK firms hold Google Reviews alone.

Community discussion absence. Reddit and professional forums carry measurable source weight for UK professional services queries (AuthorityTech, 2025). Website improvements do not compensate for community discussion absence.

Infrequent PR. One press release per quarter produces a historical record, not the citation density AI treats as active market presence (Semactic, 2025).

What AI-Visible UK Firms Have in Common

When we analyse UK professional services firms appearing in ChatGPT, Perplexity, and Claude recommendations, the pattern is earned media breadth: editorial mentions in sector publications, ratings on multiple review platforms, active presence in professional community discussions, and commentary distributed across regional and trade press (Averi, 2025; Semactic, 2025). The characteristic is source diversity across platform types.

Firms with decade-old websites can carry stronger AI citation rates than modern sites built to current SEO standards when earned media coverage differs (WhiteHat SEO, 2026; AuthorityTech, 2025). AI visibility tracks source diversity rather than website technical quality.

Ask yourself this: if you searched your service category and UK city on ChatGPT or Perplexity today, how many cited results reference your firm from a source you do not control? Fewer than three means your firm fails the earned media check before AI evaluates authority or content quality.

This is one filter in a layered AI retrieval system. When earned media is absent, the source type check removes a firm from the recommendation pool before AI considers authority, entity recognition, or content quality. Improving website schema without addressing earned media coverage addresses the wrong stage of retrieval.

AI-referred sessions grew 527% in the past 12 months (LLMRefs, 2025; Enrichlabs, 2025). For UK businesses working to improve AI visibility, the starting point is independent corroboration: editorial coverage in sector press, presence on at least two review platforms, and verifiable community mentions that give AI retrieval systems external evidence to include your firm in the candidate pool (Averi, 2025; AuthorityTech, 2025). Build those before optimising anything else.

References

• Anthropic (2024) Model calibration and hallucination risk reduction in Claude. Available at: anthropic.com (Accessed: 31 March 2026).
• AuthorityTech (2025) Earned media and AI citation analysis: how generative models select sources for recommendations. Available at: authoritytech.io (Accessed: 31 March 2026).
• Averi (2025) AI citation behaviour: source type weighting in generative search results. Available at: averi.ai (Accessed: 31 March 2026).
• Enrichlabs (2025) Generative AI traffic growth: annual analysis of AI-referred sessions. Available at: enrichlabs.ai (Accessed: 31 March 2026).
• LLMRefs (2025) AI-referred traffic tracking: referral sessions from generative AI platforms. Available at: llmrefs.com (Accessed: 31 March 2026).
• OpenAI (2024) ChatGPT search: how the model retrieves and ranks sources. Available at: openai.com (Accessed: 31 March 2026).
• Semactic (2025) Earned media priority in AI-generated recommendations: source type analysis. Available at: semactic.com (Accessed: 31 March 2026).
• WhiteHat SEO (2026) UK B2B AI search research: AI discovery readiness in professional services. Available at: whitehat-seo.co.uk (Accessed: 31 March 2026).