Two Studies. One Answer. Why Entity-First SEO Is the Only LLM Strategy That Works.

18 May 2026 · Chris Dolan

Two pieces of research landed this month that, read together, paint a clearer picture of AI search visibility than anything published in the past two years. The first was Google's official AI Optimization Guide, released May 15th. The second was a large-scale empirical study from OPP Alerts — 105,000 LLM prompts, 145 industries, 1.1 billion crawled web pages, measuring which signals actually correlate with being recommended by ChatGPT.

Google told you what to do. The OPP Alerts data told you why it works. Here's how RankingFactory is built around the signals that both agree on — and where the gaps are.

What the OPP Alerts Study Actually Measured

Most "LLM SEO" research is anecdotal. This wasn't. The OPP Alerts team generated buyer personas across 145 industries, asked ChatGPT who it recommends, then ran Spearman rank correlations between 13 measurable signals and those recommendation scores. The result is the most credible ranking factor data available for AI-driven search.

The top signals, in order:

• Search engine appearances — how often a domain shows up in Google results (ρ=+0.241, STRONG)
• Best search engine rank — the highest position that domain reaches in SERPs (ρ=+0.238, STRONG)
• SE outbound links — how many search result pages link out to the domain (ρ=+0.230, STRONG)
• Backlink count and authority — total links and PageRank (ρ=+0.204 and +0.200, STRONG)
• Common Crawl web presence — how broadly the domain appears across the crawled web (ρ=+0.123, CONFIRMED)
• Wikidata — whether the entity has a structured knowledge graph presence (ρ=+0.120, CONFIRMED — and DOMINANT in 30+ individual industries)
• Reddit signals — genuine community mentions (ρ=+0.111 comments, +0.096 posts, CONFIRMED/EMERGING)

The headline: if you rank in Google, you get recommended by ChatGPT. They draw from the same indexed web. Google said the same thing in different words — "our generative AI features rely on our core Search ranking systems." The OPP Alerts data puts numbers on it.

Where RankingFactory Lines Up

The Top Three Signals — Covered by Design

SERP appearances, SERP rank, and SE outbound links are the three strongest signals. They're also the direct output of what SmartStack does.

Every campaign RankingFactory runs has one goal: get the client's target pages ranking higher and indexed more reliably. Higher rankings mean more SERP appearances. More appearances in more search results means more outbound links from SERPs pointing at the domain. The entire signal chain — the three strongest predictors of LLM recommendation — falls out of the same strategy. This isn't alignment by coincidence. It's alignment by logic.

Backlinks and Authority — Partially Covered

Backlink count and authority are STRONG signals. RankingFactory builds links through Google-owned properties — Drive files, Google Sites, Blogger posts, Cloud pages. These are real, indexed, publicly accessible pages on google.com-class domains. They contribute meaningfully to backlink count and authority.

The honest caveat: they're all from the same domain ecosystem. True link diversity from independent third-party domains is stronger. SmartStack builds the foundation — clients who layer earned links or PR on top will see compounding returns on the signals this data says matter most.

Common Crawl Presence — Covered

Common Crawl measures how broadly a domain appears across the general crawled web. Every Drive document, Cloud page, Blogger post, and Google Sites page created by SmartStack is publicly accessible and captured in Common Crawl snapshots. Each campaign deployment expands the client's CC footprint. This is direct, measurable coverage of the sixth-ranked signal.

Wikidata — The Most Underrated Signal in the Study

Here's the finding that most people will skip past and shouldn't.

At the global level, Wikidata shows a CONFIRMED correlation of ρ=+0.120. That sounds modest. But when you look at individual industries, Wikidata is DOMINANT — the single strongest predictor — in over 30 of the 145 sectors studied. The correlations in those industries are extraordinary:

• Homeowners insurance: ρ=+0.872
• Collision-repair centers: ρ=+0.723
• ERP software: ρ=+0.655
• Healthcare IT: ρ=+0.575
• CRM software: ρ=+0.577
• Hotels & resorts: ρ=+0.650
• Furniture stores: ρ=+0.706

These aren't weak correlations in niche categories. These are among the strongest signals measured in any industry in the entire study — and they're all Wikidata.

What this means practically: ChatGPT and other LLMs are trained on knowledge graph data. When a business has a verified Wikidata entity, the LLM can confidently identify and recommend that entity because it has a structured, cross-referenced record. Without it, the business is just a website. With it, the business is a known entity — a thing the model has seen defined, described, and linked across its training data.

RankingFactory's Entity Stack is the only feature in the platform that directly targets this signal. It walks users through creating their Wikidata item, validating their Q-number, and connecting it to their Google Sites Entity Hub and schema markup. Most SEO platforms don't touch Wikidata at all. The OPP Alerts data says that's a significant miss — especially for any client in a B2B, healthcare, financial services, or brand-heavy vertical.

This is also why the Entity Stack step comes first in the onboarding flow. Before any signal deployment, before any campaign, before any content — the entity needs to exist. Everything else references it.

The One Gap: Reddit

Reddit is a CONFIRMED signal globally (ρ=+0.111 for comments) and DOMINANT in specific industries — Enterprise AI platforms (ρ=+0.528), personal banking (ρ=+0.545), fertility clinics (ρ=+0.358). For clients in those verticals, authentic Reddit presence measurably improves LLM visibility.

RankingFactory doesn't automate Reddit, and it won't. Google's own guide explicitly flags inauthentic community mentions as a tactic their spam systems target. The signal is real — but the only version of it that works long-term is genuine. A brand that actually participates in relevant subreddits, answers real questions, and builds a reputation in communities earns this signal naturally. A brand that manufactures it gets caught.

The practical recommendation for any client in a Reddit-sensitive industry: build a real presence. Create a brand account. Answer questions in your niche. That's a 30-minute-a-week activity that compounds over months into one of the stronger LLM signals in your category. SmartStack handles the technical foundation — Reddit is the one signal that has to be earned by humans.

What Both Studies Agree On

Google's guide and the OPP Alerts data converge on the same answer from different directions.

Google says: rank well, be technically sound, have a real entity with real expertise. The empirical data says: the strongest predictors of LLM recommendation are SERP presence, SERP rank, and a coherent entity footprint across the indexed web — with Wikidata as the highest-leverage single signal in industry after industry.

The entity-first architecture isn't a theory. It's the correct answer, measured across 105,000 prompts and 1.1 billion web pages.

RankingFactory was built around that architecture before this data existed. The data now confirms it.

Want to see where your entity currently stands against these signals? The free AI Visibility Audit checks your Wikidata status, schema coverage, indexation, and GBP presence — the exact signals this research says matter most for LLM recommendation.