AI Strategy & GEO

Relevance engineering: a working evidence audit

Fernando Angulo
Senior Market Research Manager at Semrush, an Adobe company
10 Min Read
Nov 30, 2024 · Updated Jun 11, 2026

Relevance engineering connects content quality, retrieval and evaluation. The five steps below are my working method, informed by Mike King’s discussion of the discipline. They are not a disclosed training algorithm or a guarantee of citations.


Map the evidence

Start with questions your audience needs answered and evidence you can publish responsibly. Distinguish original observations, external studies and your interpretation. Avoid claiming that a competitor is invisible without a dated comparison.

Explain the claim

Write a clear answer with enough context to preserve its meaning. Use readable headings, accurate identity facts and source links. Google does not require special AI markup, a fixed passage length or exact keyword variants.

Seek corroboration

Independent sources can help a reader check a claim. Earn relevant references through useful research and accurate contributions. No count of independent pages is established here as a citation-probability threshold.

Maintain useful sources

Keep canonical URLs, author information and substantive facts accurate. Update evidence when it changes. A quarterly review may suit your workflow, but it is not a training-inclusion requirement or universal freshness rule.

Observe the result

Measure mentions and citations separately for a dated prompt panel. Record the platform and setup, repeat observations and keep business outcomes distinct. A model’s output does not disclose its internal trust assessment or training history.

Sources and scope

Practitioner analysis by Fernando Angulo. Proposed examples are not measured outcomes.

Mike King / iPullRank: AI Search Manual

Google: current AI search guidance

Google: Generative AI performance report

Recommended Reading

Latest Insights

View all articles