My SEO system: get found, get understood, get chosen
I start with the person making a decision and the answer they need. A useful site helps them find it, understand it and choose what to do next. The three outcomes are how I test the work. The six pillars are how I do it.
- Start from the person
- Tell it as a story
- Be the source worth citing
- Never ship a wrong fact
- Measure, then double down
- Turn every lesson into a system
Below: the six pillars, the specific checks under each outcome, how they compound, how I measure it and the seven steps to build it.
The six pillars of my SEO system
Get found, get understood and get chosen are what a page has to achieve. These six pillars are how I get there. They came from sorting thousands of my own messages to the people and tools I work with, not from a whiteboard. Under each one are two of those messages, word for word.
-
Start from the person
“But every idea needs a real person with a real search behavior behind it.”
in the areas where people are tryign to understand what i even do, we need to think about simplifying it so everyone can understand, whether...
haydenschuster.com, Oct 11, 2026who's going to be on this page? on a blog, could be someone from linkedin or in the industry reading, [...] I also want to make sure my...
haydenschuster.com, Oct 11, 2026
My full-time job. At work, every request gets the same first question: who is searching, what did they type, and what do they need to do next? Every other choice on the account answers to that. How I work
How it works in practice4 parts
- Name the reader and their moment
- A named human per page or template, never "the user"; journey stage and entry channel (chatgpt, ai overviews, google, reddit, a shared link) decide what the page does; funnel stage decides page shape, so a diagnostic page never carries a buy button.
- Answer the question actually asked
- Harvest real queries, people also ask and fan-out queries, never invent faqs to fill schema; read social, reddit, youtube, press conferences and rival sources for what the audience is asking this week; model what people are planning (group costumes, pet costumes).
- The psychology of action
- Sell the why before the what; give a personal reason to act ("people are selfish"); shrink the ask (one tap beats a form).
- Put the next step where the decision resolves
- Every page ends in one deliberate next action (an affiliate link, a sign-up, the next book, a sibling page); affiliate links sit beside a resolved decision, never before it; tools go at the research stage, never interrupting the buying moment.
Read more: Why informational content stalls
Where it shows up
-
Tell it as a story
“Being right is the floor. Being genuinely helpful is the product.”
could you do a copywriting and voice audit, like you were a veteran who's studied copywriting and brand voice for 60 years in the industry,...
The Crawler's Cookbook, Oct 8, 2026id like you to follow the patterns of this duck team early in the season and start thinking critically about what this season's theme and...
The Quack Report, Sep 22, 2026
The Quack Report. The case study opens where the site did, with "I was going to Autzen Stadium before I could walk." The reason comes first; the numbers back it up after. The Quack Report story
How it works in practice5 parts
- Story order and the first screen
- Answer first, then the why, then the next step; a table with a takeaway sentence above it; current before historical, resolved before speculative.
- Mobile first, designed on its own
- Design at 390 pixels first and check 390, 768, 1440; build mobile separately, not shrunk; mobile menus are his named recurring weak spot.
- A human voice, opinion labeled as opinion
- Nothing may read as ai-generated; no "if this then that" hedging, no uniform sentence rhythm; no internal language, tool names or seo talk on reader pages.
- Tone that matches the reader's moment
- Read sentiment before writing; gameday energy, honesty after a loss, kindness about injuries; credit the opponent.
- Design with intent, never AI slop
- No generic boxes, arrows or reused ai content blocks; themed and branded to the site; color carries meaning, never decoration.
Read more: Content structure for answer-first search
-
Be the source worth citing
“Rankings are one distribution channel. The durable position is being the source that AI Overviews, ChatGPT, Perplexity, and Gemini cite when they answer the query [...]”
and on the about page, really think through what questions people will be curious about, what they search for, how we can dominate my name...
haydenschuster.com, Oct 11, 2026another north star i'd like to say is that SEO is GEO, and any sort of AEO or other abbreviation, at it's core, is still seo [...] AI is...
haydenschuster.com, Oct 11, 2026
WireRef. WireRef answers electrical code questions in plain tables. Microsoft Copilot cited it about 320K times in three months, by Bing Webmaster Tools. The WireRef story
How it works in practice5 parts
- Win the question nobody answered
- Concede branded and official searches; win the long tail, the conversational query and the unanswered question; position a site around a question with no good answer, not a topic with traffic.
- Architecture is intent made structural
- Every page has a parent, two siblings and one downstream action, zero orphans; hubs with real depth; natural bridges written as the reader's next step, not widgets.
- Readable by machines
- Anything behind a tab, toggle or accordion still in the html; interactive results baked as static text at build time; schema mirrors visible text exactly and gets the same verification as prose.
- Moats are commitments
- The moat test: would this still matter if a competitor copied the template tomorrow; moats ranked from template (weakest) to proprietary data assembly to creative reframing (strongest); complete registries.
- Bing and AI answers are first-class channels
- Bing and the ai surfaces first, google is upside; any google-only read is wrong on its face; track citations by engine (chatgpt, copilot, perplexity, gemini).
Read more: How AI search understands your content, Why AI Overviews appear in Google
-
Never ship a wrong fact
“A plausible estimate presented as fact is a failure condition. Never verify one AI-generated claim against another AI-generated claim.”
only use industry research from the last like 3 months max, this industry moves so fast
haydenschuster.com, Oct 12, 2026if you find pre-existing defects, and triple verifgy and confirm tghey are wrong on our site, and do not just lbvindly change , in the past...
WireRef, Sep 6, 2026
The Crawler's Cookbook. A later source check found an overclaim on the Book 9 page. I corrected it on August 2, 2026, said so, and tightened the check around the page. The correction
How it works in practice6 parts
- Primary source, matched claim, current date
- Three checks per fact (find the primary source, confirm it says the claim, check its date); every number traces to a file and row or a source actually read; classify a claim before verifying it (absence, universal, comparative, temporal, derived).
- Prove it, never trust it
- Ai output, subagents, plugins, codex, chatgpt and vendor support ai are leads, not facts; ask "agree or disagree" and break each finding; audit the audit.
- State the denominator, sweep the class
- Check every page, not a sample; say "checked 12 of 47" or "sampled"; name the method, never a bare "verified".
- Freshness is part of correctness
- Staleness covers every fact type (scores, names, times, tv channels, records) and the narrative, not just numbers; homepage, menu, footer and hubs included; an event cascades to every page it touches.
- Honest uncertainty
- Label every claim fact, inference or unknown; null beats wrong; a blank cell with a note beats a plausible estimate.
- The bar rises with the stakes
- On electrical, water, gauge and medical sites the bar is "would a professional making a real decision be right to trust this"; safety errors are fixed before anything else; unverifiable safety claims are held however differentiated.
Read more: Personal audit: where AI output fails
-
Measure, then double down
“Data is evidence for patterns, not a fence around ideas. Use it to understand why things work mechanically, then reason from those mechanics.”
i'm using peec.ai for ai visibility reports, ahrefs brand radar for citationa nd share of voice, GSC to pull csvs and analyze large...
haydenschuster.com, Oct 12, 2026when it comes to anything on the site from the past, i want you to pull the best elements out yes, but on a granular level, like data level,...
haydenschuster.com, Oct 11, 2026
My full-time job. At work, I finish small before going wide. Depth earns trust from people and from search systems, so priorities get set by what can be done properly first. Rules I work by
How it works in practice5 parts
- Every channel, cross-checked
- Search console, bing webmaster tools (including first-party ai citations), ga4, clarity, pagespeed and crux, screaming frog crawls, ahrefs brand radar, peec ai; first-party bing data verifies sampled third-party ai tools; name which source is authoritative for which question.
- Honest comparisons
- Complete periods only; equal windows anchored on a real event; volume floors so tiny samples can't make news.
- Follow the data to the winners
- Double down on what works instead of cutting; top pages first for both accuracy and growth; improve existing pages before adding new ones.
- Measure what actually earns
- Revenue per site with its own tracking id; explain every spike down to product and url; instrument before deciding when two explanations fit.
- A dashboard that says where to look
- Simple and visual, not an internal spreadsheet; headline metric near the top with a baseline; insight text generated from fresh data, never hand-written.
Read more: Search Console as a visibility dashboard, Why impressions rise and clicks stay flat
-
Turn every lesson into a system
“Audits find a problem once. Gates prevent recurrence.”
never trust agents and plugins naturally, everything needs to be checked with a fresh pair of eyes. [...] any handover documents or...
haydenschuster.com, Oct 11, 2026write a handoff to codex, where it isn't demands, but a thought document, to let codex explore, come up with new or better ideas, break the...
The Crawler's Cookbook, Oct 8, 2026
This site. Moving this site off WordPress, every one of 148 old addresses got a new home chosen by hand, so nothing anyone had linked to breaks. The decision log
How it works in practice6 parts
- Gates over audits
- A build gate that blocks the deploy beats an audit that writes a report; every gate fails on a known-bad fixture first (ideally the real shipped bug); zero inspected fails, never passes.
- Every mistake becomes a fix to the system
- Fix the generator or template, never the generated page; every change is a template change or it is debt; the unit of repair is the field, so enumerate every reader of it.
- Automation that runs without him
- The terminal state for a seasonal site is that it updates itself accurately; a self-healing loop beats an alert to someone on vacation; failures alert loudly and never report green on a no-op.
- Plan, break the plan, ship what survives
- Check what already exists (live site, repo, decision notes, recent work) before proposing; plan mode means no edits; steelman, then attack with evidence.
- Change safely where it earns
- Earning sites get small, reversible diffs and a stated blast radius and revert path; read-only audits on top pages first; design-only changes proven to leave search-visible content identical.
- Two AIs, checked against each other and against me
- I set the plan and the standard; one ai speeds up the build, the other reviews it with an open, creative eye, then they swap; every handoff opens with a read-critically note, so the next one questions instead of rubber-stamps; a shared log of what changed, how it was verified and what the other should check.
Read more: SEO strategy vs SEO systems, How I structured my old site for search
Where it shows up
Outcome 1 of 3
Get found
Can the person with this need find the right answer?
I look for a real unanswered need, then make the useful page reachable. Structure, links, rendering and redirects all serve that match.
-
Crawl and index health
Part of C. Be the source worth citing
Can Google and AI crawlers reach every page that matters, and store the right version of it?
Car Fact Sheet and CheckMyTap Bing went from 293 to 1.1K indexed Car Fact Sheet pages, and from 560 to 1.4K on CheckMyTap, in April 2026 alone. Bing Webmaster Tools, index counts, data through Oct 4, 2026
Example and details
CheckMyTap. Google had 1.6K CheckMyTap pages indexed in April 2026 and 218 by October. I measured the gap, wrote up the miss, and added a check that stops the site from publishing when it finds a known kind of error. The CheckMyTap story
- Sitemaps and crawl paths that point at the pages that earn
- One clear version of every page, so search doesn't pick for me
- The gap between crawled and indexed, which is where sites lose pages quietly
Where it shows up:CheckMyTapTechnical: structure, templates and checksMore writing: Technical SEO
-
Rendering and speed
Part of C. Be the source worth citing
Does a crawler see what a person sees, and does it load fast on a phone?
The Crawler's Cookbook Lighthouse gives it a 68 on mobile. Real readers on phones get the main content in 0.9 seconds, with 98% of visits rated good. I read field data before spending time on speed work, because the lab simulates a slow phone and the field is what people feel. PageSpeed Insights and Chrome UX Report, phone, 28 days, data through Oct 11, 2026
Example and details
This site. This site sends plain HTML with no embeds, loads its only outside scripts (analytics) after the page, and sets every image size, so nothing jumps while it loads. How this site is built
- Content that only appears after JavaScript runs can be invisible to a crawler
- Every script has to earn its weight
- Speed measured on a phone, not a desktop on office wifi
Where it shows up:Technical: structure, templates and checksHow this site is builtMore writing: Technical SEO
-
Site architecture
Part of C. Be the source worth citing
What pages should exist, what is each one for, and how do they connect?
Three reference sites On GageRef, Car Fact Sheet and CheckMyTap, the top ten Bing searches bring only about a tenth of the clicks. The rest come from hundreds of different long-tail searches, which is what a page template built for scale is supposed to do. Bing Webmaster Tools, search queries, four weeks, data through Oct 8, 2026
Example and details
My full-time job. At work, content is infrastructure: each page has a job, a place in the architecture and a reason to be cited, settled before anything gets drafted. How I work
- Topic hubs with real depth, not a pile of posts
- Important pages a few clicks from the front door
- Decided before design, while it's still cheap to change
Where it shows up:Enterprise SEO workMore writing: Content strategy
-
Internal link strategy
Part of C. Be the source worth citing
Does every link take someone to their next question?
Example and details
This site. Every post on this site links to the topics it's filed under, and every topic page links to every post filed under it. A check fails the build when a link is missing in either direction. The checks behind this site
- Every page gets a parent, siblings and a way onward, so nothing is a dead end
- Links written as the reader's next step, not a related-posts widget
- Links from the pages people trust pass weight to the pages that need it
-
Migrations and redirects
Part of F. Turn every lesson into a system
When addresses change, does anything people linked to break?
This site Copilot was already citing 9 of my old posts. All 9 old addresses redirect to their new pages, so the citations carry over. Bing Webmaster Tools, AI Performance, and this site's redirect file, data through Aug 26, 2026
Example and details
This site. Moving this site off WordPress, every one of 148 old addresses got a new home chosen by hand, so nothing anyone had linked to breaks. The decision log
- Every old address goes to its most relevant new home, never just the homepage
- No chains of redirects that slow every visit
- Checked again after launch, because migration damage shows up weeks later
Where it shows up:Technical: structure, templates and checksHow this site is builtMore writing: Enterprise SEO
Outcome 2 of 3
Get understood
Can they quickly tell what the answer means for their situation?
A direct answer, clear structure and sources help a person understand it. They also help search and AI tools describe it accurately.
-
User intent strategy
Part of A. Start from the person
Who is searching, and what are they trying to get done right now?
The Crawler's Cookbook I rebuilt the costume guide around the reader's real job, a costume ready by Halloween, with a quick answer first. Over the next four weeks its Google clicks per day grew 3.7 times, while the rest of the site grew 1.6 times. Google Search Console, pages, 28 days before and after Aug 17, 2026, data through Sep 14, 2026
Example and details
My full-time job. At work, every request gets the same first question: who is searching, what did they type, and what do they need to do next? Every other choice on the account answers to that. How I work
- Learning, comparing, buying and finding a brand each need a different page
- Where someone is in a decision shapes the page, so a page that diagnoses a problem doesn't push a sale
- Real questions from real searches, never invented ones
Eli Schwartz's advice for AI-era search starts with dropping the keyword as the first step and starting from what the customer is trying to do. That's the order I work in. Eli Schwartz, Oct 8, 2026
- The Quack Report Its injury report pages averaged position 4 on Bing for the weekly 'who's hurt' question, against 22.5 on Google, over the same four weeks of the season. Bing Webmaster Tools and Google Search Console, search queries, data through Oct 8, 2026
- Four of my sites 83% of the Crawler's Cookbook's AI citations, and 73% of CheckMyTap's, came from questions five words or longer: the conversational searches AI answers run on. Bing Webmaster Tools, AI Performance, grounding queries, data through Aug 26, 2026
Where it shows up:The Crawler's CookbookGageRefThe Quack ReportCheckMyTapUpOrbitPsychology: building for the person searchingMore writing: Search behavior and intent
-
Content strategy and topic clusters
Part of C. Be the source worth citing
Which questions deserve their own page, and which belong together?
The Crawler's Cookbook Before building a separate Halloween page, I checked: the costume guide already drew 99.9% of costume and Halloween search impressions. A second page would only have split them, so the Halloween angle went inside the guide. Google Search Console, query and page data, Sep 1 to Oct 2, 2026, data through Oct 2, 2026
Example and details
The Quack Report. I almost built a separate depth chart page for The Quack Report. Those searches were already landing on the roster page, so I kept one page instead of splitting it in two. The Quack Report
- One page per question, so my own pages never compete with each other
- Hubs that connect the questions people ask in order
- Pages built only when there's a real reader for them
- The Crawler's Cookbook Leave the Book 9 page out entirely and the rest of the site still went from 3.1K Google clicks in July to 7.2K in September. The cluster carries itself. Google Search Console, pages, data through Oct 7, 2026
Where it shows up:The Quack ReportContent: answers, data and voiceMore writing: Content strategy
-
Answer-first content
Part of A. Start from the person
Is the answer on the first screen, in the words the reader would use?
Car Fact Sheet Its 102 answer-first pages earned a median of 558 Copilot citations each, about three times the 172 for its 108 spec-sheet pages. Bing Webmaster Tools, AI Performance, top pages, data through Aug 26, 2026
Example and details
The Crawler's Cookbook. Fans wanted a release date for Book 9 before one existed. The page said what was confirmed, what wasn't and when I last checked, and drew 43% of the site's Google clicks. The Book 9 story
- The answer first, then the why, then the next step
- Headings that say what the section answers
- Written so a person in a hurry and a machine both get it
Chris Long's blueprint for an AI-ready page puts four to six sentences that answer the query outright at the top. My own Car Fact Sheet data points the same way. Chris Long, Sep 21, 2026
- Ten of my sites On Bing, question-shaped searches in the top three positions were clicked 31.7% of the time, against 12.5% for other searches at the same positions. Bing Webmaster Tools, search queries, four weeks, data through Oct 8, 2026
Where it shows up:The Crawler's CookbookThe Quack ReportContent: answers, data and voiceMore writing: AI and modern search
-
Structured data and entities
Part of C. Be the source worth citing
Can a machine tell exactly what this page is, and who's behind it?
Example and details
My full-time job. At work, I fix the source, never the output. Structured data goes in at the template, so one upstream change reaches every page built on it. Rules I work by
- Schema that matches what's visible on the page, never more
- Clear names for the people, places and things a page is about
- Tables a machine can quote without guessing
In Cyrus Shepard's survey of SEOs on what earns AI visibility, structured data ranked 12th of 12 factors, behind crawl access and a direct match between the question and the answer. So I use schema to describe a page honestly, never as a shortcut. Cyrus Shepard, Sep 16, 2026
Where it shows up:WireRefEnterprise SEO workTechnical: structure, templates and checksMore writing: Technical SEO
-
AI search and citations
Part of C. Be the source worth citing
Would an AI answer quote this page as its source?
The eleven sites in Bing's AI report Copilot citations grew tenfold in five months: 20.8K in the first 30 days of Bing's report, 208K in the most recent 30. Bing Webmaster Tools, AI Performance, data through Aug 26, 2026
Example and details
WireRef. WireRef answers electrical code questions in plain tables. Microsoft Copilot cited it about 320K times in three months, by Bing Webmaster Tools. The WireRef story
- Facts stated plainly, with a source and a date
- A site that's clearly about one thing
- Being named in an answer and being cited as the source are different wins, and I want both
- Title Town PDX When it answers, it tends to own the answer: an average 43% share of the citations across its 56 AI queries. Bing Webmaster Tools, AI Performance, citation share, data through Aug 26, 2026
Where it shows up:WireRefGageRefMore writing: AI and modern search
Outcome 3 of 3
Get chosen
What makes this answer worth trusting and acting on?
Distinct value and evidence make the next step feel earned. I watch what people actually do and change the page when the evidence says I should.
-
Trust and sources
Part of D. Never ship a wrong fact
Would a careful reader believe this, and could they check it?
Two sports sites Copilot keeps citing the roster pages I keep current: 88% of Title Town PDX's AI citations and 72% of The Quack Report's were for live-event questions. Bing Webmaster Tools, AI Performance, query intent, data through Aug 26, 2026
Example and details
The Crawler's Cookbook. A later source check found an overclaim on the Book 9 page. I corrected it on August 2, 2026, said so, and tightened the check around the page. The correction
- Every number with a source and a date
- Saying "we don't know yet" when we don't
- A real person behind the page, with real experience
Where it shows up:The Crawler's CookbookCheckMyTapContent: answers, data and voiceSystems: the checks behind every siteDecision logMore writing: AI and modern search
-
The path to a decision
Part of A. Start from the person
Once they're convinced, is the next step obvious?
WireRef On a phone, the safe, corrected column of the ampacity table fell off the screen, so the most visible number was the wrong one. I moved the answer column onto the first screen across 51 tables. That page's engaged-session rate went from 57.9% to 65.0% while the rest of the site held at 61%. GA4 landing pages, matching 29-day windows before and after Sep 6, 2026, data through Oct 5, 2026
Example and details
My full-time job. At work, content is structured around how the reader actually decides: the answer first, the context second, and the next question answered on the way. The content pattern
- The next step sits where the question gets resolved
- No popups, fake urgency or gates
- Designed for the phone they're holding, in the moment they searched
Marie Haynes points out that Google's updated content guidance now puts the page's main content front and center. I'd reached the same place on WireRef from the data, by putting the answer on a phone's first screen. Marie Haynes, Oct 2, 2026
Where it shows up:CheckMyTapUpOrbitPsychology: building for the person searchingMore writing: Search behavior and intent
-
Priorities by business
Part of E. Measure, then double down
What does this business need first: local, e-commerce, leads or content?
Title Town PDX Instead of fighting the official team sites for schedule searches, Title Town PDX answers the game-day questions they skip. Its game-day guides are clicked 1.50% of the time they're shown; its schedule pages, 0.46%. Google Search Console, pages, Jun 12 to Oct 7, 2026, data through Oct 7, 2026
Example and details
My full-time job. At work, I finish small before going wide. Depth earns trust from people and from search systems, so priorities get set by what can be done properly first. Rules I work by
- Priorities change with the industry and the business model
- Locations, products and services each need their own structure
- Let the official site win the official search; win the questions it never answered
- Eight of my sites Over 195 days, my reference sites got 16 Google clicks and 16K from Bing, while the fan sites ran the other way, 37.9K from Google. I know which engine each kind of site lives on, and plan for it. Google Search Console and Bing Webmaster Tools, Mar 27 to Oct 7, 2026, data through Oct 7, 2026
- Nine of my sites Leave out the one site Google favors, and Bing sent 88% of my search clicks in September 2026: 4.9K against Google's 657. I put the effort where the readers were. Bing Webmaster Tools and Google Search Console, data through Oct 8, 2026
Where it shows up:Enterprise SEO workMore writing: Industry SEO
-
Digital PR and mentions
Part of C. Be the source worth citing
Why would anyone else talk about this?
Example and details
My full-time job. On my team at work, digital PR is part of SEO: we build the roadmaps, strategy decks and measurement together, so the pages PR pitches are ready for the attention they earn. How I work
- Digital PR and SEO as one team, so links and content get planned together
- Original data or a story worth repeating
- The pages being pitched already carry the links and context the outreach promises
Where it shows up:Enterprise SEO workMore writing: AI and modern search
-
Using data to improve the page
Part of E. Measure, then double down
What did people actually do, and what should change because of it?
Hear Homer In launch week, Google sent 117 clicks and GA4 recorded 20 visits. I checked one tool against the other, found a security setting silently throwing GA4's data away, and fixed it before anyone made a decision on bad numbers. Google Search Console and GA4, Jul 19 to 24, 2026, data through Jul 24, 2026
Example and details
WireRef. When Google almost stopped showing WireRef in March 2026, the data showed Bing still sending readers, and growing. I moved the effort there instead of fighting it. The WireRef story
- Compare complete periods, never a partial week
- Label what's measured and what I think it means
- Change course when the data says so, and write down why
- The Quack Report Clarity showed people clicking cards that looked clickable and weren't. I fixed the whole class at once, about 1.6K cards, and dead clicks per session fell by about two thirds, from 0.42 to 0.15. The football season started in the same window, so I won't claim all of it. Microsoft Clarity, non-overlapping 3-day windows, data through Oct 11, 2026
- All my sites I track AI visits by the engine that sent them: 4.3K sessions so far, 83% from ChatGPT. Citations and visits are measured separately and never mixed. Google Analytics 4, AI referrals, data through Oct 10, 2026
Where it shows up:WireRefGageRefSystems: the checks behind every siteDecision logMore writing: SEO measurement
How the pieces compound: one site, start to finish
Each piece gets the next one closer. Here's that chain on Car Fact Sheet, a car reference site I built, with every number from my own data.
-
1
Get found
Bing went from 293 to 1.1K indexed pages in April 2026, so the pages existed where people search.
Bing Webmaster Tools, Oct 4, 2026
-
2
Get understood
Pages built answer-first earned about three times the AI citations of spec sheets: a median of 558 each against 172.
Bing AI Performance, through Aug 26, 2026
-
3
Get chosen
2.4K Bing clicks from April through September, with none from Google, because I put the effort where the readers were.
Bing Webmaster Tools and Search Console, through Oct 8, 2026
-
4
Measure
The page-type split showed what to build more of: answers, not spec sheets.
-
5
Carry it forward
The next site starts with what this one taught me.
How I know: the measurement layer under every piece
Every piece above gets measured, in three layers: collect the data, check one source against another, then decide.
-
1
Collect
First-party data straight from the platforms, plus third-party tools for what the platforms can't see.
- Google Search Console, exported and analyzed in bulk with AI
- Bing Webmaster Tools, including first-party AI citation data
- GA4 for revenue and key events
- Microsoft Clarity for what people do on the page
- PageSpeed and Chrome UX Report for the speed real visitors feel
- Screaming Frog to crawl a site the way a search engine does: broken links, redirect chains, missing tags, orphan pages
- Peec AI for AI visibility reports
- Ahrefs Brand Radar for citations and share of voice
-
2
Verify
No single tool gets the last word. Bing's first-party AI data checks what Peec AI and Ahrefs report from sampled prompts, and Search Console and GA4 tie visibility to the key events and revenue a business actually cares about.
In the fieldMike King makes the case that AI search tracking can't be accurate, only consistent: the value of the number comes from measuring the same way every time. That's why no single tool gets the last word here. Mike King, Sep 17, 2026
-
3
Decide
It all lands in month-over-month reports and roadmaps. Every big decision gets written down with what happened after, so the next round starts smarter.
Seven steps to build it, in order
The three outcomes say what a site needs. These steps say how I build it. Each one has to be right before the next one starts, because a mistake in the structure costs a sentence to fix early and a rebuild to fix late.
Building this site is the example I can show all of. Under each step: what it took here, and what had to be true before I moved on.
-
Step 1
Research the person, not the keyword
Before anything gets built, I find out who is actually searching, what they're trying to do, and what nobody has answered well yet.
On this siteMy own Search Console showed that 22 of the 25 Google clicks with a visible search behind them came from people searching my name, and when I searched my name on Google on October 11, 2026, my homepage didn't show up at all. So the reader came first: someone deciding whether my work holds up, who wants to know who I am and whether I'm good. I also studied 18 sites run by SEOs I respect, and read every page of my old site.
Before I move onI can name the reader, their question and what they'd do next.
-
Step 2
Map the architecture before any design
Every page gets a reader, a question and a next step, and every page has a parent, siblings and a way onward. The structure is decided while it's still cheap to change.
On this siteBefore a single page was designed, I mapped all 53 pages this way, grouped the writing into eight topics from my old site, and gave each of 148 old addresses a new home, so nothing anyone had linked to would break.
Before I move onNo page exists without a reason, and no page is a dead end.
-
Step 3
Design for the moment they arrive
Phone first, story first. The first screen answers the question the reader came with, and the design has one language, where every color means the same thing everywhere.
On this siteEvery page here is designed from a 390-pixel phone up, and is checked at four screen sizes before it can go live.
Before I move onSomeone on a phone gets the answer without scrolling.
-
Step 4
Write it in a real voice, with sources
The words have to sound like a person who did the work, and every fact has to hold up when someone, or an AI answer, checks it.
On this siteWhere I'd already said it, the site uses my own words, checked against the originals so they can't drift. Every number carries a source and a date. The hard numbers come from my own sites.
Before I move onEvery claim can be traced to where it came from.
-
Step 5
Move faster with AI, on standards I set
AI speeds up my thinking and covers ground one person couldn't: research, data and the repetitive build work. That frees my time for what matters most, and the decisions, the voice and the creative calls stay mine. Two AI models review each other's work, and neither is trusted on its own.
On this siteThis site runs 23 checks every time it changes, and nothing goes live until all of them pass. That frees my time for the parts no check can do: what to build, how it should feel, and what to say.
Before I move onNothing goes live that a check could have caught.
-
Step 6
Measure, then decide again
After something ships, the data says whether it worked. I compare complete periods, label what's measured versus what I think it means, and change course when the data says so.
On this siteEvery receipt on this site is a real screenshot from Search Console or Bing, and the decision log lines up each big change with what happened after it.
Before I move onI know what changed, and I can say whether it worked.
-
Step 7
Learn from the miss, and fix the system
When something goes wrong, I don't just fix the page. I find out why it happened and change the system, so that mistake can't happen again anywhere.
On this siteEach check this site runs is there to stop a specific mistake: a misquote, a broken link, a page that fails on a phone. Several began as mistakes I actually made. The first version of this site passed every one and still wasn't good enough, so the critique became the checklist.
Before I move onThe fix is a change to the system, not a promise to be more careful.
How a mistake makes the system better
Everyone makes mistakes. The useful question is whether the same one can happen twice. I fix the system, not just the page, so each lesson stays learned.
-
- What went wrong
- On a site with a page for every city, careful reviews still let errors through.
- What I changed
- I added a check that reads every page before it publishes and stops the whole site when it finds a known kind of error.
- What it means now
- That kind of mistake stops before anyone sees it, on every page at once.
-
- What went wrong
- A pop-up kept getting in readers' way on phones. My first fix made it disappear faster, and I assumed that solved it.
- What I changed
- Tap data from Microsoft Clarity said otherwise: its close button was still the most-tapped thing on the page, 495 of 2.6K taps. The real cause was how often it appeared, so I fixed that and added a check that re-creates the old bug on every build.
- What it means now
- I measure my own fixes instead of trusting them.
-
- What went wrong
- I retyped one of my own old lines from memory and got a word wrong.
- What I changed
- Lines quoted from me are now pulled from the originals and compared word for word every time the site changes.
- What it means now
- A misquote of me can't go live, no matter who edits the page.
-
- What went wrong
- The first version passed every check, and it still wasn't good enough.
- What I changed
- The critique became a written checklist, and every page gets rebuilt against it, one at a time, judged on a phone.
- What it means now
- Passing the checks is the floor. Whether it's good is a separate question, and a person answers it.
-
- What went wrong
- I logged a new page as a decision in July. It was finished, but it sat on a branch that was never merged, and Google never showed it once.
- What I changed
- Now a decision only counts as shipped when the page is live, checked by loading it, not when the work is done.
- What it means now
- Pushed isn't live. Live is live.
That's how it compounds. Every fix stays fixed, so each project starts where the last one finished, and the mistakes I made on the first site didn't repeat on the second, and didn't repeat on the third.
Same system, different sites
The steps don't change. What each one turns up does, because every site starts from a different reader.
| Site | Research: who's asking | Architecture | Design | What the data changed |
|---|---|---|---|---|
| The Crawler's Cookbook | Fans of a book series kept asking questions no official source answered. | Hubs by book, character and floor, with spoiler-safe sections. | The answer first, confirmed facts separated from rumor, the date it was checked. | One answer-first page drew 43% of the site's Google clicks, so more pages got built the same way. |
| WireRef | An electrician halfway through a job needs one number, fast. | Pages by state, by appliance and by project, each built around the number people need. | One column, the number first, tables a machine can quote. | Google almost stopped showing it while Bing and AI answers kept citing it, so the effort moved there. |
| CheckMyTap | A homeowner deciding what to do about their tap water. | A page for each city, at a scale only a system can keep accurate. | Problems before products: check your water, learn what it means, then see solutions. | Google indexed fewer pages over time. I added checks that stop a wrong number before it publishes, and wrote up the miss. |
At my full-time job it's the same seven steps, with a team and a bigger site: how I work there.
See it on real sites
The same system runs on every site I build, and the portfolio shows where it worked and where it didn't.