Purpose and the 90+ Quality Standard

This guide has one goal: producing content that ranks on Google, Bing, and AI search engines, that readers trust, and that the industry cites as a reference. Our target is numeric and unambiguous: every article must score at least 90 out of 100. A writer or editor who reads this guide from start to finish should be left with no open questions; if a question remains, the gap is ours. Ask your editor and we will add the answer to the guide.

How an article is scored (the 100-point system)

Before publication, the editor scores every article against this table. 90 or above: it publishes. 80-89: it goes back for revision. Below 80: it is rejected and re-planned from scratch.

What 90+ means in practice

• A reader who knows the topic should finish the article having learned something new.
• When a skeptical editor picks any sentence and asks "what's the source for this?", the answer should be in the text.
• When an AI search engine lifts any paragraph as a citation, that paragraph should remain accurate and meaningful on its own.
• When an editor at a rival publication reads it, they should think: "I wish we had written this."

How to use this guide

1. Every writer and editor reads this guide once, completely, start to finish.
2. The writer follows the structure and requirements table in Chapter 3 for the article's content type.
3. Every article goes through the checklist in Chapter 12 before publication. No exceptions.
4. If any item marked "P0" fails, the article does not go live. This is not up for debate.
5. The editor scores the article on the 100-point table; anything below 90 does not publish.

The Golden Rules

Even if you forget the rest of this guide, keep these ten rules. Each one closes off one of the most common failure patterns that push crypto publications out of the rankings.

YMYL & E-E-A-T: Why Crypto Plays in the Hardest League

In Google's eyes, crypto content is not ordinary content. It sits in the finance category and is held to the highest quality bar that exists. You cannot write in this industry without understanding that.

What is YMYL?

YMYL = Your Money or Your Life. Google applies this label to topics that can affect a person's health, money, safety, or wellbeing, and it evaluates these pages against its strictest quality standard. Crypto content lands squarely in the "your money" half:

• Price analysis and predictions → readers may trade on them.
• Project reviews → readers may invest based on them.
• Regulation news → readers may make legal or financial decisions because of them.

Since the 2018 "Medic" update, every core update has hit health and finance sites harder than everyone else. The March 2026 and May 2026 core updates followed the same pattern. In plain terms: the sloppiness an average food blog gets away with will kill a crypto site.

E-E-A-T: four letters, four questions

Google's three questions: Who / How / Why

Google's own guidance asks three questions of every page. Every article we publish must answer all three clearly:

• Who wrote it? A visible byline, a clickable author page, and a bio with real credentials. "X is a journalist with a passion for the crypto world" is not a credential; it's a motivation statement. "X has covered derivatives markets since 2019 and previously worked as an analyst at exchange Y" is a credential.
• How was it made? Where did the data come from? What tool verified it? If AI was used, how? Since September 2025, Google's quality raters have been explicitly assessing whether content appears AI-generated.
• Why does it exist? Does this page exist to help a reader, or to catch search traffic? Publishing a batch of same-template articles on the same day is the worst possible answer to "Why."

The author page standard

• Real name and real photo. A pen name is allowed only for safety reasons and with editor approval, and even then a real, verifiable profile must stand behind it.
• Verifiable credentials: past roles, publications, LinkedIn/X profiles.
• A stated specialty, as specific as "DeFi and stablecoin regulation."
• Person schema markup with sameAs links to social profiles.
• Conflict-of-interest disclosure: if the writer holds the asset they're analyzing, the article says so.

Article Anatomy: The Ideal Structure

Fixed templates (TOC → H2s → Glossary → Conclusion → FAQ → bare references → Disclaimer) are not used. In their place is a structure that flexes with the content type but always carries the same quality principles.

Content types and their mandatory requirements

Decide the article's type before you start writing. Every type has its own baseline requirements; if any element marked as mandatory in this table is missing, the article does not go to the editor.

Headline (H1) and meta

• The headline must carry the claim without becoming clickbait: never promise in the headline what the article doesn't prove.
• The primary key phrase appears once in the headline and once in the first 100 words, naturally. Don't force it.
• Meta description, 150-160 characters: it summarizes what the page actually contains. If it promises an answer, the answer is really inside.
• URLs are short and permanent: no dates, no filler words.

The intro: the first-60-words rule

The answer to the reader's question (and the AI engine's) goes in the first paragraph. This is called answer-first writing, and it's the single strongest technique for both Google AI Overviews and Bing Copilot citations.

Body structure

• H2 headings should be statements, not a stack of questions. An article where all six H2s are questions looks like an FAQ template leaking into the body, which is a classic AI-production signal. A question heading is fine occasionally (especially when it mirrors a real search query), but never wall-to-wall.
• Each H2 section covers one sub-topic and delivers its main finding in the first sentence or two.
• Paragraphs run 2-4 sentences. No walls of text.
• Comparisons get tables, processes get numbered lists, data gets charts. Research shows pages with lists, tables, and statistics appear 30-40% more often in AI answers.
• 3-5 internal links per 1,000 words (relevant, with descriptive anchors) and at least 2-3 external authority links.

Data and visuals are mandatory

• Price or market analysis: an annotated chart is required (source such as TradingView, timeframe visible).
• On-chain analysis: a dashboard screenshot plus the data provider's name and the date range are required.
• Every image gets meaningful alt text. Publishing images as bare "image.webp" placeholders is banned.

The close

The "In conclusion" formula and generic "time will tell / we'll keep watching" endings are banned. A good close does one of three things:

• Gives the reader specific metrics to watch ("the two data points to track are the ETFs' daily net flows and total assets under management").
• States what would invalidate the analysis ("a daily close below the $63 support cancels this scenario").
• Points to the next known date ("ESMA's transition period ends July 1; that's when the real decision lands").

Banned template sections

Sourcing & Research Rules

The most common and most expensive mistake in crypto media is single-source reporting: no independent, verified primary source beyond the covered company's own announcement. This chapter fixes that permanently.

The source hierarchy

Hard rules

• Minimum source counts: News = 3+ independent sources. Analysis = 2+ data sources plus a stated method. Reviews = first-hand testing plus 2 external sources.
• Every number gets a date: not "according to a recent study" but "according to RWA.xyz data from June 26, 2026". Bing's AI grounding system treats undated figures as unreliable.
• Chase every statistic to its primary source: if CoinDesk mentions a report, find the report itself and link to it. Quoting a quote of a quote multiplies errors.
• Cross-verify: no big claim ships on a single source. If Lookonchain shows a transfer, confirm it on Arkham or a block explorer.
• Cross-page consistency: if two of our articles give different figures for the same thing, AI engines may stop citing both. When new data lands, update the older article.
• Quote integrity: when quoting an executive, reproduce the words exactly and link the source. Paraphrasing in your own words and wrapping it in quotation marks is banned.

The on-chain data standard

• Name the data provider: Glassnode, CryptoQuant, Santiment, Nansen, Dune, RWA.xyz…
• State the query date and the time window.
• Define thresholds: if you write "transfers above 1M tokens", say whose classification that is.
• Respect the limits of interpretation: rising exchange inflows "may signal selling"; they do not prove selling happened. Transfer size is not proof of institutions (it could be an OTC desk, a custodian, or an exchange's internal shuffle).

Writing Rules & Crypto Language

Two targets: (1) prose clean enough to have come from a native-speaking editor, and (2) flawless use of crypto terminology. A failure in either one collapses our claim to expertise.

Banned AI clichés

The following phrases are the AI-writing signals that detectors and experienced editors catch fastest. All of them are on the ban list:

Jargon accuracy table

Real mistakes commonly seen in crypto media, with corrections:

Commonly confused concepts

• Market cap ≠ money that flowed in: not "$1.7B entered the token" but "the market cap reached $1.7B".
• Circulating / total / max supply: three different things; say which one you mean.
• Inflow ≠ selling: exchange inflows are selling potential, not confirmed selling.
• Burn ≠ deflationary: burning alone may not reduce net supply; weigh it against issuance.
• TVL ≠ safety or success: locked value by itself says nothing about protocol quality.
• Approval ≠ endorsement: a regulator's approval is not a recommendation; report its actual scope.

Language mechanics

• Average sentence length 15-20 words; never three long sentences in a row.
• Prefer the active voice: "the FSA approved it", not "approval was granted by the FSA".
• No em dashes (the long dash) and no emojis, anywhere in the article. Replace a long dash with a comma, a colon, or a period. Heavy long-dash and emoji use is one of the fastest AI-writing tells there is.
• Use the same phrase at most 4-5 times per article; beyond that it reads as mechanical repetition.
• Consistent price precision: the same level cannot be $2.04 in one paragraph and $2.0440 two paragraphs later.
• Consistent number formatting: "$1.7B" or "$1.7 billion": one style per the style guide, never both.

AI Usage Policy

Our core policy is simple: articles are not written by AI. Every article has a human writer. AI is not a replacement for the writer; it's an assistant that speeds up the writer's research. The boundaries below protect that principle.

What's allowed, what's banned

The writer's declaration

With every submission, the writer confirms the following (checklist group F):

1. I wrote the entire text myself; no sentence was taken or paraphrased from AI output.
2. I used AI only for research support and post-writing language checks.
3. I checked every figure, name, date, and quote against its primary source.
4. I read the full text aloud before submitting it.

Technical SEO & Schema

The minimum technical knowledge for writers, and the configuration standard for the site team. AI engines only cite pages they can crawl, render, and index; technical health is the price of entry.

The schema standard

Technical items the writer owns

• One H1; an H2 → H3 hierarchy with no skipped levels.
• Descriptive alt text and compressed files (prefer WebP/AVIF) for every image.
• Internal links with descriptive anchors ("click here" is banned).
• External links point to the specific page, never the homepage.
• datePublished and dateModified reflect reality; refreshing the date without updating the content (fake freshness) is a spam signal.

Standards the site team owns

• Core Web Vitals targets: LCP < 2.5s, INP < 200ms, CLS < 0.1.
• Content is server-side rendered; the main text must be readable without JavaScript (many AI crawlers don't execute JS).
• Confirm robots.txt doesn't block AI crawlers: Googlebot, Bingbot, GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot. (Check CDN/WAF rules too; Cloudflare bot protection can block them silently.)
• XML sitemap stays current with accurate lastmod values.
• Duplicate content and canonicals: every page carries the correct canonical tag (self-referencing where appropriate). Per Google's July 2026 guidance, pages can stay in a duplicate cluster for up to 2 weeks after a fix, so be patient after correcting rather than stacking repeated fixes.
• llms.txt: optional and experimental. Google doesn't use it; it costs nothing to publish, but it substitutes for nothing.

GEO: Visibility in AI Search

A growing share of traffic now comes from AI answers rather than blue links: Google AI Overviews and AI Mode, ChatGPT, Perplexity, Copilot, Gemini. On these surfaces the goal isn't "ranking"; it's being cited as the source of the answer.

Facts first

• Google's official AI guidance, published in May 2026, says AI Overviews and AI Mode run on the same foundations as classic Search. No special AI rewrites, no special passage lengths, no special schema.
• 2026 research shows 54% of AI Overview citations overlap with Google's top 20 organic results, but 48% come from outside the top 100. Classic SEO helps, but it isn't sufficient on its own.
• Pages with lists, tables, and statistics appear 30-40% more often in AI answers; multi-modal content (text plus images, video, and structured data) shows the strongest correlation with citation.
• As of July 2026, Gemini 3.5 Flash is the default model behind AI Mode; for most queries the primary output is now an AI answer, with classic blue links pushed below the fold. Citation share is the new baseline metric.
• Search Console's Generative AI Performance reports track our impressions in AI Overviews and AI Mode, and AI systems distinguish between a brand being "used" (cited) and merely "mentioned". The editorial team reviews this report monthly.
• Google is adding AI image generation to AI Overviews; original, high-quality visuals still improve the odds of being cited.

Six techniques for citable writing

1. Answer first: the first 30-60 words of every H2 section should answer that section's question on their own. AI engines lift passages out of context; each passage has to stand alone.
2. Build quotable sentences: clear, data-carrying sentences of the form "X happened on date Y, and it means Z" get cited. Vague sentences with three layers of hedging don't.
3. Attach a date and source to every fact: "according to RWA.xyz data from June 26, 2026…". AI systems reuse sourced, dated sentences with confidence.
4. Think in question-answer pairs: the concrete questions a reader would ask ("who can invest in a spot Bitcoin ETF?") should have explicit, unambiguous answers somewhere in the text.
5. Add expert quotes: an opinion attributed with name, title, and organization is a strong authority signal for AI systems.
6. Clarify your entities: the brand, authors, and key concepts should carry consistent names and schema; AI engines match at the entity level.

The don't-waste-your-time list

Google's own guidance confirmed these five tactics provide no special advantage:

• llms.txt files (Google doesn't use them)
• Content "chunking" for AI
• AI-specific rewrites
• Chasing inauthentic brand mentions
• Special structured data marketed as "AI schema"

The zero-click reality

• Many searches now end without a click. Success includes being cited in the AI answer; track citation share in the GSC Generative AI reports.
• Leave extra value on the page that the AI citation can't carry: the full data set, charts, calculators, or the deep analysis that pulls readers through to the site.
• Build owned channels: a newsletter and a Telegram or Discord community are insurance against algorithm swings.
• Test every new content idea: "Would this be valuable even if AI Overviews didn't exist?" If the answer is no, revise or discard the idea.

Bing, Copilot & IndexNow

Bing is usually an afterthought, but Microsoft Copilot's answers are fed by the Bing index, and ChatGPT's web search is tied to Bing infrastructure as well. Bing visibility now is AI visibility.

What changed in 2026

• Bing rewrote its Webmaster Guidelines: the rules now cover classic search together with Copilot and grounding (AI answer) results, and GEO is defined by name.
• In February 2026, Bing Webmaster Tools gained an AI Performance report: it shows which pages get cited in Copilot answers, how often, and which query phrases triggered the citations. The editorial team reviews this report monthly.
• Bing's grounding system practices "abstention": if the evidence is missing, stale, or contradictory, it simply doesn't cite. A page can be indexed and still never appear in an answer.

Bing-specific content rules

• One page, one topic: Bing's guidelines say single-topic pages are more likely to be selected for grounding.
• Essential information near the top: lead with the core facts.
• Name sources in the body text: not in footnotes but inside the sentence: "according to ESMA's June 25, 2026 statement…"
• Explicit dates on figures: "according to a March 2026 study", never "according to a recent study".
• Cross-page consistency: if two of our pages answer the same question differently, Copilot may refuse to use either.
• A visible updated date: "Last updated: …" in the text, plus dateModified.

The technical side

• IndexNow: fire a notification the moment content is published, updated, or deleted. It collapses discovery lag from days to minutes. Never use it for unchanged pages; it's a freshness signal, not a ranking lever.
• Meta directives: NOARCHIVE removes content from Copilot entirely; NOCACHE limits Copilot to URL, title, and snippet. Neither belongs on content we want cited.
• Bing Webmaster Tools verification, a clean sitemap, and regular URL Inspection checks are standing tasks.

The 2026 Updates & Spam Policies

We have to know what we're writing against. Here is the verified 2026 algorithm calendar and the spam policies that concern us directly.

The 2026 calendar (verified against Google's Search Status Dashboard)

The spam policies that concern us

What to do after a core update

1. No panic analysis before the rollout finishes. Wait a week after completion, then compare a full week of data.
2. If there's a drop, break it down by page type, query intent, and content cluster.
3. There is no "core update penalty" to remove; the system re-decided which pages best satisfy each query. The fix isn't patching individual pages; it's raising the quality of the weak cluster as a whole.
4. After a spam update drop: run an honest internal audit against the spam policy list. Recovery can take months.
5. After a drop, the first move is an audit, not a rewrite: check for thin or scaled content, duplication, and E-E-A-T gaps first, and correlate GSC warnings and impression drops with recent content changes.
6. Review the external link and mention profile: are all citations natural and contextually relevant? Any suspicious paid or spammy link gets addressed.

Editorial Workflow

Nearly every published error shares one root cause: nobody read the text from start to finish before it went live. This workflow makes that structurally impossible.

The four-gate system

Standing rules

• Nobody gives final approval to their own writing. The two-person rule is absolute.
• Breaking news gets a faster path, never a shorter one: all four gates still apply; the clock shrinks, the checks don't.
• Post-publication corrections are noted under the article ("Correction, Aug 1: the figure ... was corrected to ..."). Silent edits are banned.
• A quarterly content audit: the weakest pages get updated, merged, or removed. A pile of weak pages drags the strong ones down with it.
• GSC and Bing AI Performance reports are reviewed monthly; shared traits of cited pages are circulated to the team.