An operators' library
Magnitude Authority Engine
The operators' library on AI answer optimization
How pages get lifted, cited, and linked by AI answer engines — documented by the team at Magnitude that builds and runs three production authority sites. Everything here is a rule we actually operate under, enforced by build gates, not a best-practices listicle.
The library
The operator's guide to GEOThe flagship pillar — every operating rule we run in production, in one place.GuidesQuestion-shaped explainers: answer sizing, machine surfaces, schema, gates.GlossaryThe vocabulary of AI answer optimization, one defined term per page.Field reportsWhat actually happened on production builds — traffic shifts, recoveries, incidents.CostsWhat authority builds actually cost, line by line. Computed figures, never estimated.
Latest pages
Guides
Question-shaped explainers — each page answers one question the way people actually ask it, with a liftable answer up top.
AEO vs GEO vs LLMO vs AIO: Five Names, One PracticeAEO vs GEO vs LLMO vs AIO: 4 acronyms plus AI SEO, 1 practice. Where each term came from, who uses it, and how to pick without changing the work.AI Crawlers: The Complete List of Documented BotsAll 12 documented AI crawler tokens from OpenAI, Anthropic, Perplexity, Google, and Microsoft — what each does, robots.txt behavior, and how to verify them.Are AI Citations Worth Anything?Pew: ~1% of AI-summary visits click its links. Seer: cited brands get 120% more clicks per impression. Both are true — here is what a citation is worth.Bytespider: ByteDance's Undocumented AI CrawlerBytespider is ByteDance's crawler — no official docs, no IP list, traffic down 71.45% since July 2024 (Cloudflare). What it does and how to block it.Can Search Console Show AI Overview Impressions? Yes, With LimitsYes. Since Aug 31, 2026, every site has Search Console's Generative AI report: AI Overview and AI Mode impressions by page and country. No clicks, no queries.Can You Trust AI-Visibility Tools' Data?Same prompt, different answers: why 1 sample is noise, the monthly sampling protocol to run instead, and the limits no tracker can fix at any price.ClaudeBot, Claude-User, Claude-SearchBot: What Anthropic's Crawlers DoAnthropic documents 3 agents: ClaudeBot gathers training data, Claude-User fetches when users ask, Claude-SearchBot improves search. All 3 obey robots.txt.Cloudflare's AI Crawler Controls: What Each Setting Actually BlocksSince Sept 15, 2026, Cloudflare's Block for AI training also stops Googlebot, Bingbot and Applebot; Disallow AI Training is the training-only opt-out.Do AI Assistants Read JSON-LD at Answer Time?Testing reported by Ahrefs found AI systems extract only visible HTML at retrieval, ignoring JSON-LD — and its 1,885-page study found no citation lift.Do AI Crawlers Render JavaScript?None of the 3 AI platforms that publish crawler docs mentions JavaScript rendering. Bing does — it says don't hide content behind client-side rendering.Does DefinedTerm Schema Help Glossary Pages Get Cited?Our day-3 glossary page was cited in an AI Overview with DefinedTerm schema — but Ahrefs' 1,885-page test says the definition text, not markup, did the work.Does E-E-A-T Apply to AI Search? Docs vs ProjectionE-E-A-T has 4 components and is not itself a ranking factor, per Google — whose AI-search guide never uses the term. What is inherited vs what gets projected.Does Google Penalize AI Content? Policy vs PracticeNo documented Google penalty targets AI content as a method — the spam line is purpose. Plus our own audit: 40 of 445 AI-assisted claims corrected.Does Schema Markup Help AI Citations? What the Evidence ShowsAhrefs' 1,885-page test found no causal citation lift from adding schema, and Google documents no special markup. Where structured data still earns its keep.Does Updating Content Help It Get Cited by AI?No study on our source list isolates freshness as a cause of AI citations. What we run instead: 182- and 365-day review gates, and a date that never moves.GEO vs SEO: What Genuinely Differs and What Doesn'tGoogle says AI search runs on the same core systems. The 7 things that genuinely differ between GEO and SEO — and the fundamentals that don't change.GPTBot vs OAI-SearchBot: Should You Block Either?OpenAI runs 3 main agents: GPTBot trains models, OAI-SearchBot powers ChatGPT search, ChatGPT-User fetches for users. Blocking each has a different cost.How AI Search Works: How LLMs Choose What to CiteAI answer engines pick citations in 3 stages — retrieval, chunk selection, synthesis. What each stage rewards, per the Princeton GEO paper and platform docs.How Do AI Overviews Behave in YMYL Niches Like Finance and Insurance?No source on our list measures YMYL-specific AI Overview behavior. What changes is the evidentiary bar — 445 claims audited, 40 corrected, 0 fabricated.How Do You Get Cited by ChatGPT?Allow OAI-SearchBot, keep answers in raw HTML, write a 40–75-word liftable passage: the ChatGPT playbook, and what carries over to other AI search engines.How Do You Get Recommended by Perplexity?Allow PerplexityBot and write extractable answers; no study we can source shows freshness drives citations. Expect 186 crawls per click (Cloudflare, 2025).How Do You Show Up in Google AI Overviews?No special markup exists — Google requires indexed, snippet-eligible pages. The 5-part checklist we run, and the 1 AIO citation our fleet has earned.How Long Should a Direct Answer Be?Our contract sets the answer block at 40–75 words and a build gate fails the page outside it. No engine publishes a length — here is why we chose ours.How Many People Actually Use AI Search? The Measured DataPew: 18% of Google searches in March 2025 produced an AI summary. Every AI search usage figure on this page traces to a primary study — nothing laundered.How to Add Source Citations So AI Engines Can Verify Your PageCitation-adding lifted visibility around 40% in the Princeton GEO benchmark. Where citations go, what counts as a source, and how we audit 400+ pages.How to Attribute Value When AI Citations Don't Get ClickedPew: ~1% of AI-summary visits click its links. Seer: cited brands get 120% more clicks per impression. A 3-surface attribution model, no invented multipliers.How to Check Whether Competitors Block AI CrawlersA public-data method: read competitors' robots.txt against the 11 crawler tokens that matter, tabulate who blocks what, and read it as eligibility only.How to Create Content AI Engines Actually CitePublish numbers only you have. The Princeton benchmark measured up to 40% visibility lift from adding statistics — and our own data pages cost 1 audit each.How to Do Demand Research Before You Have Any Search DataSearch Console has nothing to mine on day 0. The cold-start method our 3 builds ran: 280 categorized questions first, real query data by day 4.How to Evaluate AI Visibility Tools: 8 Questions to AskJudge an AI visibility tool by its method, not its dashboard: 8 questions on engines sampled, prompt control, repeat runs, cited URLs and raw exports.How to Fact-Audit Content at Scale: The Refute-First ProcessThe fact-audit process behind 445 checked claims and 461 recomputed figures: ledger every claim, try to refute it from its own source, gate every error.How to Fix False Claims When the CMS Discards Your EditsWhen a page builder re-serializes content and no revision history exists, repair at render — under 1 hard rule: every requester gets the identical page.How to Fix What AI Assistants Say About Your BrandThe 6-step workflow for correcting wrong AI answers about your brand: document, trace the source, fix it, retest — and why training-data answers lag.How to Measure AI Share of Voice Without Paid ToolsA fixed battery of 20–30 prompts, run monthly across 4–5 answer engines, scored in a spreadsheet. A free method that costs an afternoon a month.How to Measure Whether a Change Earned an AI CitationSearch Console's AI report shows pages, not queries, so citation proof is dated manual checks plus GSC proxies — the 6-step method behind our own win.How to Mine Search Console for AI-Shaped QueriesSearch Console returned mineable queries by day 4 on a new domain. The weekly mining method we run on 3 builds: question filters, gap scores, a queue.How to Optimize for Microsoft CopilotMicrosoft has no separate Copilot ranking doc. Its Bing Webmaster Guidelines cover Copilot across 22 sections — and 4 directives decide what Copilot may quote.How to Publish 60+ Pages in Parallel Without ContradictionsParallel content production that held: binding tranche briefs, canonical stances, and 2 gates let one build ship 67 reviewed pages in 2 days.How to Rate Limit AI Crawlers Without Losing CitationsOf the 3 AI platforms publishing crawler docs, only 1 documents a throttle. RFC 9309 has no rate provisions — and a 503 on robots.txt reads as full disallow.How to Retire a Claim So It Cannot Come BackA claim is retired when the library can no longer produce it, not when the pages are fixed. One retired convention leaked back into 6 pages [our data].How to Set Up Article Schema on Editorial PagesArticle schema in 5 steps — and the QAPage mistake we shipped, then reverted across a whole library. Ahrefs' 1,885-page test (2026) found no AI citation lift.How to Set Up IndexNow: The 20-Minute GuideIndexNow setup in 3 steps: an 8–128 character key file at your site root, then GET or POST submissions to api.indexnow.org — up to 10,000 URLs per request.How to Ship Valid Schema Across Hundreds of PagesGenerate markup from 1 frontmatter block, validate rendered HTML in CI. We emit 4 schema types across 400+ pages — scale buys validity, not citations.How to Structure a Content Library for AI SearchCluster design is retrieval design: one question per URL, 5 clusters, glossary as the definitional layer — the shape behind 3 production content maps.How to Structure a Page So an LLM Can Quote ItAnswer block first, takeaways second, question-shaped H2s, 2–4-sentence paragraphs, 1 real table. The page order we ship across 400+ production pages.How to Track ChatGPT and AI Referral Traffic in GA4GA4 files ChatGPT and Perplexity visits under generic Referral. The custom channel group fix in 6 steps, with the regex from our production referrer list.How to Verify Real vs. Fake AI Bots in Your LogsAnyone can send 'GPTBot' as a user agent. Verify against the IP data 3 AI vendors publish, plus Google's 3-part check — step by step, with the failure modes.Internal Linking for a Content Library: Graph Design Against OrphansInternal linking is graph engineering: our naive build left 74 of 176 pages with zero inbound links while 3 pages hoarded 83 each. The fix, as steps.Is AI SEO a Scam? An Operator's AnswerNo. Google's May 2026 guide calls optimizing for AI search 'still SEO.' The scam is the sales layer: guarantees, untraceable stats, fake Google data. 5 checks.Is FAQ Schema Still Worth Adding?Ahrefs' 1,885-page test found no AI-citation lift from schema, yet we ship FAQPage on all 3 of our builds. The honest case for keeping it cheap.Is GEO Worth Doing on a Small Site?On a small site 3 free changes carry most of the available benefit and a tracking subscription carries almost none. Plus when to skip this entirely.Is SEO Dead? What the 2025–2026 Click Data Actually SaysNo — but clicks fell hard: −34.5% CTR under AI Overviews (Ahrefs, March 2025), 8% vs 15% in Pew's panel. What changed, what rebounded, what we reallocated.Is Speakable Schema Markup Worth Adding? Our Negative ResultWe ship speakable markup on every page of 3 production builds and can attribute no effect to it. A negative result, plus the one case where it is worth adding.llms.txt: What It Is and How to Set It UpHow to write and deploy an llms.txt file in about 30 minutes: the proposal's format, a working example, and what our 3 sites' logs say about its value.Nosnippet, Max-Snippet, Data-Nosnippet: The Only Granular AI Overview Controlsnosnippet blocks a page as 'direct input' for AI Overviews and AI Mode; max-snippet caps it by characters. What Google's 4 snippet controls trade away.Organization Schema Example: Setting Up Your Entity HomeA working schema.org Organization example: 1 node with a stable @id, emitted once, referenced by every page's publisher. Plus the 1,885-page null result.PerplexityBot and Perplexity-User: What They Do on Your SitePerplexity documents 2 agents: PerplexityBot builds its search index; Perplexity-User fetches for users and may ignore robots.txt. ~195 crawls per click.RAG for Marketers: The Pipeline Behind AI AnswersRAG grounds AI answers in retrieved web pages. The 3-stage pipeline — retrieve, select, synthesize — and the 2 stages a publisher can actually influence.Robots.txt for AI Crawlers: A Decision FrameworkDecide robots.txt by the bot's job — training, search, or user-fetch — not the company. A per-bot decision table for 4 business models, per RFC 9309.Server Log Analysis: Which AI Bots Crawl Your SiteAnalytics can't see AI bots — server logs can. How to grep for the 8 documented AI crawler tokens, verify them against vendor IP lists, and read the counts.Should Every Page Have a Call to Action?No. On our leasing build 11 of 12 distress-scenario pages carry zero CTAs and end on free-help resources — a trust decision, not a visibility tactic.Should Publishers Block AI Crawlers? A Decision FrameworkCloudflare's July 2025 data: Google crawled 5.4 pages per referred visit, OpenAI ~1,091, Anthropic ~38,000. The block-or-allow decision, by business model.Should You Post on Reddit to Get Into AI Answers?Participate, never manufacture. The sock-puppet play risks your domain, not just an account — plus the 5-phrase monitoring setup we actually run.Should You Publish Experiments That Didn't Work?Negative results are the one content type this niche cannot fake. Our 2 published nulls, the format that keeps them honest, and what they do not prove.The GA4 AI Channel Group Regex, From Our Production Referrer ListThe 9-hostname GA4 channel regex derived from the AI-referral matcher live in our production tagging since 2026-08-12, Semrush's additions, and upkeep.The Princeton GEO Study: Is the +40% Result Real?The KDD 2024 paper reports up to 40% visibility lift inside its own benchmark. What it tested, which levers moved, and why nobody has replicated it.What a Realistic 90-Day GEO Plan Looks LikeThe ordered sequence 3 production builds actually ran — contract and gates first, then map, audit, indexing, measurement, mining — mapped onto 90 days.What Is a Google AI Overview? Anatomy and EligibilityA Google AI Overview is an AI-generated summary with source links shown above results. Eligibility per Google's docs, plus the 4 snippet controls you have.What Is AI SEO? The Umbrella Term, UnpackedAI SEO is 1 practice sold under 5 names: making pages AI answer engines can retrieve, quote, and cite. What it covers, and how to spot a repackaged retainer.What Is Answer Engine Optimization (AEO)?Answer engine optimization (AEO) explained: what answer engines are, how AEO relates to GEO and SEO, and why 1 discipline sits under all the names.What Is CCBot? What Blocking Common Crawl Actually DoesCCBot (CCBot/2.0) is Common Crawl's crawler for a free web archive kept since 2008. Blocking it stops future crawls; pages in past snapshots stay.What Is Generative Engine Optimization (GEO)?Generative engine optimization (GEO) defined in 50 extractable words, where the term came from, and an honest table of what GEO can and cannot prove.What Is Google AI Mode — and How Is It Different From AI Overviews?AI Mode is a conversational Google search you open on purpose; AI Overviews are summaries on a results page. Since Aug 31, 2026, 1 setting opts out of both.What Is Google-Extended — and Should You Block It?Google-Extended is a robots.txt token, not a crawler: it opts you out of Gemini training, not Search. AI Overviews' own opt-out reached all sites Aug 31, 2026.What sameAs Actually Does for Entity RecognitionsameAs points at reference pages that unambiguously identify an entity. Which targets disambiguate, which are noise — and why 0 studies tie it to citations.What Should Robots.txt Say About AI Crawlers? 3 Copy-Paste TemplatesName AI crawlers by job: training, AI search, user fetches. 3 copy-paste robots.txt templates, every token verbatim from vendor docs, checked Oct 2026.When Is a Content Library Finished?Not at a page count. Across our 3 builds the query miner started driving new pages at 188, 298 and 106 files — on day 21, day 14 and day 38.When Reputable Sources Disagree: The Answer-Variance MethodPublish the reconciliation, not a verdict. The registry method: 20 contested questions on one build, 18 on another, all resolved before writing starts.Why Does a Page Rank Well and Get No Clicks?Three causes look identical in Search Console, with opposite fixes. Seer saw organic CTR on AI Overview queries fall from 1.41% to 0.64% (Jan 2024–Jan 2025).Zero-Click Search: What the Numbers Actually SayA zero-click search ends on the results page. Pew measured 8% vs 15% click rates with AI summaries; Ahrefs found −34.5% CTR. The numbers, reconciled.