AI SEO is the discipline of winning both distribution systems at once — Google rankings and AI citations — with a single content operation, and in 2026 the order of operations has flipped: citations come first. For twenty years the playbook was rank, then get traffic. Now a page can be quoted by ChatGPT Search or Perplexity within weeks of publishing, while Google still has the domain in quarantine. If you build for the old order, you spend six months invisible. If you build for the new one, the answer engines carry you while Google makes up its mind.
This is the AI SEO guide I wish existed when this site launched: the dual playbook, tested live. 500k.io is the experiment — a new domain publishing daily, currently earning its traffic from GEO citations while the Google clock runs. Every technique below is deployed on this site, and where the honest answer is “this didn’t do much,” I say so. One disclaimer: answer-engine behavior shifts monthly; treat the specifics as a 2026 snapshot and the principles as durable.
This is the hub for the AI SEO cluster. Each section links to a dedicated deep-dive.
The two systems, and why the distinction drives everything
System 1: search engines rank pages
Google and Bing evaluate pages against queries and rank them. The currency is authority — backlinks, domain age, topical depth — accumulated slowly. A new domain sits in an effective sandbox for 3-6 months regardless of content quality; that’s not a penalty, it’s Google waiting for evidence you’re real. The work that compounds here: topical clusters with pillar hubs, internal linking, technical hygiene, and patience. The terminology landscape around all this is mapped in AEO vs GEO vs SEO — three names, one converging discipline.
System 2: answer engines cite passages
ChatGPT Search, Perplexity, Google AI Overviews, and Claude compose answers and attach citations. The currency is extractability — can this passage be lifted, verbatim or near-verbatim, into an answer? Authority matters far less: how AI engines choose citations breaks down the selection mechanics, and the short version is that a specific, sourced, self-contained paragraph from a DR-0 site beats a vague one from a DR-70 site more often than SEO intuition predicts. That asymmetry is documented first-hand in how to rank in Perplexity with zero domain rating.
One operation, two systems: Google pays out in months on authority; answer engines pay out in weeks on extractability. Build for both simultaneously or lose one.
The citation formula: what makes a passage quotable
Structure: answer first, then explain
Every section that wants to be cited must answer its own heading in the first one or two sentences — the inverted pyramid, enforced ruthlessly. Answer engines extract passages, not pages: a 150-word block that opens with the claim, supports it with a number, and closes with the nuance is the ideal citation unit. Bury the answer in paragraph three and the engine quotes someone else. This is also why a TLDR block at the top of every article is non-negotiable — it’s a pre-packaged answer the engine can lift whole. The complete on-page checklist is in the GEO playbook: how to get cited by ChatGPT and Perplexity.
Specificity: numbers are citation bait
“AI tools lower cost per lead” gets paraphrased without credit. “Documented HVAC campaigns dropped CPL from $115 to $38” gets cited, because engines prefer to attribute specific claims. Every important paragraph on this site carries at least one number, price, percentage, or threshold — not as decoration, but because specificity is the difference between being the source and being the background. Pair each number with where it came from; unsourced precision reads as fabrication to both engines and readers.
Access: the crawlers must be able to read you
GPTBot, PerplexityBot, ClaudeBot, and OAI-SearchBot need explicit permission in robots.txt and server-rendered HTML — a client-side React page is a blank document to most of them. This site allows all major AI crawlers explicitly and ships every page as static HTML. While you’re at it, publish an llms.txt index — with calibrated expectations: the llms.txt counter-position explains why it’s cheap insurance rather than a ranking lever, and why structure and specificity do the heavy lifting.
The formula in one line: answer in 2 sentences, prove with a sourced number, package in a self-contained block, serve as plain HTML to crawlers you’ve explicitly allowed.
The schema stack: machine-readable trust
Schema.org markup is how you hand engines your content pre-parsed — and stacking types on one page is the technique that separates 2026 implementations from 2020 ones. An article here ships Article + FAQPage + BreadcrumbList + Organization + Person in one JSON-LD graph: the article for content, the FAQ for question-matching (the single biggest GEO driver in our testing), the breadcrumb for structure, and the entity nodes so engines know who is claiming what — the E-E-A-T wiring.
Two deep-dives cover the implementation: schema markup for AI search for the seven types that matter and the mistakes that kill visibility, and the schema.org deep-dive for the full @graph architecture you can copy. If you implement one thing this week, make it FAQPage on your ten best articles — it’s the highest citation-per-effort ratio in the stack.
The entity layer deserves its own mention: consistent Organization and Person nodes across every page, linked to real profiles, are how a brand starts being mentioned by engines even without links — and brand mentions vs backlinks argues that mention-building is the new link-building for the answer-engine era.
Measurement: you can’t improve what you don’t probe
Google gives you Search Console; answer engines give you nothing — so you build your own probe. The method: define 30 queries your audience actually asks, run them through ChatGPT, Perplexity, and AI Overviews on a weekly schedule, and log every citation of your domain. Track share-of-voice per query over time. It’s manual-ish (an hour a week, scriptable to less), and it’s the only way to know if your GEO work compounds. The full methodology — query selection, logging format, what counts as a citation — is in the AI citation tracking methodology.
On the Google side, the loop is classic but underused: weekly GSC review of queries sitting at positions 30-50 with low difficulty, then refresh those articles (title, intro, two sections, updated date). Google has already shortlisted you for those queries; a refresh is the cheapest push it responds to.
The honest numbers from this site’s experiment: hundreds of monthly GEO-attributed visits within 60 days of launch, while Google organic remained near zero — exactly the two-clock pattern this guide is built around. Your ratios will differ; the pattern shouldn’t.
Production: quality at cadence without tripping the classifier
Publishing daily AI-assisted content and surviving Google’s helpful-content systems is a solved problem if three constraints hold. Quality gate: every draft scores against a rubric before publish (this site’s threshold: 85/100, human-reviewed below that). Velocity plausibility: ramp cadence like a growing team would — a day-one site publishing 10 articles daily reads as spam to the classifier; 1/day ramping to 3-5/day reads as growth. Original signal: every article carries something that isn’t aggregation — a first-person number, an experiment, a documented failure.
The machinery behind this site’s cadence — pipeline stages, quality gating, the $100/month total compute bill — is documented in the AI SEO playbook, Claude Code SEO at scale, and the content engine process. The production tooling itself is covered in the Claude Code guide — the two pillars are siblings: that one covers the engine, this one covers the distribution.
Cadence is a multiplier on quality, not a substitute for it. The gate (≥85), the ramp (1→5/day), and the original-signal rule are what keep the multiplier positive.
The 90-day AI SEO plan
Days 1-14 — foundations. Technical pass: robots.txt allowing AI crawlers, server-rendered HTML, sitemap, schema stack on every template (Article + FAQPage minimum), llms.txt. Define your 30 probe queries. Baseline everything: GSC positions, zero-point citation count.
Days 15-45 — extractability sprint. Retrofit your best 20 pages with the citation formula: TLDR blocks, answer-first sections, one sourced number per paragraph, FAQ schema. Start the weekly citation probe. Publish new content at a sustainable ramp with the quality gate on. Expect first citations in this window — they arrive before rankings, every time.
Days 46-90 — authority and loops. Build the pillar-cluster architecture (hubs like this page, spokes like the deep-dives it links). Start the mention layer: directories, canonical republishing, community presence. Run the weekly GSC refresh loop on positions 30-50. By day 90 you should see: citations on 5-10 of your 30 probe queries, and the first Google impressions curve bending upward as the sandbox starts to lift.
What this plan doesn’t promise: Google traffic in month one. Nothing legitimate does. The plan’s job is to make sure the months the sandbox costs you anyway are spent building the two assets that pay after it lifts — extractable content and real authority.
The bottom line
The AI SEO guide version of the truth, compressed: you’re optimizing one operation for two clocks. The fast clock (answer engines) pays in weeks and rewards extractability — TLDR blocks, answer-first sections, sourced numbers, FAQ schema, open crawler access. The slow clock (Google) pays in months and rewards authority — clusters, pillars, links, mentions, patience. Every article you publish should feed both: quotable enough to be cited this month, structured enough to rank next quarter.
Start with the deep-dive that matches your bottleneck: getting cited if you have content but no citations, the schema stack if your markup is thin, citation tracking if you’re flying blind, or the production playbook if you can’t sustain cadence.
FAQ
What is AI SEO in one sentence?
AI SEO is optimizing for two distribution systems at once: classic search engines (Google, Bing) that rank pages, and answer engines (ChatGPT, Perplexity, AI Overviews, Claude) that cite sources — with one content operation feeding both.
Is GEO different from SEO?
GEO (Generative Engine Optimization) is the answer-engine half of the job: getting quoted inside AI-generated answers. The overlap with SEO is large — clean structure, real expertise, extractable facts — but the ranking logic differs: answer engines reward quotable passages and specific numbers more than backlinks.
Can a new site get AI citations before it ranks in Google?
Yes, and it's the normal order now. Answer engines cite passage-level quality with far less regard for domain authority — this site earned hundreds of monthly GEO visits while its Google traffic was still near zero, because citations don't wait for the sandbox to lift.
What actually gets a page cited by ChatGPT or Perplexity?
Extractable, specific, sourced claims: a TLDR block up top, self-contained sections that answer in their first two sentences, real numbers with sources, FAQ schema, and clean HTML that crawlers (GPTBot, PerplexityBot, ClaudeBot) can read without JavaScript.
Does llms.txt actually matter?
It's cheap insurance, not a ranking lever. Publish one — it costs an hour — but don't expect it to move citations by itself. The evidence that engines heavily weight it is thin; the evidence that structure and specificity drive citations is strong.
How long does AI SEO take to show results?
Two clocks run in parallel. Citations: first ones in 2-8 weeks if your content is extractable. Google rankings: 3-6 months minimum on a new domain regardless of quality (the sandbox is real). Plan distribution to cover the gap.