AI search visibility advice has a coherence problem. One expert says schema markup is everything. Another says it is brand authority. A third says original research. A fourth says you need to be mentioned on Reddit. They are all partly correct. And they are all giving you one pillar of a four-pillar system without telling you it is one of four.

Once the framework is on the table, the contradictions resolve. The advice was never wrong — it was incomplete. Every piece of GEO tactical advice published in the last eighteen months falls into one of four categories. Picking which category you need to work next is the entire job of a competent GEO strategy.

This post lays out the four pillars, what lives in each, how AI engines weight them differently, and the practical framework for deciding where to invest your next month of work.

Pillar 1 — Technical readiness

The prerequisite floor. If AI crawlers cannot read your site cleanly, nothing else matters. You are not in the candidate pool to begin with.

What lives here:

  • Schema markup — Organization, FAQPage, Article, HowTo with valid JSON-LD and accurate sameAs arrays
  • Content structure — question-formatted H2 headings, 120 to 180 word answer blocks, scannable lists and tables that AI extractors can lift verbatim
  • Crawler access — robots.txt permissions for GPTBot, ClaudeBot, PerplexityBot, Google-Extended, OAI-SearchBot, AppleBot-Extended, and Bingbot
  • Entity foundation — consistent business name, founding year, headquarters city, one-line description across Wikipedia (if applicable), LinkedIn, Crunchbase, Google Business Profile, and your homepage
  • llms.txt — the emerging convention some engines look for, a sitemap-equivalent that tells AI which pages matter most
  • Page-level extractability — clean HTML, fast load times, no JavaScript walls that block AI crawler access to actual content

Pillar 1 is what Reffed measures. The audit you can run right now from the homepage scores Pillar 1 in detail because Pillar 1 is the one place where AI search visibility behaves like a system you can audit deterministically: same input, same score, and the score moves when the underlying signals move.

Pillar 1 is also the cheapest pillar to fix. A weekend of focused schema and structure work can take a site from a Reffed score of 40 to 80. The other three pillars do not move that fast.

Pillar 2 — Brand authority signals

The single biggest factor in AI citations after technical readiness. And the pillar most likely to surprise people coming from classic SEO, because the signals here are not the ones backlink tools track.

What lives here:

  • Backlinks from authoritative domains — but quality dominates quantity. One TechCrunch or major industry publication link outweighs a hundred low-authority backlinks.
  • Brand mentions across the open web — even unlinked mentions in articles, podcasts, and social posts count toward entity verification. AI engines have transcript-level indexing of public conversation.
  • Wikipedia presence — AI engines weight Wikipedia disproportionately, and Wikipedia requires real notability under the platform's strict editorial standards. There is no shortcut. Either you qualify or you do not.
  • Structured directory presence — Crunchbase, LinkedIn company page, industry-specific databases. AI engines cross-reference entity claims across these structured sources to disambiguate "is this the real X or just a marketing page?"
  • Press coverage — niche industry publications often outweigh general business press for B2B and B2B SaaS. A feature in a category-specific newsletter is a Pillar 2 win.
  • Brand search volume — multiple 2025 industry studies found this is one of the most strongly correlated signals in LLM citations, outweighing backlinks.

Pillar 2 is the explanation for the most common confusion in GEO: why a site with a Reffed score of 67 like Stripe gets cited by ChatGPT constantly, while a competitor with a score of 90 gets cited rarely. Stripe's brand authority dwarfs the technical gap. AI engines have so many cross-references for the Stripe entity — Wikipedia, news coverage, GitHub, structured directories — that the entity is verified at a confidence level that no amount of schema cleanup will replicate quickly.

Pillar 2 takes months to years to build. It is also the most leveraged pillar over a long horizon. One Wikipedia entry, one press hit, one analyst report can produce a step-function change in citation rate that no amount of Pillar 1 work would generate.

Pillar 3 — Content authority

The pillar where most "content marketing for AI" advice fails. Content authority is not about keyword targeting or post volume. It is about the depth and structure of what you publish.

What lives here:

  • Long-form pillar content — 2,000+ word comprehensive guides on category-defining topics. The kind of content other people cite when they write about your category.
  • Original research and data — proprietary statistics, surveys, benchmarks. The most leveraged Pillar 3 investment available, because original data becomes the citable reference for everyone else in your category.
  • Statistics density — adding statistics to existing content was one of the top-performing interventions in the Princeton GEO research (Aggarwal et al., KDD 2024), among the methods that lifted AI visibility by up to 40%.
  • Expert quotation — adding quotes from credible sources was likewise among the strongest interventions Princeton tested. AI engines are designed to provide evidence-based responses, and direct quotation is the easiest form of evidence to lift.
  • Topical authority clusters — multiple related pages on the same theme signal subject-matter expertise to engines that cross-reference category coverage.
  • Named authors — content attributed to credentialed humans with linked LinkedIn or social profiles gets weighted higher than anonymous corporate content, particularly by Claude and Perplexity.

Content authority is where solo operators and small businesses can punch above their weight. You cannot buy Wikipedia notability. You cannot fake brand search volume. But you can publish one piece of original research with 50 data points that becomes the citable reference in your category — and that single asset can produce more AI citations than your homepage ever will.

The trap on Pillar 3 is producing volume without depth. Ten thin posts will not match one deep one. AI engines are looking for fact density and citable structure, not blog post counts.

Pillar 4 — Community signal

The pillar most underweighted by traditional SEO and most heavily weighted by AI engines that were trained on the open social web.

What lives here:

  • Reddit and forum discussion — AI engines crawl these heavily, especially for product recommendation queries. In multiple 2025-2026 citation studies, Reddit ranked second only to Wikipedia for total AI citation volume.
  • GitHub stars, npm downloads, PyPI metrics — quantitative authority signals for technical products that AI engines treat as verifiable proof of adoption.
  • Product Hunt and Hacker News mentions — high-signal communities with transcript-grade indexing and persistent visibility in AI training corpuses.
  • YouTube and podcast features — transcript indexing means spoken mentions count as much as written mentions, especially for evergreen technical content.
  • Review aggregator presence — G2, Capterra, Clutch for B2B; Yelp, TripAdvisor for local; sector-specific aggregators for niche verticals. AI engines treat aggregator presence as third-party validation.
  • Discord and Slack community presence — increasingly indexed, harder to fake, and an emerging Pillar 4 signal as AI engines extend their training data sources.

Pillar 4 is the slowest pillar to build authentically and the easiest to destroy by trying to manipulate. There is no shortcut. Either you participate in your category's communities for real, or you do not. Reddit moderation is aggressive, AI engines learn to discount obviously promotional content, and one obviously-paid Reddit thread can poison your community signal for months.

How AI engines weight the four pillars

Engines weight the pillars differently. The patterns are observable in citation outcomes but they shift as the underlying models update. The May 2026 best-current-understanding consensus, drawn across the audit data Reffed collects every week:

  • ChatGPT — heaviest on Pillar 2 (brand authority) and Pillar 4 (community signal, especially Reddit and Wikipedia). The hardest engine to win on technical fixes alone.
  • Perplexity — heaviest on Pillar 3 (content authority, fact density, recency) and Pillar 1 (extractability of fact-dense passages). The most rewarding engine for original-research investment.
  • Google AI Overviews — heaviest weighting on Pillar 1 (schema markup is most consequential here) and Pillar 2 (Google's existing entity graph). The engine closest to traditional SEO in signal mix.
  • Claude — heavy on Pillar 2 (Wikipedia and authoritative source verification) and Pillar 3 (named expert quotation). The engine where credentialed authorship matters most.
  • Gemini and Copilot — inherit weighting from Google and Bing indexes respectively, layering their own synthesis on top. Pillar 1 dominant in both, Pillar 2 secondary.

No engine weights any pillar at zero. A site that is strong on one pillar and weak on three will get cited intermittently — visible in some engines, invisible in others, with citation patterns that look random. Sustained visibility across all six major AI engines requires real presence across all four pillars. There is no single-pillar shortcut.

The framework for deciding what to work next

When you can see the four pillars clearly, the strategic question becomes simple. Where is your business weakest? That is what you work next.

A diagnostic that takes ten minutes:

  • Pillar 1 score: run a free Reffed audit. Anything under 70 is the obvious bottleneck. Fix this before anything else.
  • Pillar 2 score: count your Wikipedia entry (yes or no), top-tier press hits in the last 12 months, structured directory presence (Crunchbase, LinkedIn, industry databases), and brand search volume trend. If you have one Wikipedia entry plus one major press hit plus a Crunchbase profile, you have a baseline. If you have none of those, Pillar 2 is your real gap.
  • Pillar 3 score: count your pillar content pieces over 2,000 words on category-defining topics with original data or expert quotation. Three or fewer is thin. Ten or more starts to build topical authority.
  • Pillar 4 score: count your category's top three communities (probably Reddit, plus one or two others). Do you have authentic, non-promotional presence in each? Do people mention you organically without your prompting? If the answer is no, Pillar 4 is your gap.

The pillar with the lowest score is your bottleneck. Work that pillar for a full quarter before moving to the next. Most businesses try to advance all four simultaneously and end up moving none of them. Pick one. Move it. Then move the next.

Where this leaves Reffed in the picture

Reffed measures Pillar 1 deeply and reports Pillar 1 weakness in detail. The free audit and Reffed Watch both score schema markup, content structure, crawler access, FAQ extractability, llms.txt presence, and the engine-specific patterns that govern technical extractability. We surface a mention-check that gives directional read on Pillars 2 and 4, but the deep work on building authority, content depth, and community presence happens outside the audit.

Here is the honest framing: Reffed solves the prerequisite. The other three pillars are work you do over months and years across your business — and that is what Reffed Academy teaches. Foundations is free, walks through the framework end-to-end, and includes a version of this lesson with worked examples. Start with Foundations if you want the full operating model.

If you are coming to AI search visibility from an existing SEO background and want a working framework that resolves the contradictions in the advice you have been reading, the four pillars are it. Score your business on each pillar honestly. Pick the bottleneck. Spend a quarter on it. Then re-audit.