Getting cited is the new getting ranked.
For twenty-five years, being found online meant ranking a page. Increasingly, it means being the sentence an AI model quotes when it answers someone directly: a different game, with different rules. This page covers what those rules are, where they came from, and exactly what Beyond Visibility measures to help you win at them.
The shift
A search engine hands someone ten blue links and lets them do the reading. An answer engine reads for them, and once it has answered, most people never click through to check its sources.
Only one of these rows has a chance of being you. The question the rest of this page answers: how do you get into it?
AEO and AISO, decoded
AEO and AISO get used almost interchangeably online, but they're not the same question. Beyond Visibility's two products map onto exactly this split.
Answer Engine Optimization
- In plain English
- Getting your content picked as the answer, or cited inside one, when there's no single blue link to click.
- Technically
- Structuring and writing content so it can be extracted as a direct answer, or pulled into a generated summary as a cited source, by a retrieval-augmented system.
- What it asks
- “If someone asks this question, would an AI actually lift an answer from my page, or cite it while answering?”
AI Search Optimization
- In plain English
- The layer underneath AEO: making sure AI systems can find and read you at all.
- Technically
- The prerequisite technical layer (crawlability, structured data, machine-readable metadata) that decides whether a page is eligible for retrieval in the first place.
- What it asks
- “Can an AI crawler even reach my page, parse it, and understand what it is?”
AISO gets you in the door. AEO gets you the answer, and the citation that can come with it. The Web Grader scores AISO's prerequisites: crawler access, schema, metadata. The Content Grader scores AEO: whether what's actually on the page is answer-shaped and citation-worthy.
How we got here: SEO → AEO
None of this happened overnight. Each era optimized for a slightly different signal, and the signal AI answer engines optimize for today is the direct descendant of all of them.
- 1997–2011
Keywords & backlinks
Rank by matching query keywords and counting inbound links. You optimized for the word, not the meaning.
- 2011–2015
Content quality & E-E-A-T
Panda and Penguin punish thin, spun content. Depth, expertise, and trustworthy sourcing start to outweigh keyword density.
- 2015–2018
Mobile-first & voice
Mobile-first indexing and voice assistants push writing toward direct, conversational answers to spoken questions.
- 2018–2022
Featured snippets & answer boxes
Search engines start answering directly on the results page. The boxed answer, nicknamed "position zero," becomes the real prize, not the #1 blue link.
- 2022–2024
Generative answer engines arrive
ChatGPT, Perplexity, and AI Overviews start synthesizing answers from multiple sources instead of listing links. The results page itself starts to disappear.
- 2024–now
AEO / AISO
The target shifts from ranking a page to being the passage an AI model quotes, cites, or paraphrases when it answers someone directly.
What carried over, what's new
AEO isn't a rejection of SEO. Most of the discipline transfers. The target just moved.
Carried over
- Clear, accurate, well-organized writing still wins
- Technical crawlability (robots.txt, sitemaps, fast responses) is still table stakes
- Authority and trust signals still matter, just evaluated per-passage now instead of only per-domain
- Fresh, maintained content is still rewarded over stale content
New in AEO
- The unit of ranking is a 100–300 word passage, not a whole page
- There's no click-through or dwell-time signal: a model either quotes you or it doesn't
- Structured data tells a model what a passage is, not just what it says
- A handful of named AI crawlers decide if you're visible at all, before quality is ever judged
How an answer engine actually reads your page
Every answer engine (ChatGPT, Perplexity, Claude, Google's AI Overviews) runs some version of the same retrieval pipeline. A page has to survive every stage to end up quoted. Miss one, and the rest of the page's quality never even gets evaluated.
Fetch
The crawler requests your page. If robots.txt blocks it, the pipeline stops here: the model never sees a single word.
AI Crawler AccessParse
HTML gets stripped to content. Heavy client-side rendering or a thin text shell behind a JS app can leave nothing worth parsing.
ExtractabilityChunk
The page is cut into passages, typically 100–300 words. Clean heading hierarchy survives this step intact; a wandering wall of text gets chunked badly.
Structure & ChunkingEmbed
Each chunk becomes a numeric fingerprint of its meaning, so it can be matched against a query by similarity, not keyword overlap.
Semantic CoverageRetrieve
The engine pulls the passages closest to the query. Topical completeness, not keyword stuffing, decides whether you're even in the running.
Semantic CoverageRerank
Candidate passages are re-scored for relevance, authority, recency, and structural quality before the model reads them.
Freshness & MetadataSynthesize
The model drafts an answer from the surviving passages. Direct, answer-first writing is far easier to lift into a sentence than a slow windup.
AI Answerability & ClarityCite
If the model quotes or attributes you, a citation or your name appears in the answer. Named sources and verifiable claims measurably raise the odds.
Citations & FactualityThere is no position #1 in a retrieval pipeline, only retrieved or not retrieved. A page that's perfect at every later stage is invisible if the very first stage, Fetch, blocks the crawler.
What Beyond Visibility measures
Every weight below is a live number from the grading engine, not a marketing round-up. It's pulled straight from the same rubric config the grader runs against.
Content Grader
Does each section answer the question up front, in a form an AI can lift straight out and quote?
Clean heading hierarchy, right-sized passages, one idea per chunk: the shape retrieval actually reads.
Named sources, attributed statistics, real quotations: the things that measurably lift citation rate.
Whether the full topic is covered, or an answer engine has to go fill the gaps from a competitor.
Prose an LLM can parse, summarize, and attribute accurately, without filler or hedging.
Claims stated in checkable form, internally consistent, free of red-flag statistics.
Reading-ease scores, sentence length, and scannability: the parseability floor underneath everything else.
Web Grader: Technical Score
JSON-LD present and valid, matching the right type, with the fields that type requires.
Whether robots.txt and meta-robots actually let GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, and Google-Extended in.
Title, description, canonical, Open Graph tags, language, landmarks: the basic contract every page owes a crawler.
A machine-readable publish or modified date, and whether it's actually recent.
Whether the content can be read as text without running JavaScript, and how much of the page is boilerplate.
How a grade actually gets produced
- 1
Ingest
Paste raw text or Markdown, upload a file (.md, .txt, .html, .docx), or point the Web Grader at a live URL.
- 2
Chunk
Content is split into passages the same way a retrieval pipeline would: the unit every check below actually runs against.
- 3
Score
Deterministic analyzers (heading structure, reading-ease formulas, schema validation) run alongside an LLM judge scoring against a versioned rubric: a hybrid, not one black-box model call.
- 4
Aggregate
Category scores are weighted and summed, then two hard caps apply: any category under 30 caps the overall at 75, and detected factual red flags cap Factuality at 40, so a single critical weakness can't be averaged away.
- 5
Recommend
Every failed or weak check becomes a recommendation: the exact passage that caused it, why it matters, a concrete rewrite, and an estimated point impact. Sorted by impact, with the top 3 surfaced as Quick Wins.
What it's worth to you
Content & marketing teams
Stop guessing why a competitor gets quoted and you don't. Every deduction points at an exact sentence and a specific fix, not a vague "improve your content" score.
SEO & growth
AEO adds a new, largely uncontested channel: brand mentions inside AI answers don't show up in a rank tracker, but they show up in traffic and pipeline. This gives you a way to actually see and improve that channel.
Engineering & web teams
The Web Grader's Technical Score turns robots.txt rules, schema gaps, and missing metadata (the kind of thing that silently blocks an entire crawler) into a prioritized, ticket-sized checklist.
Founders & leadership
A single 0–100 number you can track release over release, the same way you'd track page speed. Except this one tracks whether AI can see you at all.
Measured lift in AI-visibility from the Princeton GEO study: 10,000 queries across 25 domains. Every one of these is a scored check in the Content Grader, not an assumption.
What we don't claim
No guaranteed citations
No grader, including this one, can promise a specific AI model will quote you. Scores predict likelihood, not certainty.
Off-page signals aren't measured
Brand mentions on forums (≈4× citation odds by the same research this grader is built on) and review-site presence (≈3×) matter, but can't be read off a single page, so they show up as an informational checklist, not a scored category.
Scores are directional, not absolute
Rubric weights get revisited as retrieval research evolves, so a score is only directly comparable to another score from the same rubric version.
We don't fact-check the live web
Factuality scores internal consistency and verifiability, whether a claim is stated in checkable form, not whether it's actually true.
Grade a page. Takes about a minute.
Paste a URL or an article and see exactly where it stands.