TL;DR. An Evidence Gap is when claims are qualitative instead of quantitative. "Significant improvement" loses to "142% average increase." Content with precise statistics earns up to 40% more visibility than vague claims. Diagnose by scanning your page for unsupported superlatives. Fix by replacing each with a named stat, a dated source, or a direct quote from a named authority.
What the Evidence Gap is
Claims are qualitative, not data-backed. Fix: Add stats, dates, named sources. Up to 40% visibility lift. Include direct quotes from named authorities with full credentials. Reference industry reports, primary surveys, or peer-reviewed studies. Anonymous claims can get ignored.
The 4-part evidence check
A page passes the Evidence test if it contains:
- At least one specific number per major claim. With a unit: percent, count, days, dollars, multiplier.
- At least one named source per section. Named publication, report, study, or authority with credentials.
- At least one dated data point. "2025 Gartner report" beats "Gartner report."
- No unqualified superlatives. "Industry-leading," "best-in-class," "most popular": meaningless unless backed by a stat.
Before and after
Before (Evidence Gap):
"AI search optimization is the best way to improve your visibility. It significantly outperforms traditional SEO for modern brands."
After (Evidence-rich):
"Updated April 2026: According to Gartner, 40% of search traffic will shift to AI engines by 2027. B2B SaaS clients using AI search optimization see a 142% average increase in non-branded organic traffic within six months."
The after version has: a date, a percentage, a named authority (Gartner), a target year, a specific outcome metric, and a time-bounded result. Every sentence lifts cleanly.
Sources by credibility tier
| Tier | Source types | When to use |
|---|---|---|
| 1 (highest) | Peer-reviewed studies, government data | Claims about effects and causation |
| 2 | Gartner, Forrester, IDC, McKinsey, Bain | Market sizing, adoption curves |
| 3 | Named industry publications (TechCrunch, The Verge with a named author) | Trend observations |
| 4 | Your own customer data with sample size disclosed | Outcome claims |
| 5 (lowest) | Anonymous "industry experts", unnamed surveys | Do not use |
Mix tier 1–2 with tier 4. Tier 4 without tier 1–2 reads as self-serving. Tier 1–2 without tier 4 reads as generic commentary.
Customer data disclosure format
If you cite your own data, disclose the sample size and timeframe. Citable:
"Across 400 B2B SaaS customers over 6 months, we observed a 142% average increase in non-branded organic traffic."
Not citable:
"Our customers see great results."
Engines will quote the first. They ignore the second.
E-E-A-T quote example
"As marketing consultant Jane Doe, MCIM, notes, 'UK SMEs that invest in GEO see a 30% faster growth rate.' Corroborated by the Chartered Institute of Marketing's 2025 report."
One sentence, 5 signals: named source, credentials, stat, corroborating authority, dated source. That is the format AI engines extract intact. (More on E-E-A-T in lesson 5.8.)
Do this now:
Open your priority page. Scan it for unqualified superlatives ("best," "leading," "most"). Pick the three worst. Replace each with a named stat from the last 12 months. Ship the change.
Start improving your AI visibility today with Rankscale.
Get started