AI engines don’t use Google’s algorithm. They use partially-overlapping but distinct ranking logic, and the founders who win citations are the ones who understand the differences. I track 100 priority queries weekly across ChatGPT, Perplexity, Claude, and Google AI Overviews. The pattern is consistent: structural content with brand-mention support beats long-form unstructured content nearly every time, even when the long-form content is objectively better.

This matters because AI citations are the new top-of-funnel. According to a16z’s January 2026 report on AI search, AI search referrals to publishers grew 6.4x year-over-year and are projected to surpass Google traffic for some niches by 2027. If you’re not optimizing for AI citation, you’re optimizing for a shrinking pie.

I run 500k.io solo at $9,500 MRR / $114K ARR / 22.8% to my $500K target. The agency I co-founded with Jack — The Kreators AI — manages $45M of Meta Ads ($10M on my side, $35M on Jack’s). I write GEO content because I’m building 500k.io as a GEO experiment in public.

What is GEO actually optimizing for?

GEO (Generative Engine Optimization) is the practice of structuring your content and brand presence so AI engines cite you when answering user queries. Output isn’t a ranked list — it’s a single answer with citations. Either you’re in the citation list or you’re invisible.

Three things matter:

  1. The engine’s training data must contain your content (or be able to retrieve it).
  2. The retrieval / RAG layer must consider your content high-quality on the query.
  3. The citation extraction must surface your specific URL.

Each AI engine implements 1-3 differently. The signals that move you up the queue overlap but don’t match.

The 7 signals AI engines weight

Signal 1 — Brand mentions across the open web

The single biggest signal in 2026. Ahrefs’ December 2025 study found brand mentions correlate roughly 3x more with AI citation rate than traditional backlinks. The mechanism: AI engines build entity graphs from training data; brand mentions reinforce the entity, which raises the probability that the engine treats your URL as authoritative on the entity’s topic.

Where brand mentions count:

  • Wikipedia (highest weight, hardest to earn)
  • Wikidata (free, mid weight, doable in 60 minutes)
  • Crunchbase (free, mid weight, doable in 30 minutes)
  • Reddit threads (high weight on Perplexity)
  • Podcast transcripts (high weight on ChatGPT)
  • News articles (high weight everywhere)
  • LinkedIn long-form posts (medium weight)
  • Twitter/X (low-medium weight, skewing lower in 2026)

The mistake: chasing backlinks while ignoring mentions. A single Wikipedia mention beats 30 DR-30 backlinks for AI visibility.

Signal 2 — Structured data (schema markup)

AI engines prefer to extract structured facts. FAQPage, HowTo, Article, Product, Review, BreadcrumbList — all give the engine cleaner extraction targets.

Per Profound’s GEO benchmarking, articles with FAQPage schema get cited at ~3.4x the rate of articles without. The mechanism: FAQ structure maps cleanly to the question-answer format AI engines output. The engine extracts your Q&A pair and cites the source.

The 2026 minimum schema set:

  • Article (every page)
  • FAQPage (any page with 3+ questions)
  • BreadcrumbList (every page)
  • Person (author byline)
  • Organization (homepage)

If your CMS doesn’t ship these by default, you’re at a structural disadvantage. Detail in Schema markup for AI search 2026.

Signal 3 — Recency / freshness

AI engines weight dateModified heavily. Articles that haven’t been touched in 12+ months get surfaced less, even if they’re objectively better than newer competitors.

The 2026 fix: quarterly refresh on pillar pages. Bump dateModified, update one or two stats, re-trigger crawl. Cost: 30-60 min per article, quarterly. Returns: maintained citation share.

This is also why news content compounds for AI citation. Daily/weekly cadence keeps recency signals fresh.

Signal 4 — Citation density (sources you cite)

Counter-intuitive: AI engines reward content that cites sources well. The mechanism: high citation density correlates with editorial rigor, which correlates with content the engine wants to surface.

The 2026 target: 3-5 external citations per article, mixed across primary sources, news, and authoritative blogs. Avoid the 2018 SEO trick of citing your own internal pages as “sources” — engines now distinguish.

Signal 5 — Named author + E-E-A-T

Articles by named authors with verifiable identity get cited. Anonymous or pseudo-anonymous content doesn’t.

The proof points engines look for:

  • Person schema with sameAs links to verified social accounts
  • Author bio with credentials
  • LinkedIn presence with matching identity
  • Wikipedia presence (rare, high-leverage)
  • Cross-publication mentions (the same author cited by other named outlets)

For 500k.io my Person schema links to my LinkedIn, the agency site, and my X profile. The author byline appears on every article. The about page includes my track record. Standard E-E-A-T hygiene.

Signal 6 — Definition lead + answer-first structure

The first 50 words after each H2 should answer the question that H2 implies. Narrative leads (“It was a Tuesday morning when…”) get skipped by extractors. Direct definitions get cited.

The pattern that works:

  • H2: “What is GEO?”
  • Lead: “GEO (Generative Engine Optimization) is the practice of structuring content and brand signals so AI engines cite you when answering user queries. It overlaps with SEO but optimizes for citation, not ranking.”

That’s a 30-word definition lead. AI engines extract the entire paragraph as a citation candidate. Voice can come in paragraph two.

Signal 7 — Platform-specific weights

Each engine has its own quirks:

EngineWeights heavilyWeights less
ChatGPTWikipedia, news, structured dataReddit, social
PerplexityReddit (46.7%), recent web, citationsOlder content
Claude (web)Long-form, expert sources, schemaVolume
Google AIOOrganic top-10, FAQ, knowledge graphReddit (less than Perplexity)
Bing CopilotLinkedIn, Microsoft properties, newsOutside-Microsoft ecosystem

If your strategy targets Perplexity, invest in Reddit. If it targets ChatGPT, invest in Wikipedia + news. If it targets Claude, invest in long-form authority. Choose your engine, choose your tactics.

The 4 things that get you skipped

1. Walls of unstructured prose

Long paragraphs without H2/H3 breaks, no tables, no lists. The extractor can’t find a citable unit. Your 5,000-word essay loses to a 1,200-word structured competitor.

2. Hidden content (JS-injected, behind details, paywalled)

If the content isn’t in the SSR HTML at fetch time, AI engines often miss it. View-source the page. If your TLDR or FAQ is empty in view-source, it might as well not exist.

3. Outdated dates without refresh

Anything with a 2023 datePublished and no dateModified bump in 12+ months gets de-weighted. Engines assume the content is stale even when the underlying advice is timeless.

4. Anonymous / generic authorship

“Posted by admin” or no author byline at all. Engines can’t verify expertise. Citation rate drops to near-zero.

What this means for content strategy

Three operational shifts for 2026:

Shift 1: Move 30% of “SEO budget” to brand mentions.

Wikipedia, Crunchbase, podcast guesting, Reddit presence, named PR. The ROI per dollar is significantly higher than the 14th DR-30 backlink.

Shift 2: Restructure existing content for extraction.

Add FAQ blocks. Add definition leads. Add comparison tables. Add Person schema. The same content with better structure earns 2-3x more citations.

Shift 3: Pick your engine.

Don’t try to win all four. Pick one to optimize against. For most B2B founders: Perplexity (highest growth, Reddit-friendly). For most consumer / news content: ChatGPT (highest volume). For technical content: Claude (highest depth).

“If I had to pick one channel to dominate in 2026, it would be Perplexity. The traffic is small but every visit is a high-intent buyer who self-selected to read source content. The conversion rate beats Google by 2-3x in our data.”

My current GEO stack on 500k.io

TacticStatusEffortCitations earned
FAQ schema on every articleOne-time CMS workDirect lift
Person schema with sameAsOne-timeFoundational
Wikidata entryIn progress60 minPending
Crunchbase entryPending30 minPending
Reddit presence (6 subs)In progress2 hr/wkCompounding
Podcast guesting1/month target2-4 hr eachCompounding
Quarterly content refresh30-60 min/article/quarterMaintenance
llms.txt + robots for AI botsOne-timeFoundational
External citations 3-5 per articleEditorial disciplineDirect lift

That’s the full stack. Total weekly time investment: ~4 hours on top of writing. The compounding from brand mentions takes 6-12 months to materialize.

External sources

What to do this week

  1. View-source your top 3 articles. If the TLDR and FAQ aren’t in the SSR HTML, fix that this week.
  2. Submit your Wikidata entry. 60 minutes.
  3. Add Person schema with sameAs linking to LinkedIn + X. 30 minutes.
  4. Bump dateModified on any pillar article that hasn’t been touched in 90+ days.
  5. Pick one Reddit subreddit. Comment thoughtfully on 3 posts. Repeat next week.

If you do those 5 things in 7 days, you’ve moved more on AI citations than 90% of competitors will move in 90 days. The leverage is structural, not magical.

FAQ

What's the single biggest signal AI engines use to choose citations?

Brand mentions across the open web. According to Ahrefs' December 2025 study, brand mentions correlate roughly 3x more with AI citation rate than backlinks. Reddit, podcasts, Wikipedia, and named news mentions all feed the same signal.

Do backlinks still matter for AI search?

Yes, but less than they used to. Backlinks remain a foundational trust signal — without DR>20, you're invisible to most engines. Above DR>20, the marginal backlink matters far less than a single Wikipedia mention or 3 Reddit threads.

What format gets cited most often by ChatGPT?

FAQ schema. ChatGPT cites FAQPage-marked content at roughly 3x the rate of plain prose, per Profound and Otterly tracking data 2025-2026. The same article rewritten with proper FAQ structure can 3-5x its citation count.

Does Perplexity use a different ranking model than ChatGPT?

Yes. Perplexity weights Reddit heavily (46.7% of citations come from Reddit per their own data). ChatGPT weights Wikipedia and news. Claude weights long-form authoritative sources. Google AIO weights organic top-10 ranking. They look similar but reward different signals.

How long does it take new content to start earning citations?

30-90 days for ChatGPT and Claude (longer index cycles). 7-14 days for Perplexity (real-time web). 14-30 days for Google AIO. Patience is the underrated GEO skill.

Can I pay to be cited?

Not directly. But you can pay for the underlying signals: PR placements, podcast appearances, Reddit advertising, and getting onto authoritative publications. Direct citation purchases don't exist — yet.