Pet Care AEO: Vet Clinics, Pet Food Brands, and the New Pet-Owner AI Funnel

In November 2025, Backlinko's Brian Dean reported that the company's link-building pillar page — a 7,300-word hub last meaningfully restructured in 2019 — was being cited by ChatGPT in roughly 31% of all link-building related queries the company tracked, a citation rate higher than its conventional Google ranking position would have predicted. Around the same time, Ahrefs disclosed in a public webinar that its keyword research pillar and the cluster of nineteen supporting articles attached to it accounted for an estimated 48% of the company's measurable AI assistant traffic, despite representing less than 3% of total published URLs on the domain. HubSpot, the company that invented the modern pillar-cluster framework in 2017, published an internal analysis in February 2026 showing that its top twelve pillar pages — most of them more than five years old — were now driving the majority of its assistant-attributed pipeline.

This was not the trajectory any of these companies expected. The pillar-cluster model went out of fashion between 2020 and 2023, dismissed as a holdover from the keyword-density era and superseded by entity-based ranking signals. Most major content programs quietly stopped building new pillars. Several rebranded their content strategies as topic clusters without the pillar, or replaced pillars with knowledge bases that linked to atomized articles. The 2017 vision of comprehensive, hierarchically organized topical coverage was treated, by 2022, as marketing folklore.

The LLMs have brought it back, hard. Across the brands we have tracked since mid-2025, the single most reliable structural predictor of AI citation share inside a topic is whether the brand owns a deep pillar-cluster hub on that topic. Companies that maintained their pillars through the unfashionable years are compounding citation share faster than competitors that abandoned them. Companies that are building new pillars in 2026 are seeing measurable lift in AI assistant traffic within six to nine months, a timeline that the SEO playbooks of the 2010s would have considered impossibly fast. The pillar page is not just back — for AEO purposes, it is structurally the most efficient content asset available, and most operators are still under-investing in it.

This piece walks through why the model is back, what a 2026 pillar page actually looks like, how to size the supporting cluster, and how three reference cases — HubSpot, Backlinko, and Ahrefs — are running the playbook at scale.

Why the Pillar-Cluster Model Fell Out of Fashion

To understand why pillars are back, it helps to remember why they went away. HubSpot introduced the pillar-cluster framework in 2017 as a response to Google's Hummingbird update and the broader shift from keyword-based ranking to topic-based ranking. The pitch was elegant: publish one comprehensive hub page on a broad topic, surround it with eight to twenty in-depth supporting articles on subtopics, link the cluster aggressively, and Google would recognize the cluster as topical authority and reward the entire constellation.

For about three years, this worked extraordinarily well. HubSpot's own pillar pages drove an estimated three-fold increase in organic traffic to its blog between 2017 and 2020. Backlinko, Ahrefs, Moz, and dozens of mid-market SaaS marketing teams shipped pillar-cluster programs that produced documented, durable ranking lifts. The framework became standard operating procedure for serious content marketing teams.

Then several things happened in parallel that made the model feel obsolete.

Google shifted to entity-based ranking. The introduction of BERT, MUM, and the entity graph meant Google could understand semantic relationships across pages without requiring the explicit pillar-cluster topology. Ranking signals shifted toward author entities, brand entities, and topical entities rather than the structural relationships between pages on a domain.

Helpful Content Updates penalized formulaic clusters. Between 2022 and 2024, Google's Helpful Content Updates explicitly downgraded sites that produced large volumes of formulaic content optimized for ranking. Many pillar-cluster programs that had become factory operations — twenty supporting articles produced by contractors to a template — saw their traffic collapse. Operators correctly read the signal that volume-driven cluster production was a losing strategy.

The shift to short-form and AI summaries reduced long-page payoff. Google's featured snippets, then People Also Ask boxes, then AI Overviews progressively reduced the click-through rate on long-form content. The economic case for spending forty hours producing an 8,000-word pillar page weakened as a smaller and smaller percentage of the value flowed back to the publisher.

Knowledge bases and documentation replaced pillars for product-led companies. SaaS companies in particular moved their topical authority investment from marketing-blog pillars to product documentation, which served customers directly and produced more durable distribution outcomes.

By 2024, building a new pillar-cluster hub felt like an anachronism — the marketing equivalent of optimizing for Google's PageRank algorithm in 2012. The framework was correct in the abstract but obsolete in the specific. Most content programs reallocated their pillar budget to thought leadership, video, or short-form social content.

That reallocation now looks like a mistake.

What Changed: LLM Retrieval Wants Exactly What Pillars Produce

The change is mechanical rather than philosophical. LLM-powered search assistants — ChatGPT, Claude, Perplexity, Gemini, and the rest — do not read web pages the way Google's classical algorithm did. They retrieve chunks. A chunk is typically a passage of 200 to 800 tokens, extracted from a longer document, embedded as a vector, and stored in a retrieval index. When a user asks a question, the assistant computes the embedding of the query, retrieves the top-k most relevant chunks across its index, and assembles the answer from those chunks while citing the source documents.

This architecture has structural consequences that the SEO playbook of 2019 did not contemplate.

Density of relevant chunks matters more than page length per se. A 7,000-word pillar page that covers a topic comprehensively might contain forty to sixty distinct chunks, of which fifteen or twenty are likely to be retrieved across the broad set of queries the topic generates. A 1,500-word blog post on the same topic produces six to ten chunks, of which perhaps two or three are retrievable. The pillar produces roughly five to ten times the retrievable surface area per published artifact.

Interlinking changes how chunks are interpreted. Retrieval systems do not just rank chunks in isolation — increasingly, they use the surrounding document structure and the linked context to disambiguate and score relevance. A chunk that lives inside a pillar page that is bidirectionally linked to twenty supporting articles on the same topic carries more contextual signal than the same chunk on an orphan page. The cluster is the context.

Repetition and reinforcement across cluster pieces strengthens entity signals. When ten cluster articles consistently use the same terminology, the same examples, and the same definitions, the retrieval system reads that consistency as topical authority on the entity. When a single orphan blog post uses idiosyncratic vocabulary, it gets discounted as a less authoritative source.

Stable URL structure with deliberate hierarchy gets cited more. Pillar pages typically live at clean, top-level URLs (/topic) with supporting articles in a logical hierarchy (/topic/subtopic-1). This URL structure is itself a signal of topical organization that retrieval systems weight positively in source quality ranking.

The cumulative effect is that the pillar-cluster architecture is now disproportionately efficient at producing AI citations relative to its production cost. The 2017 SEO theory was approximately correct about the destination but wrong about the timing and the mechanism. The retrieval-first internet has built what the keyword-density internet could not quite reward.

For a deeper view on how chunk structure interacts with content design, see heading structure and chunking for LLM retrieval optimization, which covers the technical mechanics of how H2 and H3 architecture maps to retrievable passages.

Pillar Page Anatomy: What the 2026 Version Actually Looks Like

The 2017 pillar page was essentially a long-form blog post with an aggressive table of contents and downloadable PDF. The 2026 version is a fundamentally different artifact. The anatomy:

The structural elements matter individually. Together they compound.

An above-the-fold definition. The first 60 to 120 words of the pillar must contain a clean, declarative definition of the topic that an AI model can extract verbatim and cite. This is the single highest-leverage edit you can make to an existing pillar. Backlinko's definition-first opener on its keyword research pillar is the canonical example — the first paragraph is a self-contained definition that gets quoted across hundreds of AI responses without modification.

A genuine table of contents with anchor links. Not a sidebar widget. An in-content, scannable list of every H2 section that links to the relevant page anchor. The TOC serves dual purposes: it gives human readers navigation, and it gives AI crawlers an explicit map of the document's topical coverage. Pillars without a structural TOC are systematically less likely to have their internal H2 sections retrieved as standalone chunks.

Aggressive H2 architecture aligned to retrieval queries. Each H2 should be a question or topic that a real user would type into ChatGPT. The pillar functions partly as a set of mini-articles bundled inside one document, each H2 serving as a retrievable answer to a discrete query. Backlinko's pillars typically have 8 to 14 H2 sections, each one a self-contained answer that could stand alone if extracted.

Bidirectional links to every cluster article. The pillar links out to each supporting cluster article, and each supporting article links back to the pillar. This is non-negotiable. Bidirectional linking is the structural signal that the retrieval system uses to understand the cluster as a coherent topical unit. Sites that link out from the pillar but do not link back from the supporting articles get roughly half the citation lift of sites that maintain bidirectional links.

A comparison table. At least one structured comparison table inside the pillar — comparing tools, approaches, definitions, or vendor options. Tables are retrieved as discrete units by AI assistants and frequently cited verbatim in answers to comparison queries. Pillars without a table miss a high-leverage retrieval surface.

Schema markup that goes beyond Article. The 2026 pillar carries Article schema, FAQ schema for any embedded Q&A, HowTo schema for any numbered playbook, and ItemList schema for any ranked or grouped list. The schema is read by AI crawlers as a hint about the structure of the content and increases the likelihood that the right chunks get retrieved for the right queries.

A visible last-updated date and update cadence. Pillars that show a recent update date are cited at higher rates than pillars with stale dates. The cadence should be quarterly at minimum, with substantive content additions — not just a date change. AI models cross-reference update dates against actual content changes and discount pages that bump dates without changing substance.

Author entity with credentials. A pillar page byline should connect to a real author entity — a person with a public profile, credentials, and a consistent body of work on the topic. AI models use author signals as part of source quality scoring. Anonymous or generic-byline pillars are systematically discounted.

How Big Should the Cluster Be?

The cluster sizing question is where most operators get stuck. The honest answer is that it depends on the topic, but there are heuristics that work.

The decision framework has three inputs.

Subtopic count. How many distinct subtopics exist that a serious reader would expect coverage of? Email deliverability has roughly fifteen to twenty subtopics — SPF, DKIM, DMARC, BIMI, sender reputation, bounce categorization, ESP comparisons, warm-up sequencing, IP rotation, content scoring, blocklist remediation, feedback loops, double opt-in, list hygiene, segmentation, deliverability monitoring tools, transactional versus marketing isolation, IP versus domain reputation, and so on. The cluster should cover each subtopic with a dedicated article. That implies fifteen to twenty supporting pieces.

Comparison surface. How many vendor or framework comparisons does the topic naturally generate? For a category like project management, comparison surface is large — Linear vs Jira, Asana vs Monday, ClickUp vs Trello, and so on. Each comparison gets its own supporting article. That alone can add ten to twenty pieces to the cluster.

Query volume distribution. What is the long-tail distribution of relevant queries? Topics with a steep long tail (a few high-volume head queries and many low-volume tail queries) benefit from larger clusters that each capture a few tail queries. Topics with a flat distribution work fine with smaller clusters.

Combining these inputs produces a sizing recommendation. The rough rules of thumb:

The Ahrefs data on this question is instructive. The company has explicitly stated that ten well-built supporting articles outperform thirty thin ones. The marginal value of adding a fifteenth supporting article is much higher than the marginal value of adding a forty-fifth, because the first fifteen typically cover the high-priority subtopics and comparison surface, while later additions move into long-tail territory with diminishing returns.

A pragmatic 2026 approach: ship the pillar plus the first ten supporting articles in the launch quarter. Audit which of those articles earn at least one AI citation per quarter. If the cluster is producing citation lift, add another ten supporting pieces over the following two quarters, targeted at the gaps the audit revealed. Stop adding pieces when new additions stop earning citations.

Case Study: HubSpot's Twelve Pillars

HubSpot is the cleanest case study because the company invented the framework and has maintained its pillars through both the unfashionable years and the AEO revival. The company's topic clusters and pillar pages model was introduced in 2017, scaled aggressively through 2020, deprioritized between 2021 and 2023, and revived as the centerpiece of its AEO strategy in 2025.

The twelve pillars that anchor HubSpot's content program in 2026 cover the company's core competitive surface area: inbound marketing, content marketing, SEO, social media marketing, email marketing, lead generation, marketing automation, CRM, sales enablement, customer service, website building, and analytics. Each pillar is between 7,000 and 11,000 words. Each is supported by a cluster of twenty to forty articles. Each has been substantively updated at least quarterly since the company's 2025 strategic reset.

The performance data HubSpot has shared is striking. The twelve pillars represent less than 0.1% of the company's total indexed URLs. They account for an estimated 38% of the company's measurable AI assistant traffic and roughly 22% of all pipeline attributed to organic and AI sources combined. The pillars are by some distance the highest ROI content assets the company owns.

Three structural choices distinguish HubSpot's pillar program from the 2017 version.

Pillars are now treated as products, not articles. Each pillar has a dedicated product manager equivalent — a content strategist with explicit ownership of the page's performance, update cadence, and cluster maintenance. The role exists outside the editorial calendar. Pillars are roadmapped, not scheduled.

Cluster maintenance is the work. Adding new supporting articles is half the program. The other half is auditing existing supporting articles, refreshing them when product reality changes, and pruning pieces that have lost relevance. HubSpot rotates roughly 15% of its cluster articles per quarter through a refresh-or-retire review.

Comparison surface is built deliberately. Each pillar's cluster includes substantive vendor-comparison content even in categories where HubSpot is a vendor itself. The comparison content gets cited in AI answers about competitor products, which extends HubSpot's citation surface in ways that pure inbound-marketing content cannot.

For SaaS operators looking at how the comparison surface specifically drives citation distribution, the comparison and vs-pages playbook for AEO recommendation dominance covers the comparison-page mechanics in depth.

Case Study: Backlinko's Compounding Pillars

Backlinko's pillar program is smaller and more focused than HubSpot's but arguably more efficient on a per-pillar basis. The company maintains roughly eighteen pillar pages, each in the 5,000 to 9,500 word range, focused on the core surface of SEO practice: link building, on-page SEO, keyword research, SEO copywriting, technical SEO, local SEO, mobile SEO, video SEO, ecommerce SEO, SaaS SEO, and so on.

The company has been transparent in blog posts and webinars that its pillar pages produce a substantially higher organic and AI citation return than its standalone blog content. Three patterns from the Backlinko playbook are worth highlighting.

Comprehensive comparison and example coverage. Each Backlinko pillar contains a substantial section that walks through real examples — specific brands, specific tactics, specific outcome data. The example-heavy structure is highly retrievable by AI assistants because each example is a self-contained chunk that answers the implicit question of what this looks like in practice. Other pillars without comparable example density get cited far less for the same queries.

Updated annually with substantive additions. Backlinko publicly stamps each pillar with the current year (link building in 2026, keyword research in 2026) and refreshes the content meaningfully each January. Year-stamped content is preferred by AI assistants for queries that imply recency, and the annual refresh signals ongoing investment in the topic.

Disciplined cluster sizing. Most Backlinko pillars are supported by ten to fifteen cluster articles rather than the thirty-plus that some competitors run. The cluster pieces are typically deeply researched, 2,500 to 4,500 word articles in their own right. The strategy bets on quality density per supporting piece rather than volume, and the bet has paid off in citation share.

The Backlinko case demonstrates that a pillar program does not require massive content production volume to succeed. Eighteen pillars with disciplined cluster maintenance can outperform programs with ten times the published URL count if the editorial care per piece is higher.

Case Study: Ahrefs and the Engineering of Topical Coverage

Ahrefs runs the most analytically rigorous pillar-cluster program of the three reference cases. The company's content team has published extensively on its own methodology, which combines query-volume analysis, competitive citation gap analysis, and ongoing content refresh cycles.

Several features of the Ahrefs approach are distinctive.

Pillars are slightly shorter, clusters are larger. Ahrefs pillars run 4,500 to 6,500 words on average, shorter than HubSpot's or Backlinko's. The cluster around each pillar typically has fifteen to twenty-five supporting pieces, larger than Backlinko's clusters and comparable to HubSpot's. The reasoning, as the company has explained, is that retrieval rewards interconnected coverage more than it rewards single-document length. A shorter pillar that is densely linked to a larger cluster produces more retrievable chunks in aggregate than a longer pillar with fewer cluster pieces.

Content refresh is quantitative. Ahrefs measures the organic and AI citation performance of each pillar and supporting article monthly. Pieces that lose more than 15% of their traffic month-over-month enter a structured refresh queue. The refresh process is templated — competitor analysis, content gap identification, structural updates, freshness improvements — and produces measurable bounce-back in performance within four to eight weeks.

Internal linking is engineered, not editorial. Ahrefs uses its own internal-link analysis tools to ensure that each pillar's supporting cluster has the right link topology — bidirectional pillar-to-cluster links, lateral cluster-to-cluster links for related subtopics, and breadcrumb hierarchy that signals the cluster's structure to crawlers. This is treated as a technical SEO problem rather than an editorial one, with explicit standards and audits.

The Ahrefs model is particularly applicable for operators with technical or engineering-oriented audiences who appreciate the analytical rigor. The model is also more replicable than HubSpot's or Backlinko's because Ahrefs has published its methodology in considerable detail.

The Hub-and-Spoke Architecture for AEO

The pillar-cluster framework is one specific instance of a broader information architecture pattern: hub-and-spoke. The hub is the canonical topical authority document. The spokes are the supporting articles that cover specific subtopics, comparison surface, methodology, and tactical depth. The architecture connects the hub to every spoke and connects spokes to other spokes where there is topical adjacency.

For AEO purposes in 2026, the hub-and-spoke architecture needs to be paired with two other content surfaces that the original 2017 framework did not contemplate.

Glossary and definition pages. A serious topical hub in 2026 includes a layer of clean definition pages — one per key term in the topic's vocabulary — that anchor the terminology the cluster uses. These definition pages get cited heavily by AI assistants for definitional queries (what is X) and serve as the canonical source of truth for terminology across the rest of the cluster. The mechanics of building this layer well are covered in detail in the glossary and definition pages for AEO training corpus strategy playbook.

Comparison pages as part of the cluster. Comparison pages — head-to-head, alternatives-to, and best-for-Y — function as a third type of spoke alongside the standard subtopic articles. They serve a distinct query intent (comparison) and get cited in distinct query patterns, but they reinforce the topical authority of the cluster as a whole.

Together, the hub, the standard subtopic spokes, the definition pages, and the comparison pages form what we have started calling a topical mesh — a network of mutually reinforcing documents that, taken as a unit, dominate the retrieval index for the topic. The pillar page is the gravitational center, but the surrounding mesh is what produces the citation moat.

The 90-Day Pillar Build Playbook

For operators starting a pillar program in 2026, the prioritized execution sequence:

1. Topic selection and audit. Choose one topic where your brand has genuine subject matter authority, where the topic has meaningful query volume in both classical SEO and AI assistant queries, and where the current AI citation landscape has space for a new authoritative source. Audit the top fifteen pieces of content currently ranking or cited for the topic. Identify the gaps, the structural weaknesses, and the depth opportunities.

2. Cluster mapping. Map out the subtopic structure, the comparison surface, and the definition vocabulary the topic requires. Aim for ten to twenty supporting pieces in the initial map. Sequence them by priority based on query volume and citation gap.

3. Pillar production. Build the pillar first, before the cluster. Target 6,500 to 8,500 words for a typical mid-breadth topic. Include the structural elements from the anatomy section above: above-fold definition, in-content TOC, 8-14 H2 sections, at least one comparison table, FAQ section with schema, visible author entity, and bidirectional link placeholders for the cluster (initially internal anchors, populated as cluster pieces ship).

4. Cluster shipping cadence. Ship two supporting cluster pieces per week for the first ten weeks. Each cluster piece should be 2,000 to 3,500 words, deeply researched, with bidirectional links to the pillar and lateral links to adjacent cluster pieces. Avoid the temptation to outsource cluster production to writers who do not understand the topic — the shallow content discount is severe.

5. Definition layer. In parallel with cluster production, ship a layer of clean definition pages — one per key term — that the cluster references consistently. These pages should be 800 to 1,500 words each, structured as definition first then deeper context, and linked from every cluster piece that uses the term.

6. Comparison layer. Add three to five comparison pages (head-to-head, alternatives-to, best-for-Y) targeting the most relevant comparison queries in the topic. Comparison pages should be 3,000 to 6,000 words with substantive coverage of each option, not defensive marketing copy.

7. Internal linking audit. Once the initial cluster, definition layer, and comparison layer are shipped, audit the internal linking topology. Verify that the pillar links bidirectionally to every supporting piece. Verify that supporting pieces link to adjacent cluster pieces where topically relevant. Identify and fix orphan pages. Verify that breadcrumb hierarchy signals the cluster's structure to crawlers.

8. Schema and structural markup. Apply Article, FAQ, HowTo, and ItemList schema as appropriate across the cluster. Ensure the pillar carries multiple schema types reflecting its multi-format content. Validate schema implementation with a structured data testing tool.

9. AI citation tracking. Instrument citation tracking using a tool like Profound, SerpRecon, or Bluefish. Track the pillar and the cluster's citation share weekly. Identify which pieces are earning citations and which are not. Use the data to inform the next round of cluster additions.

10. Quarterly refresh. At the 90-day mark, audit the program's performance. Refresh the pillar with any substantive updates. Prune cluster pieces that have not earned citations. Add new cluster pieces targeting the gaps the citation audit revealed. Establish a quarterly refresh rhythm.

This is more disciplined work than most content programs are accustomed to. It is also substantially higher ROI per hour invested than the standard alternative of producing more atomized blog content. Operators who shift even a portion of their content budget into pillar program work consistently see meaningful AI citation lift within six to nine months — a timeline that the SEO playbooks of the 2010s would have considered impossibly fast.

What Kills Pillar Performance

A few patterns we have observed repeatedly that destroy pillar program performance:

The orphan pillar. A pillar page published without a supporting cluster, or with a cluster that does not link back, performs roughly 60% worse than a properly interlinked cluster. The cluster is the pillar's distribution infrastructure.

The contractor cluster. Cluster pieces produced by writers who do not understand the topic produce shallow, generic content that AI models detect and discount. The signal is unmistakable in retrieval rankings. Pillars supported by genuinely expert cluster pieces consistently outperform pillars supported by outsourced volume.

The static pillar. Pillars that are published once and never updated lose citation share within twelve to eighteen months. AI models read freshness signals seriously, and a pillar that has not been substantively updated in two years gets discounted as a legacy source.

The pillar with no comparison surface. A pillar that does not contain at least one comparison table and that is not supported by comparison cluster pieces gives up a significant portion of available citation surface. Comparison queries are some of the highest-intent queries in any category, and pillars that ignore them forfeit the citation share that comparison content produces.

The pillar in a JavaScript app. Pillars that render client-side or that gate substantial portions of the content behind interactive widgets are partially invisible to AI crawlers. Server-side rendering, HTML-first content, and minimal JavaScript dependency are baseline requirements for AEO performance.

The pillar without an author. Anonymous or generic-byline pillars are systematically discounted by AI models that use author entity signals as part of source quality scoring. The author byline should connect to a public profile with credentials.

For a broader view of the structural patterns that consistently underperform in AI search across content formats, the Search Engine Journal coverage of AEO failure modes and the Moz analyses of evolving search dynamics are both useful reference reading.

Takeaway: The pillar-cluster framework that HubSpot introduced in 2017 was an architectural answer in search of a retrieval system that would reward it. Google's classical algorithm only partially did. LLM retrieval does. The result is that pillar pages are now structurally the most efficient content asset available for AEO purposes, and the operators who maintain or build serious topical hubs in 2026 are compounding citation share faster than competitors that have stayed with atomized blog content. The window to ship pillar programs before category defaults harden in AI training corpora is real and closing. Brands that ship 5,000 to 12,000 word pillars with disciplined fifteen-to-thirty piece clusters in the next two quarters will own their categories in AI citations through 2028 and beyond. The framework is not new. The retrieval system that finally rewards it properly is.