THE TOOLKIT
Appendix C — What the Evidence Says
Every tactic in the book, graded A to F against the published evidence. Where studies conflict, both are named rather than reconciled. Use this table to decide what to fund, what to trial with a measurement plan attached, and what to decline regardless of how confidently it is sold.
From Becoming the Answer by Jeremy Osborn · 347 words
Every tactic in this book, graded. Where studies conflict, both are named.
| Tactic | Evidence | Grade |
|---|---|---|
| Server-side render primary content | Vercel (1.3B crawler requests, 2024); searchVIU (23 crawlers, 69% cannot execute JS, 2025) | A |
| Fix AI-crawler 404s and redirect chains | Vercel: ~35% of ChatGPT and Claude fetches hit 404s | A |
| Correct crawler access control by user agent | First-party documentation from OpenAI, Anthropic, Perplexity, Google | A |
| Product feed completeness and the eight Merchant Center attributes | Google first-party specification, stated AI-surface purpose | A |
| Never seed or sentiment-incentivize reviews | FTC rule effective October 2024; enforcement live 2026 | A |
| Report rolling averages only | St. Gallen: run-to-run source overlap 0.34–0.42 | A |
| Entity consistency and Wikidata | Documented grounding mechanism; strong correlational support | A− |
| /.well-known/ucp manifest | Google first-party specification; Universal Cart live in the US | A− |
| Earned third-party mentions | Ahrefs: strongest correlate; Muck Rack: 82% of citations earned | B+ |
| Digital PR aimed at cited outlets | Muck Rack; third-party listicles ~80% of listicle citations | B+ |
| Publishing prices and specifications | Largest controlled experiment: price presence among the strongest factors, in a synthetic testbed | B+ |
| Agent-completability testing | Agentic traffic composition; 70% product and search routes | B+ |
| Organizing coverage around real question-form queries | Washington University: 6.8x activation on question-form queries (query form, not page formatting) | B |
| Original data, statistics and named sources | Academic GEO study; partially challenged by C-SEO Bench | B |
| Content recency and refresh cycles | Seer: 65% of AI bot hits target content under a year old | B |
| Passage self-containment | Documented chunking mechanism; direct effect unmeasured | B− |
| YouTube presence | Strongest single correlate (r ≈ 0.74); causality unestablished | B− |
| Structural rewriting for chunkability | Largest controlled experiment: negligible. C-SEO Bench: sometimes harmful | C |
| JSON-LD schema as a citation lever | Ahrefs difference-in-differences (n=1,885): −4.6% to +2.4%; Google: no special markup needed | C− |
| Mass-generated permutation pages | Content volume is the weakest correlate measured (r ≈ 0.18) | C− |
| llms.txt | Zero frontier-crawler fetches in a 900-domain, seven-month log study | D |
| Blocking Google-Extended for AI Overview control | Google documentation: does not affect Search inclusion or AI features | D |
| Keyword stuffing | −8%, the worst-performing method tested | D |
| Manipulating AI recommendations | Works, then degrades for everyone; plus regulatory exposure | F |