Most “build topical authority” advice describes the shape of good content. This is the math underneath it — why distance from your core topic costs you more than thin writing ever will.
In January 2025 I killed forty-one articles on a finance affiliate site I’d been building for fourteen months. Not “merged” or “refreshed” — deleted, redirected, gone. The site had been growing the entire time I was writing them. It just wasn’t growing in the way that mattered, and it took me an embarrassingly long time to understand why.
The forty-one articles weren’t bad. Most of them were competently researched, reasonably well written, and targeted at real search queries with real volume. The problem was that they covered forty-one different things. Budgeting apps one week, business credit cards the next, a detour into crypto tax software, then back to budgeting apps from a completely different angle. Keyword tools kept telling me there was demand for all of it, so I kept writing toward the demand. Eighteen months later the site had 70-some pages and a Domain Rating that had crawled up to 19. A competitor launched four months after I did, wrote sixty pages about exactly one thing — comparing cashback business credit cards — and passed my entire site’s organic traffic within seven months on a domain that started from zero.
That site is the reason this article exists. What follows isn’t another description of pillar-and-cluster content (you’ve read that one, probably several times, and our own deep dive on affiliate SEO content systems already covers the five operational systems that sit underneath good execution). This is the architecture decision that has to happen before any of that — the question of where your next page should go, expressed as something closer to a model than a vibe.
- The 2026 environment: why “spread it across more topics” is now actively dangerous
- The Topical Gravity Model — an original mental model for resource allocation
- The Authority Compounding Curve — a 24-month simulation
- The Entity Coverage Ratio — measuring depth instead of guessing it
- The Three-Layer Authority Stack
- The Capital Allocation Matrix — deepen vs. expand, with real unit economics
- Two failure modes: the Scatter Trap and the Single-Topic Trap
- The unpopular take: “go deep, never broad” is wrong after a certain point
- A 90-day build sequence
- Where this system breaks down
- FAQ
Nobody outside Google has Google’s actual ranking algorithm, and anyone who tells you they’ve reverse-engineered it is selling something. Every model in this article — the gravity formula, the compounding curve, the ROI matrix — is a planning heuristic I built to make resource-allocation decisions concrete enough to act on, not a claim about how Google’s systems literally work internally. Where a number comes from a documented external source, it’s cited. Where it’s my own model, I say so and show the assumptions, so you can disagree with them.
The 2026 environment: why scattering is now actively dangerous
Affiliate sites went into 2026 already bruised. The March 27–April 8, 2026 core update — arriving two days after a dedicated spam update on March 24–25 — produced what one tracking firm’s analysis of roughly 600,000 monitored pages described as the highest negative-impact rate of any site category: affiliate properties saw ranking drops far more often than e-commerce, SaaS, or local-business sites in the same dataset.1 AI-generated content farms with no original testing reportedly lost 60–80% of their traffic in the same window.1 The May 2026 core update that followed reinforced the same message rather than introducing a new one: sites with demonstrable topical depth gained ground, and affiliate pages offering little beyond a product link and a generic summary kept losing it.2
None of this is Google quietly hating affiliates. It’s the predictable result of a multi-year shift that’s now fully load-bearing in the ranking systems: depth in one subject reads as a trust signal, and breadth without depth reads as the opposite. Google’s own developer documentation on core updates puts it in plainer terms than most SEO commentary does — these are broad recalibrations of an entire scoring system, not penalties aimed at individual sites, and the self-assessment they recommend asks directly whether your content is comprehensive on a topic or just technically present on it.3
The AI Overview piece matters specifically for affiliate content because it changes who benefits from being “the most complete source” on a topic. Content cited inside AI-generated answers reportedly earns meaningfully more clicks than comparable uncited content on the same query — one analysis put the gap at roughly 35%1 — and summarization systems are, structurally, biased toward sources that cover a subject comprehensively rather than partially, because comprehensive sources are cheaper to extract a confident answer from. That’s a separate argument for depth that has nothing to do with classic ranking and everything to do with surviving the next interface change. We’ve written more specifically about that shift in our AI Overview recovery research if you want the GEO-specific angle.
So: the environment rewards depth and punishes scatter more clearly than it has at any point in the last three years. The advice to “build topical authority” is, at this point, close to universal. What’s much rarer is a way to decide, concretely, whether the page you’re about to commission tomorrow is building that authority or quietly spending it.
The Topical Gravity Model
Here’s the mental model I wish someone had handed me before I wrote those forty-one finance articles. Picture your site’s core topic as a mass sitting in space. Every page you publish is a smaller object you’re choosing to place somewhere around it. Pages placed close to the core — sharing its entities, its vocabulary, its reader intent — get pulled into a stable orbit by everything you’ve already built. Pages placed far away just drift. They don’t get demoted, exactly. They get ignored, which in practice looks the same and costs more, because you still paid to produce them.
The reason distance matters more than almost anything else is that its effect isn’t linear. It behaves like gravity in the literal physics sense — the pull falls off with the square of the distance, not the distance itself. Move a piece of content twice as far from your established core, and you don’t lose half its ranking support. You lose three-quarters of it.
L — Link Density: contextually relevant internal links connecting the new page back to that core, normalized 0–10
D — Semantic Distance: how far the new page’s topic sits from the core, on a 1–10 scale, based on shared entities and reader intent
Pull — a relative score, not a ranking position. Useful for comparing two content decisions against each other, not for predicting an exact outcome.
Run the numbers on three real decisions from a hypothetical pet-tech site whose core pillar is “best robot vacuums for pet hair” (M = 80, a genuinely thorough 4,200-word pillar with real testing notes and 12 internal links already pointing at it):
| Candidate page | Distance (D) | Link density (L) | Pull score | What it means |
|---|---|---|---|---|
| “HEPA filters in robot vacuums, explained” | 2 | 6 | 120.0 | Tight orbit. Shares entities directly with the core (filtration, allergens, pet dander). |
| “Best upright vacuum for pet hair” (non-robot) | 5 | 2 | 6.4 | Same buyer, different product category. Pull collapses almost twentyfold versus the first option. |
| “Smart home automation starter guide” | 9 | 1 | 0.99 | Plausible traffic on paper. Structurally invisible to the authority you’ve already built. |
The inverse-square term is the whole point of building this as a model instead of a slogan. It explains something that pure intuition gets wrong: the jump from “closely related” to “loosely related” is far more expensive than the jump from “loosely related” to “barely related,” because most of the damage happens early. Most operators sense that distant topics rank worse. Almost nobody budgets for how much worse, which is why “I’ll just throw in a few tangential posts for extra traffic” keeps feeling like free money until the analytics six months later say otherwise.
This is also the part of the model that owes the most to existing work. The terms “topical authority” and “topical map” were coined and developed by Koray Tugberk GÜBÜR, whose framework around source context and what he calls the “contextual bridge” between topics is the most rigorous public treatment of why semantic relatedness matters for ranking.4 The Gravity Model isn’t a replacement for that work — it’s my attempt to compress the qualitative version of it into a number two people can argue about over a content calendar, instead of a feeling one person has about whether a topic “fits.”
The Authority Compounding Curve
Gravity explains the resource-allocation decision for a single page. It doesn’t, by itself, show why scattering feels productive for the first several months and then quietly stops working. For that you need to look at compounding over time, not just distance at a point in time.
Below is a simulation, not a tracked dataset — I want to be explicit about that distinction since it’s the single most important caveat in this article. I built two hypothetical 24-month builds with identical starting conditions (200 sessions/month, ~70 total pages by month 24, same publishing cadence) and let them diverge only in how concentrated the pages were:
| Assumption | Compressed build | Scattered build |
|---|---|---|
| Topic spread | 1 core topic, ~70 pages, average D ≈ 2.3 | 18 loosely related topics, ~4 pages each, average D ≈ 6.1 |
| Internal link density | High — every page links to 2–3 siblings and the pillar | Low — most pages link mainly within their own mini-cluster |
| Growth pattern modeled | S-curve: ~14%/month months 1–12, decelerating to ~6%/month months 13–24, reflecting compounding internal link equity | Roughly linear: flat addition of new-page traffic with little lift to older pages |
The shape of that gap — small early, large late — is the part that gets operators in trouble, because for the first four or five months a scattered build and a compressed build can look almost identical in a dashboard. The divergence is structural, and structural problems take longer to show up than content-quality problems do. This is the actual mechanism behind the “topical depth beats breadth” pattern that’s now showing up consistently in 2026 core-update analyses; it isn’t that Google has a rule against broad sites, it’s that broad sites rarely accumulate enough internal Pull on any single subject to clear the bar for genuine depth on that subject.
The Entity Coverage Ratio
“Build topical depth” is advice everyone agrees with and almost nobody can measure. The Entity Coverage Ratio is my attempt to turn it into a number you can actually track over a content calendar, by borrowing directly from the entity/attribute framing in Koray Tugberk GÜBÜR’s topical-map methodology4 and giving it a simple denominator.
Worked example: an audit of “standing desks” across the top 25 ranking pages, PAA boxes, and a couple of buyer forums turned up 86 distinct entities and attributes worth covering — motor types, weight capacities, certain certification standards, specific assembly complaints, warranty terms, desk-mat compatibility, cable-management features, and so on. A typical thin affiliate roundup in this niche covers maybe a dozen of those. A site with one deep pillar and eight or nine well-evidenced cluster pages might cover 30–35.
| ECR band | Coverage | What it usually looks like |
|---|---|---|
| Thin | <25% | A roundup post and not much else. Reads as a single opinion, not a knowledge base. |
| Developing | 25–50% | A pillar plus a handful of clusters. Enough for some long-tail rankings; not enough to win competitive terms. |
| Competitive | 50–75% | Most buyer questions answered somewhere on the site, with real evidence layers, not just mentions. |
| Dominant | 75%+ | Rare, and usually the result of 12+ months of deliberate building rather than a content sprint. |
The honest limitation here: building the denominator is manual work. There’s no API that hands you “the entity universe of standing desks.” You’re doing close reading of the top of the SERP, which takes a few hours per topic. I think that’s a feature, not a bug — the act of building the list is itself the research that should inform your content brief, so the audit and the content plan happen in the same sitting rather than as separate steps.
The Three-Layer Authority Stack
Most affiliate sites that “fail at E-E-A-T” haven’t actually failed at all three E-E-A-T pillars equally. In practice I see the same pattern repeatedly: sites get one layer of authority mostly right, partially fake a second layer, and almost never build the third at all.
- Lexical Authority — using the right vocabulary, entities, and attributes for the topic. This is what most “SEO content” optimization actually targets, and it’s the easiest layer to fake with AI assistance, which is exactly why it’s worth the least on its own in 2026.
- Structural Authority — internal links and site architecture that demonstrate the topical relationships actually exist, rather than just being claimed in body copy. A page that mentions twelve related concepts but links to none of your other content isn’t demonstrating depth, it’s listing it.
- Behavioral Authority — the signal that comes from readers actually finding what they needed: dwell time, return visits, branded search growth, lower pogo-sticking back to the SERP. This is the layer Google’s own core update guidance circles back to constantly, and it’s also the layer covered in detail in our analysis of Google’s evaluation stack. It’s the hardest layer to build and the only one you can’t shortcut with better writing alone — it requires the content to actually be the best answer, not just look like one.
The Capital Allocation Matrix
Eventually this stops being an abstract modeling exercise and becomes a budget conversation: deepen the topic you already have, or expand into something adjacent. Here’s the unit-economics version of that decision, using plausible but illustrative numbers for a mid-tier affiliate page.
Payback period = Production cost ÷ Expected monthly value
| Deepen (D=2, near core) | Expand (D=9, new topic) | |
|---|---|---|
| Production cost (writing, editing, hands-on testing) | $180 | $180 |
| P(rank) within 12 months | ~85% | ~20% |
| Traffic once ranked, monthly | 240 visits | 150 visits |
| Conversion rate / avg. commission | 3% / $34 | 3% / $34 |
| Expected monthly value | $208 | $31 |
| Payback period | ≈ 0.9 months | ≈ 5.9 months |
Two failure modes: the Scatter Trap and the Single-Topic Trap
I built the Gravity Model after living inside the first failure mode for eighteen months. It’s worth being precise about what actually goes wrong, because “don’t spread too thin” is true but too vague to act on.
The Scatter Trap
Picture publishing 100 pages across 20 loosely related topics, five pages per topic. None of those clusters individually reaches the page count needed for meaningful internal Pull, so the probability of any single page reaching a top-10 position stays low across the board — in my own tracked experience on the finance site mentioned at the top of this article, roughly 8% of pages reached page one within a year, and most of those were low-volume long-tail terms that didn’t move revenue. The total investment looks like 100 shots at the lottery. It behaves more like 20 separate underfunded attempts, each too small to win.
The Single-Topic Trap
Less discussed, and the one most “go deep” advice ignores entirely: a topic has a ceiling. Once you’ve covered the available entities reasonably well — once your ECR is genuinely north of 60–70% — additional pages on the exact same subtopic stop adding meaningful Pull and start competing with your own existing pages for the same SERP real estate. I model this as a logistic saturation curve: achievable traffic from a single topic approaches a ceiling as page count increases, and the marginal Pull per new page falls off well before most operators notice.
The unpopular take
“Go deep, not broad” is correct for roughly the first 60–70% of a niche’s revenue ceiling and starts becoming wrong after that. I’ve watched operators — including, for a while, myself on a second site — refuse to ever expand into an adjacent topic out of fear of “diluting authority,” long after their Entity Coverage Ratio on the original topic had crossed into the Competitive band and started generating diminishing returns per new page. That’s not discipline. That’s leaving a real revenue ceiling on the table because a heuristic that was right at month six got treated as a permanent rule at month eighteen. The skill isn’t picking depth over breadth once. It’s noticing when you’ve crossed the saturation threshold in Figure 7 and treating expansion into a closely adjacent topic — not a randomly chosen one — as the next deepening move, not a betrayal of the strategy. Our broad authority sites framework covers what that expansion phase looks like once you’re actually past the threshold, rather than guessing you might be.
A 90-day build sequence
Putting all six pieces together into something you can actually schedule:
- Weeks 1–2 — Build the entity universe. Run the Entity Coverage Ratio audit on your core topic before writing a single new page. This becomes your content brief source of truth.
- Weeks 3–4 — Score your existing archive. Run every published page through the Gravity Model. You will find pages with surprisingly high Pull that are under-linked, and pages with low Pull that have been eating budget for months.
- Weeks 5–10 — Deepen first. Use the Capital Allocation Matrix to sequence new pages by expected ROI, prioritizing anything inside D≤3 of your core before touching anything further out.
- Weeks 11–12 — Re-link, don’t just re-publish. Every new page should add 2–3 internal links to existing siblings and the pillar. This is the step most teams skip under deadline pressure, and it’s the one with the highest leverage in the entire formula.
- Ongoing — Watch for the saturation threshold. Once your ECR clears roughly 60%, start scouting the nearest adjacent topic rather than continuing to add marginal pages to a saturated one.
Where this system breaks down
- The Gravity Model assumes you can reasonably estimate Semantic Distance. In genuinely novel or fast-moving niches — anything tied to a brand-new product category, for instance — there may not be enough SERP precedent to estimate distance with confidence, and the model will feel arbitrary until more data exists.
- The Capital Allocation Matrix’s P(rank) inputs are estimates, not guarantees, and they’re far less reliable in YMYL categories (finance, health, legal) where Google’s quality bar for new entrants is structurally higher regardless of topical distance. Our research on AI-resistant niche selection goes deeper on which categories behave differently here.
- None of this replaces actual product testing, original photography, or the evidence-layer work that earns trust at the sentence level. A perfectly architected site full of thin reviews is still a site full of thin reviews.
- Saturation thresholds are niche-specific and I don’t have a clean formula for predicting where one sits in advance — you mostly find it by tracking marginal traffic per new page and watching the curve flatten in your own analytics.
Frequently asked questions
What is a topical authority system for an affiliate website?
It’s a deliberate content architecture — not just a list of articles — that decides which subtopics to cover, in what order, with what internal linking pattern, so the site builds measurable depth in one subject area before expanding into adjacent ones.
How is this different from a content cluster or pillar-cluster model?
Pillar-cluster models describe the shape of the content. Topical authority describes the outcome you’re trying to produce with that shape. A site can have a textbook cluster structure and still lack topical authority if the entities it covers don’t match what the topic actually requires.
How many articles does it take to build topical authority?
There’s no fixed number — niches vary enormously in entity count. Measuring Entity Coverage Ratio against top-ranking competitors is a more reliable gauge than counting pages.
Does this still matter with AI Overviews everywhere?
More than before, not less. Depth is also what gets a source cited inside AI-generated answers, because comprehensive sources are easier for summarization systems to extract a confident answer from.
Should a brand-new affiliate site start broad or narrow?
Narrow, almost without exception, until you can demonstrate (via ECR) that you’ve reasonably saturated the topic — then expand to the nearest adjacent one, not a random one.
More from ContentEvaluator
Sources
- JetDigitalPro tracking data (~600,000 pages) and Seer Interactive AI Overview CTR research, both cited via LaunchCodex’s March 2026 core update analysis: launchcodex.com
- Google May 2026 core update impact analysis on affiliate content: asclique.com
- Google Search Central, official documentation on core updates: developers.google.com
- Koray Tugberk GÜBÜR, Topical Authority and Topical Maps methodology, Holistic SEO: holisticseo.digital
- Seer Interactive, content pruning case study (+23% YoY organic traffic recovery): seerinteractive.com
- Search Engine Land, guide to Google core updates: searchengineland.com
- Post Affiliate Pro, affiliate marketing industry size 2025–2026: postaffiliatepro.com — note that market-size estimates for the affiliate industry vary widely by methodology (figures from $14B to $37B+ appear across current reports), which is a genuine measurement problem in this space rather than a number we’re picking to suit the narrative.