BUXS framework: brand and user-first semantic methodology for SEO & LLMO

I developed BUXS after years of watching companies burn budgets on SEO that never compounds. The pattern was always the same. Start with keyword research. Build a list. Publish content around that list. Wait. Traffic either comes and plateaus, or it doesn’t come at all. And nobody (I mean nobody) asks the question that should come first: what is this company’s strategic position, and how does organic search serve it?

BUXS Framework is a strategic SEO methodology created by Szymon Słowik, SEO consultant, founder of takaoto.pro, and international SEO speaker. It combines Brand positioning, UX optimization, and Semantic content networks into one integrated system for organic growth. Unlike keyword-first approaches, BUXS begins with competitive positioning using Michael Porter’s differentiation principles and Warren Buffett’s “brand is a moat” idea, then builds topical maps where semantic, behavioral, and sales funnel logic are designed together rather than bolted on afterwards.

TL;DR:

• BUXS integrates Brand positioning, UX journey design, and Semantic architecture into one system. Not three separate workstreams. Goals are: better retrieval, ranking and acquisition.
• Every content decision is filtered through a strategic brief that defines what you are and what you are not. Keyword volume alone doesn’t decide your map. Consistent narrative, framing, angle instead of catchy topics.
• Every topic gets decomposed twice: EAV (what the subject is made of) and query frames (how people ask about it). Crossing the two gives a coverage matrix where gaps and cannibalization become visible cell by cell, backed by live SERP evidence.
• Demand is verified on two separate axes: what the evidence says, and what we decide to do about it. Zero-volume, high-intent queries can be pursued as explicit strategic bets with a recorded reason and a review date. Not wishful thinking, a tracked promise.
• A behavioral layer models why people buy: twelve motivational types (fears, doubts, trigger events and more), verified against the voice of the market, not against keyword tools.
• The pipeline is vibe-coded with AI but uses constitutional quality gates, territory checks, and a post-generation repair layer. LLM output is never trusted raw.
• Topics that don’t pass the topical map filters become LinkedIn posts, newsletter segments, Reddit threads. Nothing gets wasted, everything compounds.

A plain-English key (because I’ll keep referring to these)

A lot of this article ties a method to a payoff, and the payoffs have jargon names. Here’s what they actually mean, in human terms, so the rest reads cleanly:

• Cost of retrieval — how hard a search engine or an AI has to work to figure out what your page is about and whether to trust it. When the system instantly “gets” who you are, it surfaces you more cheaply and more confidently. High cost means it keeps re-evaluating you from scratch and often picks someone clearer instead.
• Empty traffic — visits that look good in a report (impressions, sessions, time-on-page) but never turn into a lead or a sale. Usually the wrong audience, or the right audience with no path to do anything next.
• CRO alignment — short for conversion rate optimization. In plain terms: is the page built so a reader can actually take a step toward becoming a customer, at the moment they’re ready?
• Engagement — people clicking deeper into your site and staying, instead of bouncing straight back to Google to try another result.
• AEO / AIO presence — Answer Engine Optimization and AI Overviews. Being the source that ChatGPT, Perplexity, and Google’s AI answer box actually quote. A different game from blue links: the AI reads a handful of trusted, clearly-written sources and paraphrases them. You want to be one of those sources. You’ll also see this discipline called LLMO (LLM optimization) or GEO (generative engine optimization). Same game, different acronyms; the industry hasn’t picked a winner yet.
• Disambiguation — making it unmistakable which entity you are. Same name, different people or companies happens more often than you think. If a retrieval system can’t tell you apart from your namesakes, it won’t cite you with confidence.
• Strategic bet — a page built for demand that no tool can measure yet: zero reported search volume, but clear, high-value intent. It ships with a recorded reason and a review date, and real data later confirms it or kills it.

Keep those in mind. Almost every design decision below exists to move one or more of them.

What you’ll find in this article:

1. Why keyword-first SEO fails as a strategy
2. The three pillars: Brand, UX, Semantics
3. Two ways to decompose a topic: EAV and query frames
4. The behavioral layer: why people actually buy
5. Demand verification and the strategic bet
6. How the methodology works: the full pipeline, stage by stage
7. Quality gates that prevent the pipeline from drifting
8. The snowball effect: how the topical map feeds every channel
9. How BUXS evolved: three generations of the same idea
10. The mindset behind BUXS: pragmatism, rational doubt, evolution, compounding

The core difference from traditional SEO? Business strategy theory applied to search. Content architecture treated as an investment problem, not a publishing schedule.

Content without architecture does not compound. You are spending, not investing.

Why keyword-first SEO fails as a strategy

Most SEO projects begin backwards. An agency pulls keyword data from Ahrefs or Semrush, sorts by search volume, groups by topic, delivers a spreadsheet. Sometimes rankings improve. But each piece sits in isolation, competing with a thousand other pages chasing the same terms with the same angle.

Drucker wrote that strategy is deciding what not to do. Porter underlined the importance of trade-offs. That principle is completely absent from keyword-first SEO. Start with a keyword list; you’re letting search volume decide your strategy for you.

The result is competing on every front instead of dominating one semantic territory. The framework starts from the opposite direction. Define your strategic position first. Decide what you are and (just as importantly) what you are not. Then let that positioning guide every content decision that follows.

Here’s the practical cost of getting that wrong, in the language above: chasing volume produces a lot of empty traffic and a high cost of retrieval. You publish pages that pull in readers who were never your buyers, and because those pages don’t connect to a clear identity, the search engine keeps treating each one as a stranger. Effort in, very little business out.

Hm, and here’s why this matters for business buyers especially. A CMO justifying SEO budget to a board needs to know what the organic growth engine is building toward. “We published 15 articles targeting high-volume keywords” is activity reporting. “We established authority over three strategic topic clusters that map directly to our product positioning” is strategy. Different conversation entirely.

The three pillars: Brand, UX, Semantics

BUXS stands for Brand, UX, and Semantics. These three dimensions are integrated from the first day of any project. They work together as a system because search engines (and now AI retrieval systems) evaluate all three at once.

Brand defines your semantic territory. When search systems can unambiguously identify who you are, the cost of retrieval drops. Every new piece of content gets linked to a known name rather than evaluated from scratch. I’ve worked through this with my own disambiguation challenge. “Szymon Słowik” competes with at least four other people in the Knowledge Graph.

Building a clear signal goes beyond schema markup. Consistent co-occurrence is where it clicks. Every core page on my site mentions BUXS Framework, takaoto.pro, my conference performance or other strictly defined concept within the first 100 words.

Why bother with that first-hundred-words discipline? Think about how the system reads a page. The faster it can confirm “this is that Szymon, the SEO one, connected to these known concepts,” the less guessing it has to do, and the more confidently it can rank you or cite you. That’s the cost-of-retrieval payoff in one move.

The same discipline is what gets you into AI answers: ChatGPT and AI Overviews lean on sources they can attribute cleanly. A page that says who it is and what it covers up front is an easy, safe source to quote. A page that buries the point is a risk the AI skips.

Identity work doesn’t stop at your own domain. The framework now compares what you declare about yourself against what the off-site surfaces say: Wikidata, social profiles, directories, the knowledge panel itself. Every mismatch is a small tax on trust, because the system has to reconcile conflicting versions of you before it can cite any of them. Found inconsistencies become concrete tasks on the roadmap, not a note in an appendix.

And one rule is absolute: brand facts (positioning claims, proof points, credentials) are declared by the client or pulled from first-party documents. An AI never writes them. Your identity is the one thing in the whole system that must not be generated.

For clients, this means answering the differentiation question before writing a single word of content. What makes you different? That answer becomes the filter for every topic you cover and every topic you deliberately skip. During the Growbots project after their company pivot, the entire effort started with repositioning in search. New product, new market, but old associations were still stuck in retrieval systems. You can’t fix that with keywords alone.

In the tool this isn’t a vibe. The strategic brief carries an explicit Topical Territory: topics you own, topics that are forbidden (out of bounds even when they look adjacent and tempting), and topics that are merely adjacent, close enough that a seed could drift into them and dissolve your identity.

The forbidden list is the part most people resist and the part that pays off hardest. Every off-topic article you don’t publish is a chunk of empty traffic you never have to pay for, and a signal you never have to dilute. Stay in your lane and the search engine builds a sharp, confident picture of what you’re an authority on.

That sharper picture is what lifts rankings on the terms you actually care about. This is Porter’s trade-off logic and Buffett’s moat idea applied literally. A competitor can copy your articles. They cannot copy the accumulated, confirmed association between your name and your territory. That association is the moat.

UX means designing behavioral paths from the topical map planning stage, not making the site look nice. Let me be precise about what UX is not, in this framework: it is not UI. Not fonts and spacing, not visual semantics on the design level. Those matter, but they’re static properties of a page. UX here is dynamic and user-centred: what a specific person, in a specific state of their decision, needs to see next. A page is not an endpoint. It’s a position in someone’s process.

Most topical maps treat every page as an information delivery endpoint. User arrives, reads, leaves. I’ve audited sites with 300 informational articles and exactly zero pathways to a money page. Three hundred dead ends. The entire cluster produces impressions and time-on-site but zero pipeline impact.

That, exactly, is empty traffic at scale: a content library that performs in analytics and does nothing for revenue.

The framework fixes this by building conversion pathways into the architecture itself. Every page must be within two clicks of a money node (I call this the “Maximum Hops” rule). The plain-language reason: if a reader who just got interested has to dig through five pages to find how to actually work with you, they leave.

Two clicks keeps the path to becoming a customer short enough to survive. That’s CRO alignment built into the map, not bolted on after.

But classification alone isn’t enough. What matters is timing, placement, and context. A link to a service page dropped into the first paragraph of an educational article just annoys the reader. The same link placed after a paragraph where you’ve just described a problem the reader recognizes and given them a framework to think about it?

That’s a natural next step. Surrounding text sets up the click. CTA language matches what the reader just learned, not what you want to sell. Get this right and people move forward willingly, which is both better conversion and better engagement, because a well-timed next step keeps someone on your site instead of sending them back to the results page. And which problem to describe, which objection to answer before the link?

Those decisions come from the behavioral layer I describe below, not from a copywriter’s guess.

Internal linking in most SEO is treated as a crawl and indexing optimization. Help Google find your pages, distribute PageRank, reduce retrieval cost. Those things matter. But the primary job of an internal link is to move a human being from one stage of their decision process to the next. That’s a conversion problem, not a crawling problem. So in BUXS every internal link carries an explicit role, assigned at the map stage rather than retrofitted after publication.

There are five:

• define (points to the canonical explanation of a concept),
• deepen (adds depth on a topic the page only introduces, keeping a curious reader on-site),
• workflow_step (a sequential step in a process),
• commercial_path (routes toward a money node or CTA, the actual conversion step),
• and meaning_bridge (connects two distinct territories without blurring who you are).

Naming the job of each link is what stops linking from being decorative: if every link has to declare why it exists, you can’t accidentally build 300 dead ends, and you can’t drop a sales link where it only irritates. A link without a declared role doesn’t make it into a finalized brief.

This connects directly to ranking, too. UX is known to feed ranking systems (Navboost). In plain terms: Google watches what people do with a result. If they bounce from your page back to the SERP, that’s a negative signal.

If they click deeper into a well-designed path, the system picks up that positive feedback. So good UX isn’t just nice for humans; the engagement it produces is itself a ranking signal. UX work could well be the most underrated ranking factor in practice.

Semantics goes deeper than typical topical authority work (to learn about the core principles, follow Koray’s TA Course). One thing I picked up from Koray’s framework and built into BUXS: topical authority isn’t just about coverage. Coverage is necessary but not sufficient.

Traffic acquisition is the confirmation layer in Google. If you’ve published 40 articles in a cluster and none of them bring traffic, the search system isn’t treating you as an authority on that topic regardless of how many pages you have.

I built this tracking into the pipeline: it reads Search Console history and tells compounding clusters (positions trending up) apart from decaying ones (impressions falling faster than clicks). The point is to stop paying for empty traffic on autopilot. A decaying cluster is money leaking out, crawl budget and writing hours spent on pages the market has quietly stopped rewarding. Coverage without confirmation is just cost.

Content clusters are designed using Entity-Attribute-Value (EAV) decomposition rather than keyword grouping. The difference: keyword clustering groups terms that look similar. EAV decomposition maps the dimensions people evaluate your company on, then turns those into topic clusters.

In my engine every attribute is specifically marked. For example:

• functional (what you do),
• relational (how you relate to other entities),
• evaluative (judgments and comparisons),
• contextual (when and for whom you apply),
• oppositional (what you’re not, the myths you correct),
• or temporal (how things change over time).

Here’s why that beats keyword grouping in practice: two keywords might have similar volume and overlapping terms but serve completely different attribute relationships. Group them in one cluster and you’ve quietly created two pages fighting each other for the same spot, which is cannibalization. Google can’t decide which one to rank, so it half-ranks both, and you lose.

EAV decomposition keeps distinct needs on distinct pages and related needs reinforcing each other, so authority concentrates instead of splitting. Concentrated authority is what actually moves rankings.

BUXS uses SRO (Semantic Relevance Optimization, concept I know from Sergey Lucktinov) to reduce retrieval cost at every page. That means passage-level writing, explicit anchoring in the first 100 words, and deliberate co-occurrence patterns across the network. In human terms: write each section so it answers one thing cleanly and completely, name the things you’re talking about explicitly instead of leaning on “it” and “this,” and the page becomes easy to read for a machine.

An easy-to-read passage is a cheap-to-retrieve passage, and it’s exactly the kind of self-contained chunk that wins a featured snippet or gets lifted into an AI Overview. This is the most direct lever you have on AEO and AIO presence.

But reducing retrieval cost isn’t the end goal. The end goal is SEO ROI, and that’s where most campaigns quietly fail. They optimize for rankings and traffic without connecting those numbers to revenue.

Clusters are treated like product lines in a portfolio (BCG matrix logic): which clusters generate pipeline, which bring links, which build authority that compounds into future revenue.

If a cluster ranks well but never influences a conversion, it might still earn its place as an authority builder, but you should know that explicitly rather than assuming all traffic is equal. Treating clusters this way is how you tell real performance from empty traffic at the portfolio level, and how you decide where the next euro of content budget should go.

Two ways to decompose a topic: EAV and query frames

The EAV map I just described answers one question: what is this subject made of? Entities, attributes, values. It’s the structural skeleton, and it’s close to how static knowledge bases see the world. Wikidata and Google’s Knowledge Graph store what things are: this company, this product, these properties. Useful, and BUXS anchors to that layer deliberately.

But it’s only half of decomposition. The other half: how do people ask about it? The same attribute can be approached through different canonical query frames, the lenses a searcher can take on any topic:

• definitional (what X is),
• comparative (X vs Y),
• procedural (how to do X), evaluative (which is best),
• causal (why),
• oppositional (myth vs fact),
• preventative (risks and red flags)…

… and several others 🙂

The last one is ORM kind of frame: brand evaluation, the “is X trustworthy, is X legit” questions that people increasingly type into ChatGPT instead of Google.

Same attribute, over a dozen different intents, and each intent may or may not deserve its own page.

So in BUXS every topic gets decomposed both ways. EAV gives structure. Frames give intent. Cross them and you get a coverage matrix: each cell is one attribute seen through one frame. That cell, not the keyword, is the real unit of coverage. Two of your pages claiming the same cell? That’s cannibalization made visible before it costs you rankings.

A cell with confirmed demand and no owner is a gap, and now it has an address. And some cells you hand to someone else on purpose: comparison portals can keep “cheapest X”, that’s their game, not yours. A recorded boundary, not a gap. The matrix turns “do we cover this topic?” from a feeling into a lookup.

This double decomposition is also my answer to a debate in semantic SEO that I think is framed wrong: static versus dynamic semantics.

The static camp builds entity graphs and schema and stops there. The dynamic reality is that meaning lives in what users mean by a query, and that meaning gets confirmed or refuted by behavior over time.

Google knows what “catering” is from the static layer. Then it watches whether people who typed it clicked service pages or product pages, and reshapes the SERP to match.

BUXS sits deliberately between the two: EAV keeps you compatible with the static knowledge layer, frames plus behavioral data keep you honest with the dynamic one.

One entity can legitimately split into two clusters because the market asks about it in two different modes: in the Polish meal-delivery market, “catering dietetyczny” behaves like a service and “dieta pudełkowa” behaves more like a product. Same underlying thing, two asking modes, two clusters.

A purely static model can’t see that. A purely keyword model can’t explain it.

The behavioral layer: why people actually buy

Keyword tools show what people type. They don’t show why. And the why is where behavioral economics earns its place in an SEO methodology: buyers are not rational attribute-comparers. They act on fear and doubt, on hard constraints, on trigger events. A page that answers the query but ignores the tension behind it reads correct and converts nobody.

The newest layer of the framework, the one we’re formalizing right now in its next-generation constitution, models this explicitly. Alongside the subject side (entities, attributes, values), the knowledge model carries a motivation side with twelve canonical types:

• Goal: what the user wants to achieve. The true intent behind the query.
• Need: what is objectively required to reach that goal.
• Driver: the subjective push toward buying. A want, not a requirement.
• Problem: the obstacle or pain on the way.
• Fear: the anticipated negative outcome.
• Doubt: hesitation about a specific solution. “Will this work for a company my size?”
• Expectation: the assumed standard the solution must meet.
• Past experience: the reference point, including disappointments with a previous vendor or approach.
• Aspiration: the pull upward. Desires and ambitions beyond the immediate need.
• Curiosity: interest without a transactional goal. Still worth serving; today’s curious reader is next quarter’s buyer.
• Constraint: the hard limits: budget, time, law, skills.
• Trigger event: the life or business event that creates the need, with its recurrence pattern. September creates demand for school accident insurance every year. A funding round creates demand for outbound tooling once.

Two rules make this practical rather than academic:

1. First, the format of a piece of content derives from the motivational pair plus the user’s stage, never from one type alone. Curiosity is not automatically a listicle. A past experience with a bad vendor can produce a comparison, an avoidance guide, or a procedure, depending on what it’s paired with.
2. Second, motivations get verified against the right evidence: the voice of the market. Reviews, support logs, your own community threads, industry reports. Not keyword tools, which are structurally blind here, because a fear that stops a purchase rarely gets typed into a search box in so many words. An AI can propose motivational hypotheses, but each one enters as a hypothesis, gets reviewed by a strategist, and either earns evidence or gets rejected. Rejected hypotheses are kept with their reasons. Negative results are data.

The payoff runs through everything downstream. Doubts and fears become the objections your money pages counter, with proof attached. Trigger events tell you when demand will spike and what to have ready before it does. And expectations set the ceiling for a landing page: what it must promise, what it must never overpromise.

At this point CRO and SEO are one map, not two disciplines with a handoff between them. The psychology also connects back to Porter: differentiation could mean answering the specific doubt your competitor’s page leaves hanging. A slogan on the homepage can’t do that.

Demand verification and the strategic bet

Here’s a discipline I now consider non-negotiable, and it comes from watching both failure modes up close:

1. Failure mode one: a team publishes only what keyword tools bless, and misses the demand that tools can’t see.
2. Failure mode two: a team publishes “thought leadership” for demand that exists only in their heads, and never checks.

Both come from the same mistake: mixing up what the evidence says with what you decide to do.

So BUXS keeps them on two separate axes, and they never blend. Axis one is the evidence:

• a topic is either validated by data (volume numbers exist in Search Console or keyword tools),
• validated indirectly (no volume number, but live SERPs are clearly shaped to answer exactly this question, which is supply-side proof that Google sees the demand),
• or unable to validate (nothing observable confirms the demand).

Note the wording of that last state. It means “we cannot confirm demand.” It never means “there is no demand.” A keyword tool returning zero is a fact about the tool’s coverage, not about the market.

Axis two is the decision, and it has four values: pursue, bet, reject, or park. The interesting one is the bet.

A strategic bet is a page pursued despite “unable to validate”: zero reported search volume, but high, specific, commercially valuable intent. Persona objections translated into query hypotheses are the classic case.

If your ICP is a CMO worried about justifying SEO budget to a board, there are questions she types at 11 PM before a budget review that no volume tool has ever measured. They follow a pattern you can model, even when no single phrasing repeats often enough to register.

This is also increasingly where AEO lives: people ask AI assistants long, specific, conversational questions that never existed in a keyword database, and the source that answered that exact question is the one that gets cited. The framework treats those AI prompts as first-class demand, stored as patterns with variants, right next to classic search queries.

What makes a bet strategic rather than wishful is the paperwork around it. Every bet carries a recorded rationale (who is asking this, why we believe it, what winning looks like) and a review window. Then the data gets its turn:

If the page earns impressions and engagement by the review date, the bet is confirmed and reinforced.

If it doesn’t, that absence is treated as evidence too, and the page gets merged into a stronger one or restructured.

The default for weak-evidence topics is even more conservative: they start as a section inside an existing strong page, never as a thin standalone page, and get promoted to their own URL only after Search Console shows they’ve earned it.

Demand creation is a legitimate play. Demand assumption is not. The difference is a review date and the willingness to lose the bet.

How BUXS Framework works

The methodology covers more ground than a typical SEO engagement. It starts with strategy and ends with content briefs your team can execute from.

The pipeline runs as a sequence of stages, each feeding the next, each leaving a record of where its decisions came from. Here’s the whole thing, and for each stage, what it’s actually for.

Strategic Brief

I sit down with the founder or CMO, audit existing positioning, map the competitive environment, identify ICPs, and write down what can’t change (budget, timeline, tech stack, team capacity). The output is a strategic brief that governs every decision downstream. What this stage buys you is focus, and the absence of waste.

Working with one B2B SaaS company, defining the “not” list eliminated about 40% of their previous content calendar. Those topics weren’t bad. They just didn’t serve the positioning. Cutting them up front is the cheapest empty-traffic prevention there is, because you never write the pages that would have under-performed.

EAV Decomposition

Your company gets broken down into attributes and values. The EAV map imposes a strict semantic territory definition: keywords outside your territory get rejected, even if volume looks attractive. The purpose here is identity protection. I learned this working with a client whose previous agency had them publishing generic marketing content because those keywords had volume.

The content ranked fine, but the company’s identity was dissolving in search, and every off-topic win made the search engine less sure what they were actually about.

Rejecting tempting-but-off-territory keywords keeps the entity sharp, which is what makes the on-territory content rank more easily. The newest addition to this stage is the motivation side described above: alongside what your company is, the model records why your buyers act, so the twelve motivational types get mapped in the same pass as the attributes.

Fan-Out Enrichment

The EAV map gets expanded through keyword fan-out, GSC data, and competitor gap analysis. Search volume enters the process here (finally), but filtered through the strategic lens. Not every high-volume keyword belongs in your topical map. The job of this stage is to ground the strategy in real demand instead of guesses, so you don’t build pages for searches nobody runs.

There are several data sources that feed the fan-outs:

• DataForSEO for live SERP scrapes and keyword expansion (the SERP is the ground truth I care about; it covers AIO, PAA and other features of the result pages),
• Ahrefs for search volume estimates,
• Google Search Console for your own first-party demand on terms you already rank for,
• and Perplexity as a factual reality-check that pulls verified facts with citations and routes any contradiction of your internal claims back into review rather than accepting it silently.

That last one matters for AEO: AI answer engines reward content that’s factually tight and well-sourced, so catching a wrong claim before it ships protects the trust that gets you cited. Everything that comes out of this stage lands on the two demand axes described above: the evidence gets a status, the strategist makes a call, and the two are never confused.

Content Audit and Pruning

Before building forward, the existing site gets audited. Legacy content is analyzed against the EAV footprint using vector embeddings and cosine similarity. Pages get a disposition: keep, update, rewrite, re-arrange, merge, or prune, based on semantic fit, not just traffic. In plain terms, cosine similarity just measures how close two pages are in meaning on a 0-to-1 scale; when two of your own pages score very high, they’re competing with each other, and finding that automatically is how you spot self-competition you’d never catch by eye across hundreds of URLs.

Why prune at all?

Because removing or merging weak, overlapping pages concentrates authority on the ones that remain. I’ve watched top-3 keyword counts grow after pruning weak pages and strengthening what remained. Strategic subtraction, not addition.

Fewer, stronger pages also mean less empty traffic and a lower retrieval cost, because the search engine isn’t splitting its read of your site across near-duplicates. The audit also checks E-E-A-T signals: author credentials, trust markers, source quality, experience indicators. Every page gets a clear instruction: what to do with it and why.

Frame Analysis

This is the second decomposition axis from earlier, applied against the live market. The single most expensive mistake in topical mapping is getting page architecture wrong: splitting one topic into five thin pages that cannibalize each other, or cramming five separate needs onto one page that satisfies none of them.

Both bleed rankings and create empty traffic.

So for each entity in your EAV map, the system generates one search query per canonical query frame. It then pulls the live SERP for every frame and looks at the overlap. If one URL is omnipresent across most frames, the market is telling you this topic wants to be one comprehensive page. If each frame is served by different specialists, the topic wants to split.

The classifier turns that into a verdict: SPLIT, HYBRID, or CONSOLIDATE.

The payoffs stack up here. Matching the page to the frame the searcher (or the AI) is actually using means you answer the real question, which lifts rankings and keeps people engaged instead of bouncing.

And because AI Overviews and answer engines summarize by frame, building pages that cleanly own a frame is how you get pulled into those answers.

Deciding whether a topic is one page or five used to be the most experience-dependent judgment in the whole process. Now it’s backed by what’s actually ranking.

Topical Map

Not a content calendar. It’s a strategic architecture defining hubs, spokes, supporting pages and money nodes, with role assignments, funnel stages, bridge targets, CTA placements, query frames, and priority scoring.

The reason every node gets a funnel stage and a role is so conversion logic is designed in, not discovered later. You can see, before writing a word, that an awareness article has a path to a consideration page, which has a path to a money node. That’s CRO alignment at the blueprint stage.

Map Audit

A separate post-build pass checks the finished map against itself and against the territory. This is where the hard stops live: orphaned nodes (pages nothing links to), broken bridges, seeds that wandered into forbidden territory, money nodes nobody can reach.

The map can’t be finalized until the audit passes. Plainly, this stage exists to guarantee the map has no dead ends and no leaks before anyone invests in writing it. An orphaned page is empty traffic waiting to happen; an unreachable money node is a conversion path that doesn’t connect. Catching both on the blueprint is far cheaper than discovering them six months of content later.

Content Briefs

Many SEO strategies stop at the map and say “go write.” BUXS doesn’t.

Every priority node gets a full content brief built from live SERP analysis. The brief runs an Information Gain audit against the current top 10, but “gap” here does not mean “missing keyword.”

It means:

• What frame is nobody using?
• What conclusion does nobody draw?
• Is there a counter-argument the SERP ignores?
• A comparison table that would answer the query faster than five paragraphs of prose?
• An edge case that only a practitioner would know about?

The point of Information Gain is to make sure your page adds something instead of repeating what’s already ranked. This is increasingly the whole ballgame for both Google and AI answers: a page that just rehashes the top 10 gives the search engine no reason to rank it and the AI no reason to cite it, because they already have that information.

A page with something genuinely new is worth surfacing and worth quoting. It’s also where differentiation gets enforced at the sentence level: a unique take, a named counter-frame, an angle the market ignores. Porter’s trade-offs, executed one H2 at a time.

Six specific gap types get checked:

1. Topics no ranking page covers at all.
2. Dimensions covered too shallowly.
3. Cross-domain connections nobody makes.
4. Practitioner-only insights.
5. Original data opportunities.
6. Counter-frames where the common advice is wrong or incomplete.

Each H2 in the final brief traces back to one of these gaps. No section exists just to fill space.

The brief also builds answer blocks with passage-level entity grounding (these target featured snippets and AI Overview citations) and suggests visual semantic elements for better retrieval: diagrams, comparison tables, process flows.

These aren’t decoration. A self-contained answer block is the literal unit Google lifts for a featured snippet and an AI engine quotes in an overview, so writing them deliberately is direct AEO and AIO work.

A comparison table answers a “which is better” query faster than five paragraphs, which both helps the human and gives the machine a clean, structured thing to extract.

Content Production Kit

The complete deliverable is a Content Production Kit in agent-ready JSON/MD: strategic brief, EAV map, content audit with pruning instructions, topical map, content briefs for priority nodes, a Global Writing Guide enforcing SRO rules on all writers, zero-shot prompts for Claude-based production, and an execution roadmap.

Why structured output? Because I watched the same pattern across projects. Consultant delivers strategy. Marketing team reads it once. Six weeks later the writer is working from a Slack message that says “write something about topical authority.” The architecture gets lost in translation. Machine-readable output prevents that.

The Writing Guide is the part that protects all the upstream work at the moment it’s most fragile, when an actual writer (or an AI) turns a brief into prose. Without it, the careful entity anchoring and answer-block structure that lower retrieval cost and win citations quietly evaporate in the draft.

Hah, people use it. That alone puts it ahead of most strategy PDFs I’ve seen in this industry.

Quality gates: why the pipeline doesn’t drift

I’m building this pipeline as a working tool, vibe-coded with AI. The obvious question I get (and ask myself constantly): how do you know it isn’t making things up?

Fair question. The answer isn’t “trust the model.” It’s: don’t.

Every decision is backed by real data. GSC queries, Ahrefs metrics, live SERP scrapes, competitor inventories, client briefing materials, Perplexity-sourced facts with citations. The model never gets to invent demand.

The territory gets checked three separate times: EAV generation, map finalization, post-hoc audit!

Keywords in forbidden zones get hard-blocked. Post-generation, every structured output passes through a repair layer: demand tiers re-checked, slugs validated against a real dictionary, hierarchy tested for orphans, broken bridges removed.

This is the cleanup pass that catches the structural damage a fluent-sounding AI draft hides, the kind that quietly produces dead-end pages and empty traffic if it ships.

On top of those territory checks sit five constitutional gates. Think of them as the conditions a decision has to satisfy before it’s allowed to become real. These aren’t warnings you click past. They’re hard stops.

• Cascade enforcement. Nothing can be built on a stale foundation. If the brief or the EAV map changes, downstream work can’t quietly proceed on the old version. It stops the slow rot where the strategy says one thing and the content drifts somewhere else, which is how brands lose the clarity that keeps retrieval cheap.
• Node creation. A topic can’t become a page unless it passes four conditions at once: real demand evidence (or a declared strategic bet), distinctness from nearby pages, a declared strategic role, and competitive defensibility from the SERP recon. When no evidence exists yet, the gate doesn’t guess: it queues a real check, with its cost visible, and the candidate waits. This is the single biggest defence against both empty traffic (a page with zero demand evidence is blocked) and cannibalization (a page too similar to an existing one is blocked).
• Intent completeness. A brief can’t be approved until all four intent layers are resolved: the opening answer, the section logic and depth, the page type and schema, and the funnel role, CTA, and linking plan. A page that fully matches what the searcher wanted keeps them engaged and moves them forward, which is exactly what lifts both rankings and conversion.
• Link-role discipline. Every internal link in a plan must declare one of the five roles. An unlabeled link blocks the brief. This is what guarantees the conversion paths and on-site engagement loops actually get built, instead of being good intentions that never make it into the page.
• Additional input. When new evidence arrives (a data pull, a CSV upload, an imported conversation), it gets classified against the canonical state. Consistent evidence flows through, extending evidence creates a new attribute with provenance, and contradicting evidence is blocked from silently overwriting the model and routed to human review. It keeps the facts true over time, and factual reliability is what earns and keeps AI citations.

A seed on forbidden territory? Blocked. TOFU node with zero demand evidence? Blocked. BOFU node with no bridges leading to it? Blocked.

Hm, this might sound over-engineered. But I’ve watched enough AI-generated topical maps that looked plausible and fell apart on closer look. Clusters built on keywords nobody searches for. Nodes targeting topics outside the company’s territory because the volume was tempting. The gates and the repair layer catch all of that before it reaches the final output.

Brief sections trace back to a gap in the live SERP. Fan-out clusters cite their data source. Seeds connect to a validated demand tier. LLM output is never trusted raw.

Beyond the topical map: the snowball effect

The topical map isn’t just a publishing plan. It’s a content engine that feeds every channel you operate on.

Not every topic from the pipeline will pass all quality gates for the website. Some nodes won’t have enough demand. Some might sit in adjacent territory. Others could target a query that’s too competitive right now.

Those topics don’t get thrown away.

A topic that doesn’t pass the topical map filters might still be a strong LinkedIn post. Or a Reddit comment in a relevant thread. Or a newsletter segment for subscribers who aren’t ready to buy yet. The strategic brief already defined your positioning, your ICPs, your competitive angles. That thinking applies everywhere, not just on your blog. In other words, the work you did to keep empty traffic off the site doesn’t get wasted; it gets redirected to a channel where it fits.

Zero-volume, high-intent queries follow the strategic bet path described earlier: published with a recorded rationale, validated in GSC by their review date, and either reinforced or recycled as content fuel for other channels. Either way, you learn something a keyword tool could never tell you.

There’s also a topic-splitting safety valve. When the evidence is ambiguous, a topic can ship as a single umbrella article with its frame variations absorbed as sections, and a watch list tracks those sections in Search Console after publication. If one starts earning its own impressions, it gets promoted to a standalone page. You don’t have to gamble on architecture up front, and you don’t lose the upside later. You let real engagement data make the call.

Post-production, the loop keeps running. GSC and GA4 (sometimes MS Clarity and GTM) will feed back into the system. A page getting impressions but no clicks might need a reworked title. A page getting clicks but no conversions could be missing a behavioral bridge.

Optimize if there’s potential. Cut if it only wastes crawl budget and authority distribution!

That last move matters more than it sounds: pruning dead weight frees up the attention Google spends on your site for the pages that earn it, which lowers retrieval cost across the whole domain.

The data tightens the map for the next cycle, and each cycle costs less because the strategic foundation is already in place. This closing loop is the dynamic half of semantics in practice: what users meant gets confirmed or corrected by what they did, and the map absorbs the answer.

The more you know about your market and your audience, the cheaper each new piece becomes to produce. Not because you’re cutting corners, but because the strategic work is already done. Consistency, dedication, and doing the homework up front so everything downstream compounds faster.

BUXS vs traditional SEO

Dimension	Traditional SEO	BUXS Framework	What it buys you
Starting point	Keyword research and search volume	Brand positioning and competitive differentiation	A clear identity, which lowers cost of retrieval
Content planning	Keyword clusters sorted by volume	EAV decomposition filtered through strategic brief	Less cannibalization, more concentrated authority
Topic granularity	Split by intuition or word count	Split or consolidated from live cross-frame SERP overlap	Right page architecture, better rankings and AEO
Demand decisions	Search volume is the go/no-go	Evidence and decision on separate axes; zero-volume bets are explicit and reviewed	High-intent demand tools can’t see becomes an asset, not a blind spot
Buyer psychology	Absent, or ad hoc “pain points”	Twelve motivational types verified against market voice	Pages that answer objections, not just queries
UX integration	Separate CRO pass after content exists	Sales funnel logic built into topical map from day one	Conversion paths by design, less empty traffic
Scope control	Anything with search volume is fair game	Topical territory rejects off-topic terms	A sharper entity, fewer wasted pages
Quality control	Scrape top results, cover same ground	Information Gain audit before every brief	Something new to rank and to cite
Trust in AI output	“The tool generated it, ship it”	Five constitutional gates + repair layer	No dead-end or off-territory pages reach production
Output format	PDF audit report or spreadsheet	Agent-ready JSON/MD with Writing Guide and zero-shot prompts	The strategy survives contact with the writer
Decision logic	“It depends”	Named assumptions and named trade-offs	A budget conversation you can actually have

Traditional SEO produces results. But those results tend to plateau because nothing guides what compounds and what doesn’t. In BUXS, the strategic brief already spells out goals and limits. So “it depends on your goals” turns into: “Target this keyword because it builds territory in the highest-return cluster; trade-off is lower volume; assumption is that authority here compounds within 6 months.” Ever been in a meeting where the agency can’t give you that?

Who BUXS Framework is for

BUXS is not for every company. I’d rather be honest about that than pretend it’s universal. It works best for B2B companies that want to launch a product or they are after a pivot or rebrand, for SaaS companies competing against higher-authority players who need to find and own specific semantic territory. Professional services firms whose expertise lives in people’s heads but not in search systems are another fit. Same goes for companies entering new markets where “translate and publish” doesn’t cut it.

It requires a solid technical SEO foundation. Companies without clear differentiation may need positioning work before starting. And not every business justifies the full sprint. If you’re running a local service business with three competitors, a good technical audit might be everything you need.

I’m currently developing the appropriate logic for e-commerce, where semantics are managed in a slightly different way and category page can serve as a pillar surrounded by variously framed guides and rankings.

The mindset behind BUXS

Before the pipeline and before the gates, there is a mental model. Everything above is downstream of four commitments, and if you take nothing else from this article, these travel well beyond SEO:

• Pragmatism: the “whatever works” approach. Methods are judged by outcomes, not by loyalty to a school and not by what authorities say, Google included. SEO is full of churches: white hat versus black hat, content purists versus link buyers, “Google says” versus “Google does.” I don’t join any of them. Guidelines are one corporation’s preferences, not moral law; know the risks, know the enforcement patterns, and make your own call. If the data says a sacred practice isn’t paying, the practice goes.
• Rational doubt instead of dogmatism: every rule in the framework is a hypothesis with evidence attached, my own rules included. “It depends” is banned unless it immediately names the variables it depends on. Being wrong is acceptable; the system keeps rejected hypotheses on record precisely so we can see when I was. Staying wrong is the only real failure.
• Evolution: the framework changes when evidence changes, through recorded amendments, not through rebrands. Three generations in, the core idea has survived every revision while almost every mechanism around it has been rebuilt. That’s not instability. That’s the method applied to itself.
• Patience and compounding instead of just relying on tricks and constant pivots: tricks decay, assets compound. Organic growth pays like an index fund, not like a slot machine, and the biggest cost in this channel is cashing out early to chase the next shiny thing. Every pivot resets the clock on authority you had already partly paid for.

The market rewards this mindset structurally, not sentimentally. Search engines and AI systems are pattern-accumulating machines: they build confidence in entities that stay consistent over time. A trick works until the system learns it. A compounding asset works because the system learned it.

Inspirations and thanks

BUXS is a patchwork, and I say that with pride. Nobody builds a methodology alone. Three people shaped its foundations more than anyone else. Koray Tugberk Gübür, whose topical authority framework taught me to treat coverage and semantic structure as an engineering discipline, not a content calendar. Sergey Lucktinov, whose Semantic Relevance Optimization sits behind the passage-level writing rules in every BUXS brief. And Jason Barnard, whose work on brand SERPs and Knowledge Panels convinced me that disambiguation and entity clarity are brand strategy, not technical trivia.

A separate thank-you goes to the colleagues whose peer reviews, discussions, and feedback shaped my current understanding of SEO, semantics, UX and brand strategy: Michał Suski, Maciej Tesławski, Piotr Socha, Maciej Markowski, Erfan Azimi, Robert Niechciał, Damian Sałkowski, Kevin Maguire, Damian Kozłowski, Maciej Chmurkowski, among many more. Our community is something outstanding.

BUXS Framework FAQ

Ready to see how BUXS applies to your specific situation? The BUXS strategy sprint walks through the full diagnostic process, from strategic brief through topical map delivery.

Powodzenia!

Article created by Szymon Słowik (narrative, know-how, strategy, cases, sources and quote selection) and edited with LLM, but could you actually see any difference? 🙂