SAVI & GEO Score Methodology

How a site is scored

Every audited site is fetched live with a neutral, headless browser, so JavaScript-rendered content is seen the way an AI crawler would see it. The audit then evaluates:

• On-page structure: HTML, headings, schema markup, metadata, internal links and content architecture.
• AI access: robots.txt rules for the AI search and training crawlers (OAI-SearchBot, ChatGPT-User, Claude-SearchBot, ClaudeBot, PerplexityBot, GPTBot, Google-Extended), plus llms.txt presence.
• External signals: presence and consistency across the sources AI systems cross-reference, including Wikipedia, Wikidata, LinkedIn, Crunchbase and review platforms.

The result is a GEO Score from 0 to 100, computed from 130+ individual AI-visibility signals grouped into eight weighted categories:

The tiers

Scores fall into five 20-point bands, lower bound inclusive:

• AI-Ready (80–100): AI engines can reliably find, understand and trust the site.
• Strong (60–79): well-optimised, with specific gaps remaining.
• Emerging (40–59): foundations present, not consistently citable.
• Low Visibility (20–39): rarely usable by AI engines.
• Invisible (0–19): effectively unreadable to AI search.

Earlier SAVI editions used different band cuts; bands were recalibrated to these 20-point steps in 2026, and each published report states the bands it used. Published editions are historical records: their figures are preserved as published, under the methodology of their day.

Editions and publication

SAVI is published as dated editions: the global index plus sector editions (for example UK accountancy, UK dentistry and UK aesthetic clinics). Every edition states its sample, its date and its band definitions, and sector editions are linked from the SAVI report library. Headline statistics on live pages, such as the data hub, are quoted as percentages with an as-of date.

Freshness and verification

AI visibility is not static: sites change, domains change hands, and scoring signals evolve. SearchScore therefore applies a verification discipline to its published statistics:

• Top-tier re-verification. Before tier-level statistics are published or refreshed, the sites in the affected band are re-audited with fresh scans rather than quoted from stored history. The July 2026 refresh of the live-corpus figures on the data hub followed a full fresh re-audit of the top band.
• Recency limits. Statistics quoted as current reflect recent audits, and each figure carries its own date.
• No survivorship edits. Published editions and dated press materials are never retro-edited; corrections and newer readings appear in newer editions.

Limitations, stated plainly

• The GEO Score is a proxy. No external audit can see inside an AI model. The score measures observable signals that correlate with being found, understood and cited; it is not a readout of any engine's ranking, and no score, ours or anyone's, guarantees citation.
• AI answers are volatile. Independent 2026 research shows the same prompt can cite substantially different sources run to run. Citation outcomes should always be read as rates over repeated runs, never single answers.
• This is our own data. SAVI is built and published by SearchScore. We publish this methodology, our band definitions and our verification practices precisely so the work can be independently scrutinised.

Methodology change log

• July 2026: top-tier re-verification adopted as standard practice before publishing tier statistics; live-corpus figures on the data hub refreshed on fresh scans.
• 2026: tier bands recalibrated to five 20-point steps (80/60/40/20), applied consistently across the audit, reports and this site.