The 5-Category Content Architecture That Wins AI Search

Last updated: 2026-10-12

TL;DR

• Definition: The 5-Category Architecture is a content taxonomy of identity, concept, comparison, framework, and persona pages, where each category answers a different kind of buyer question.
• Why shape matters: Google says its AI features may issue multiple related searches across subtopics and data sources [1]. In Ahrefs’ March 2026 analysis of 863,000 keywords, only 37.9 percent of cited AI Overview URLs also ranked in the top 10 for the same query [2].
• Why long-tail pages matter: The average AI Mode query ran 7.22 words against 4.0 for traditional Google search in Semrush’s mid-2025 clickstream data [3].
• Our evidence: A startup team published 77 posts dated September 11, 2026, and four days later its site led Google AI Mode’s answer to a category query with buying intent [4]. That is one case with several confounders, not a controlled test.

In Ahrefs’ March 2026 analysis of 863,000 keywords, 37.9 percent of cited AI Overview URLs also ranked in Google’s top 10 for the same query [2]. The other 62 percent ranked lower. The technique that plausibly explains it is one Google documents: AI Overviews and AI Mode may issue multiple related searches across subtopics and data sources before writing one answer [1]. The link is Ahrefs’ inference, since Google does not show the searches it generates [2].

That changes what a content plan is for. If your buyer’s question is answered by several searches you never see, one excellent post can answer only one of them. The question shifts from “is this post good?” to “does a page exist for each question the engine will ask?”

This is the fifth pillar in our series on why AI doesn’t know your business. It answers one question: what content structure actually ranks across ChatGPT, Perplexity, Google AI Mode, and Claude? Nobody has proven one structure across all four. What I can offer is a five-part taxonomy, the evidence for each part, and a line between what is documented and what I infer. I’m writing for pre-Series B AI founders with a few strong posts and no system around them. If the Citation Gap is the problem, this is the shape of the fix.

What is the 5-Category Architecture?

The 5-Category Architecture is a content taxonomy of identity, concept, comparison, framework, and persona pages, built so that each category answers a different kind of question a buyer asks an AI engine. I adapted it for twosetai.com from a playbook I studied at a startup, described below [4]. Each category has one job.

Category	Job	Example URL slug (illustrative)	Query shape it answers	Target posts
A. Identity	Own your name in AI answers	/about/, /methodology/	”What is TwoSetAI?” “Who is Angelina Yang?“	5 to 8
B. Concept explainers	Define terms and become the citation anchor for new vocabulary	/blogs/the-citation-gap/	”What is the Citation Gap?“	10 to 15
C. Comparison articles	Win category-level “vs” and “best” queries	/compare/amplification-package-vs-profound/	”X vs Y,” “best X for Y”	8 to 12
D. Framework guides	Equip the buyer at the decision stage	/frameworks/how-to-evaluate-a-founder-visibility-service/	”How do I choose…,” “how much should… cost”	8 to 10
E. Persona pages	Capture long-tail queries shaped like one buyer’s situation	/for-fintech-founders/	”AI visibility for fintech founders”	15 to 25

The counts are targets from our own plan; nobody has published a tested ratio. The order matters more. Identity and concept pages come first because every later page links back to them, and persona pages come last because they borrow from the other four.

Why does AI search reward a taxonomy instead of a few brilliant posts?

Google calls this technique query fan-out [1]. Semrush’s topic cluster guide draws the practical conclusion: a well-built cluster gives an AI system a page to point to for each sub-query [5]. I call the share of those sub-questions your site can answer with a dedicated, citable page Fan-Out Coverage. You cannot measure it exactly, because Google does not expose its searches [2]. You can approximate it by writing down the sub-questions yourself and counting your pages.

Three more findings point the same way.

• Questions are getting longer. The average AI Mode query ran 7.22 words against 4.0 for traditional Google search, in Semrush’s clickstream data from almost 69 million sessions [3]. Andreessen Horowitz put the average LLM query at 23 words against 4 [6]. More context means more sub-questions.
• Depth beats volume. AI search strategist Jean-Yves Scauri, interviewed on the Microsoft Advertising blog, advises deep coverage of a small number of core topics instead of thin content [7].
• Retrieval works on passages and hierarchies. Perplexity says the index behind its Search API splits documents into sub-document units scored individually against the query [8]. Stanford’s RAPTOR, a research system and not a commercial engine, organizes text into a tree of summaries and improved the best reported QuALITY accuracy by 20 percentage points with GPT-4 [9]. Information architecture, the older discipline, treats organization and labeling as core components [10].

Here is what these findings do not show. No study I found compares a five-category site with a same-size site that has no structure. HubSpot’s documentation says creating topic clusters in its tool does not directly affect SEO [11]. Ahrefs’ 38 percent figure was about 76 percent in July 2025, but its parsing method changed in between, so the two are not a clean trend [2]. And Google says AI Overviews and AI Mode have no additional technical requirements beyond being indexed and eligible for a snippet [1]. So the architecture is not a trick. It makes sure that for each question an engine asks, a page exists that can answer it.

Why is one brilliant post not a moat?

AI engines do not reward popularity. In OtterlyAI’s analysis of 100 million+ AI citations, views, likes, and subscribers for cited YouTube videos each correlated with citation at roughly zero, while description length correlated at about 0.31 [13].

Alex Reichenbach (CEO, Structify) was describing enterprise data pipelines when he said:

“Text-to-SQL, for instance, is now sufficiently easy to do that it has basically become a commoditized thing. Doing large-scale data pipelines is a next step that we need to make sure doesn’t become commoditized.” [14]

You can hear it on YouTube, see the watch page, or read the write-up on his 2 percent accuracy moat. I’m borrowing the shape of his argument, not claiming he said it about content. In my view, one strong explainer is the text-to-SQL of content, easy for a capable writer or an AI tool to match. Fifty connected pages covering the long tail of your buyers’ questions, in vocabulary you defined, are harder to match. The work is in the edge cases.

What does each category do, and how strong is the evidence?

A. Identity: who is this?

Identity pages answer questions about you or your company by name. They carry less of the load than founders assume. McKinsey puts a brand’s own website at about 5 to 10 percent of the sources AI search references [15], and Scauri cites research putting it near 4 percent [7]. But identity pages are the share you fully control, and they decide whether the facts AI repeats are yours. Google’s guidance asks whether readers can tell who wrote the content and whether bylines lead to author information [16].

Our own audit shows why. In August 2026, When ChatGPT did not cite a YouTube video, it reached guests through third-party authoritative posts (founder profiles, twosetai.com pages). That is why the Founder Visibility Program ships a written editorial layer alongside each recording [17]. Where our video failed, identity-shaped pages carried the guest. Weight: medium.

B. Concept explainers: what does this term mean?

Concept pages define a term and give it a stable address. My template is Latent Space’s June 2023 essay “The Rise of the AI Engineer,” which argued that software engineering would spawn a new subdiscipline and named it [18]. I can’t prove the essay is why the term spread, so I use it only as a pattern: name it, define it, keep using the name.

The documented support is indirect: Google’s helpful content guidance asks whether content offers original information, reporting, research, or analysis [16], and Scauri says original data, expert perspectives, and case studies are the hardest to summarize or reproduce [7]. Our coined terms are the Citation Gap, the Buyer Bucket, the Founder-Voice Moat, the Editorial Layer, and this architecture. Whether any engine treats them as ours is untested. Weight: medium for the page type, none yet for the anchor effect.

C. Comparison articles: which one should I pick?

Comparison pages answer “X vs Y” and “best X for Y” queries. The closest evidence is about listicles, not comparisons. In Peec’s analysis of more than 1 million citations (February 2026), listicles were just under one-fifth of the URLs in the ChatGPT data, and 52 percent of those were cited more than twice in one answer [19]. Peec did not break out comparison pages, so I read that as supportive, not conclusive.

The caution is who writes the comparison. A September 2025 study found ChatGPT, Perplexity, and Gemini lean toward earned media over brand-owned content [20]. Your own “us vs them” page is brand-owned, so it has to be honest, including saying when a rival is better for a use case, and it needs corroboration elsewhere. Weight: medium to low.

D. Framework guides: how do I decide?

Framework pages help a buyer decide: how to evaluate a founder visibility service, or how to choose among a PR agency, an SEO agency, and a GEO agency. A framework is original analysis, the kind of content Scauri calls hard to summarize or reproduce [7].

This is the category where my claim is least proven. That framework content outranks generalist sites comes from our internal audits, not a published study, so I treat it as a hypothesis to test on twosetai.com. Weight: low to medium.

E. Persona pages: what about someone like me?

Persona pages answer long-tail queries that carry one buyer’s situation: “AI visibility for fintech founders.” The support is the query-length trend [3][6] plus fan-out [1]. Longer questions carry a role, an industry, and a constraint, and a page written for that situation matches more of them. That last step is my inference. Persona pages are also the likeliest to turn thin, the failure Scauri warns about [7], so ours draw on real guest conversations. Weight: medium on the query trend, low on the page-type effect.

How do the five categories link to each other?

Hub-and-spoke is the standard model, and the vendors describe it alike. HubSpot says supporting content should link back to the pillar page [11]. Ahrefs describes a hub that links to every subpage, with each subpage linking back [12]. Semrush adds links between cluster pages using keyword-rich anchor text [5]. Google lists making content easily findable through internal links as a practice for its AI features [1] and says link text should tell readers and Google what the linked page is about [21].

Our rules, taken from the playbook we studied:

• Ten edges per post. On the page we checked, each post carried five contextual body links and five Related Reading links. We render ours as plain HTML so crawlers see them. The 10-Edges-Per-Post Rule spoke has the detail.
• Concept pages get the most inbound links. Comparisons link to concepts for definitions and to frameworks for the decision. Frameworks link to comparisons for named examples. Personas link to frameworks, comparisons, and identity pages.
• Paths group the categories (/blogs/, /compare/, /frameworks/). Google says keywords in a URL path have hardly any effect on rankings beyond breadcrumbs, though grouping similar topics in directories helps it learn how often those URLs change [21]. I use paths for readers and audits; see URL Path as Topic Signal.

At 60 posts the taxonomy also does an organizational job. Hector Hernandez (Co-founder and COO, Speakeasy) was talking about companies scaling their use of AI when he said:

“That’s why the framework is so important, right? You don’t want every single individual trying to figure out how am I supposed to do this correctly.” [22]

His watch page has the full conversation, and you can jump to the moment. Again I’m borrowing the shape of his point. A content taxonomy is a framework in his sense: when every post has a category, nobody writing the fortieth post has to work out what it is for.

What happened when a team shipped the architecture all at once?

In September 2026, a startup team published 77 posts on a new domain, all dated September 11, organized into these five categories. By September 15, four days later, the site led Google AI Mode’s answer to a category query with buying intent [4]. I’m not naming the company.

It is one site, one query, one engine, and one observation date. The playbook also included coined vocabulary, comparison articles, and manual indexing requests, so I can’t credit the structure alone. I treat it as a reason to test on our own domain, not as proof.

Our best result on record is a different kind of evidence. In August 2026, ChatGPT labeled our episode with Alexander Berger (COO, Bolt.new) “Best overall,” Perplexity called it “the best fit,” and Google AI Mode cited it across multiple sections of his bio. The full stack was live: YouTube full episode and highlight, Medium posts, Reddit posts, and a watch page [23]. You can watch the conversation. That is not one brilliant asset. It is coverage across surfaces, the same idea at the scale of one guest.

We are now applying the architecture to twosetai.com: about 60 posts, with a working split of 8 identity, 12 concept, 12 comparison, 10 framework, and 18 persona. A later post in this series reports what happens, including if it does not work.

What does the architecture not do?

Three limits.

It is the owned layer, not the whole picture. Your own pages are a small share of what engines read [15][7], and ChatGPT, Perplexity, and Gemini lean toward earned media [20]. It does not replace being discussed elsewhere, or the Editorial Layer, the human-authored, positioning-anchored content on top of automated monitoring. More in Why Monitoring Tools Can’t Fix Your AI Visibility.

It needs indexing first. A page must be indexed and eligible for a snippet to be a supporting link in AI Overviews and AI Mode [1]. SEO practitioner Lily Ray said at Affiliate Summit West 2026 that ranking well in Google raises the likelihood of AI citation [24]. Seventy-seven unindexed posts help nobody.

Measure presence rate, not single runs. In SparkToro’s study of 2,961 runs, the chance of getting the same brand list twice from ChatGPT or Google’s AI was under 1 in 100 [25]. The architecture is about coverage, measured across repeated runs.

How do you build it, and in what order?

The order I recommend:

1. Positioning first. TwoSetAI runs proprietary positioning steps for every founder guest, drawing on classical PMM frameworks and adapted for AI-search citation. Every page should be able to name the one claim it supports.
2. Identity next. Home, about, methodology, founder page. State the same facts the same way everywhere.
3. Concept pages second. Define the terms you want to own before leaning on them in comparisons.
4. Comparisons and frameworks third. Write comparisons honestly. Write frameworks from real decisions your buyers made.
5. Persona pages last. Each should rest on a named conversation or case.

Then put your own voice through all of it. Fifty templated pages are the failure case; a founder’s specific stories and numbers are what a competitor cannot copy in an afternoon. That is the Founder-Voice Moat, which is why we start from a recorded conversation, and video gives the pages real material to quote.

How do you know the architecture is working?

Three checks, weekly.

1. Index check. Is each new page indexed in Google Search Console and Bing Webmaster Tools?
2. Presence rate per category. For each category, write five prompts a buyer would type. Run each several times across ChatGPT, Perplexity, Google AI Mode, and Claude, and log whether your page appears [25].
3. Fan-Out Coverage. For your five most valuable buyer questions, list ten likely sub-questions each, then count how many have a dedicated page. “Thirty-one of fifty” is a number you can improve.

When a page you expected to be cited isn’t, the question shifts to where in the retrieval chain it failed. Rodrigo Stockebrand’s O’Reilly book on answer engine optimization describes a complementary five-gate RAG visibility funnel: Discovery, Retrieval, Selection, Inclusion, and Representation. A citation can break at any gate, and the fix depends on which one broke [26]. Our architecture decides which pages to build. His funnel is where to look when one of them isn’t working.

The measurement spoke gives the weekly template. A Citation Gap audit shows where you stand today, and these checks show whether the architecture is closing it.

What should you do this week?

So, what content structure actually ranks across ChatGPT, Perplexity, Google AI Mode, and Claude? None is proven across all four. The best-supported answer I have is a site with a page for each question the engine is likely to ask, sorted into five jobs: identity, concept, comparison, framework, and persona. It favors coverage over brilliance because these systems retrieve by sub-question and by passage.

This week, list your five most valuable buyer questions and write ten sub-questions under each. Map every sub-question to a category and mark the ones with no page. Write the missing identity page first, since everything else links back to it.

That is the work we do at TwoSetAI: we turn one podcast conversation into lasting AI-search presence for AI founders whose buyers check AI before they check the website. When your buyers ask AI, be the answer.

Next in this series: Category A: Identity, Owning Your Name in AI Answers, Hub-and-Spoke Internal Linking: The 10-Edges-Per-Post Rule, and Measuring Architecture Health.

Who is TwoSetAI?

TwoSetAI is one of the top AI podcast and video studios in Silicon Valley, and the team behind this article, hosted by Angelina Yang. The show has featured 100+ AI founders and leaders on how real AI products get built, sold, and trusted. Alongside the podcast, TwoSetAI runs the Founder Visibility Program that turns each conversation into the articles, clips, and pages that make its guests easy to find. Watch full episodes on YouTube at youtube.com/@TwoSetAI, and find every guest and episode at twosetai.com.

Glossary

• 5-Category Architecture (also called: GEO content taxonomy, five-category content system). A content taxonomy of identity, concept, comparison, framework, and persona pages, each built to answer a different kind of buyer question.
• Fan-Out Coverage (also called: sub-question coverage). The share of the sub-questions behind a buyer’s question that your site answers with a dedicated, citable page.
• Query fan-out (also called: sub-query expansion). Google’s term for AI Overviews and AI Mode issuing multiple related searches across subtopics and data sources to build one answer [1].
• Hub-and-spoke (also called: topic cluster, pillar and cluster model, content hub). A site structure where a broad hub page links to detailed spoke pages and each spoke links back [11][12].
• The Citation Gap (also called: AI-answer gap, brand invisibility in AI search). The space between what AI says about your brand and what your business actually is.
• Editorial Layer (also called: GEO editorial layer). Human-authored, positioning-anchored content on top of automated AI-visibility monitoring.

About Angelina Yang

Angelina Yang is VP of AI, a fast.ai fellow, and the host of the TwoSetAI podcast, where she has interviewed 100+ AI founders and leaders about how real AI products get built, sold, and trusted. Find her on LinkedIn, talk to her AI clone on Delphi, watch the podcast on YouTube, or browse twosetai.com.

Frequently Asked Questions

What is the 5-Category Architecture? It is a content taxonomy of identity, concept, comparison, framework, and persona pages. Each category answers a different kind of buyer question, and the categories link to each other in a defined pattern. It is TwoSetAI’s term, not an industry standard [4][5].

How do I structure a website for AI search? Make sure a page exists for each question an engine is likely to ask, grouped by job and linked together. Google says AI Overviews and AI Mode may issue multiple related searches across subtopics and data sources, and that internal links help make content findable [1]. Nobody has proven one structure across all engines.

Do topic clusters still matter for AI search? Semrush says a well-built cluster gives an AI system a page for each sub-query [5]. HubSpot says creating clusters in its tool does not directly affect SEO [11], and I found no controlled test of clusters against unstructured sites. Treat clusters as a sound way to organize content, not a guaranteed lever.

What is query fan-out and why does it matter for my content? It is Google’s term for AI Overviews and AI Mode issuing multiple related searches across subtopics and data sources to build one answer [1]. A page can be cited for a sub-question without ranking for the typed query. In Ahrefs’ data, 37.9 percent of cited AI Overview URLs ranked in the top 10 for the typed query [2].

How many posts do I need before AI engines cite my site? Nobody publishes a threshold. In the one case we studied, 77 posts preceded a Google AI Mode lead four days later [4]. Scauri advises depth on a few core topics over volume of thin content [7]. Start with the identity pages, then add by category.

Do comparison articles help me show up in ChatGPT? Possibly. Listicles were just under one-fifth of the URLs in Peec’s ChatGPT data [19], but Peec did not isolate comparisons. A study of ChatGPT, Perplexity, and Gemini found a bias toward earned media over brand-owned pages [20], so write comparisons honestly and seek corroboration.

Do I need special markup to appear in AI Overviews? Google says there are no additional technical requirements beyond being indexed and eligible to appear with a snippet, and that standard SEO practices apply [1].

How do I know if my content architecture is working? Check indexing, then measure presence rate per category by repeating buyer prompts across engines, and count your Fan-Out Coverage. Repeat runs matter because the same brand list came back under 1 time in 100 in SparkToro’s study [25].

References

[1] Google (2026). “AI features and your website.” Google Search Central, last updated December 10, 2025, accessed October 12, 2026. https://developers.google.com/search/docs/appearance/ai-features

[2] Linehan, L. (2026). “Update: 38% of AI Overview Citations Pull From The Top 10.” Ahrefs, March 2, 2026 (863,000 keyword SERPs, about 4 million AI Overview URLs; Ahrefs Brand Radar data). https://ahrefs.com/blog/ai-overview-citations-top-10/

[3] Harsel, L. (2025). “Google AI Mode’s Early Adoption and SEO Impact.” Semrush, July 30, 2025 (clickstream data, almost 69 million US desktop Google Search sessions, May 1 to July 5, 2025). https://www.semrush.com/blog/google-ai-mode-seo-impact/

[4] TwoSetAI (2026). Observation of a startup’s 77-post site launch (posts dated September 11, 2026) and its Google AI Mode result by September 15, 2026. Company unnamed. Internal data; see the Methodology note.

[5] Lindley, A. (2026). “Topic clusters for SEO: what they are & how to create them.” Semrush, September 7, 2026. https://www.semrush.com/blog/topic-clusters/

[6] Cohen, Z. and Amble, S. (2025). “How Generative Engine Optimization (GEO) Rewrites the Rules of Search.” Andreessen Horowitz, May 28, 2025. https://a16z.com/geo-over-seo/

[7] Scauri, J.-Y. and Goodman, A. (2026). “All in on AI: ‘Discovery to Influence in GEO’ Part 1.” Microsoft Advertising, March 30, 2026. https://about.ads.microsoft.com/en/blog/post/march-2026/all-in-on-ai-discovery-to-influence-in-geo-part-1

[8] Perplexity (2025). “Introducing the Perplexity Search API.” Perplexity Hub, September 25, 2025. https://www.perplexity.ai/hub/blog/introducing-the-perplexity-search-api

[9] Sarthi, P. et al. (2024). “RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval.” Stanford; ICLR 2024; arXiv:2401.18059. https://arxiv.org/abs/2401.18059

[10] Rosenfeld, L., Morville, P., and Arango, J. (2015). “Information Architecture: For the Web and Beyond,” 4th ed. O’Reilly Media. https://www.oreilly.com/library/view/information-architecture-4th/9781491913529/

[11] HubSpot (2025). “Topics, pillar pages, and subtopic keywords.” HubSpot Knowledge Base, last updated November 20, 2025. https://knowledge.hubspot.com/content-strategy/pillar-pages-topics-and-subtopics

[12] Hardwick, J. (2020). “Content Hubs for SEO: How to Get More Traffic and Links With Topic Clusters.” Ahrefs, April 2, 2020. https://ahrefs.com/blog/content-hub/

[13] OtterlyAI (2026). “YouTube AI Citation Study 2026.” Updated March 2, 2026. https://otterly.ai/blog/youtube-ai-citation-study-2026/

[14] TwoSetAI (2026). Alex Reichenbach (CEO, Structify) episode; quote at 29:09; written analysis published April 10, 2026. https://www.youtube.com/watch?v=l6Byz148I8s&t=1749s

[15] McKinsey & Company (2025). “New front door to the internet: Winning in the age of AI search.” October 2025 (AI Discovery Survey, 1,927 US consumers, August 2025). https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/new-front-door-to-the-internet-winning-in-the-age-of-ai-search

[16] Google (2026). “Creating helpful, reliable, people-first content.” Google Search Central, last updated October 5, 2026. https://developers.google.com/search/docs/fundamentals/creating-helpful-content

[17] TwoSetAI (2026). Guest search audit, checked August 10 to 25, 2026, incognito, screenshots on file. Internal data; protocol in the Methodology note.

[18] swyx (2023). “The Rise of the AI Engineer.” Latent Space, June 30, 2023. https://www.latent.space/p/ai-engineer

[19] Wells, T. (2026). “What does a good citation rate look like? Benchmarks from over 1 million AI citations.” Peec AI, February 2026. https://docs.peec.ai/research/content-types-cited-most-by-engine

[20] Chen, M., Wang, X., Chen, K., and Koudas, N. (2025). “Generative Engine Optimization: How to Dominate AI Search.” arXiv:2509.08919, September 10, 2025. https://arxiv.org/abs/2509.08919

[21] Google (2025). “SEO Starter Guide: The Basics.” Google Search Central, last updated December 10, 2025, accessed October 12, 2026. https://developers.google.com/search/docs/fundamentals/seo-starter-guide

[22] TwoSetAI (2026). Hector Hernandez (Co-founder and COO, Speakeasy) episode, published October 5, 2026; quote at 22:32. https://www.youtube.com/watch?v=yhOLsoqSSEY&t=1352s

[23] TwoSetAI (2026). Alexander Berger (COO, Bolt.new) episode and three-engine check, incognito, August 2026. https://www.youtube.com/watch?v=8C3C4Vhknx8

[24] Bernstein, J. (2026). “Google Rankings vs AI Overview Citations: 2026 Ahrefs Data.” Something Inc., August 21, 2026 (reports Lily Ray’s remarks at Affiliate Summit West 2026). https://somethinginc.com/blog/lily-ray-seo-foundation-aeo-ahrefs-fan-out-data/

[25] Fishkin, R. and O’Donnell, P. (2026). “New Research: AIs are highly inconsistent when recommending brands or products.” SparkToro, January 28, 2026. https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/

[26] Stockebrand, R. (2026). Answer Engine Optimization. O’Reilly Media, July 2026 (five-gate RAG visibility funnel; Chapter 4). https://learning.oreilly.com/library/view/answer-engine-optimization/0642572275549/

Methodology note. Last verified 2026-10-12. Google statements come from Google Search Central pages dated above. Third-party studies are cited with their dates, and several (Ahrefs, Semrush, Peec, OtterlyAI) come from companies that sell SEO or AI-visibility tools, so treat them as directional. Ahrefs’ 2026 update says its parsing method changed since its July 2025 study and that the fan-out explanation is its authors’ inference [2]. RAPTOR [9] is research on retrieval systems, not a description of how commercial engines work. The “weight” ratings for each category are my judgment, not statistical measures. The TwoSetAI audit [17] covered a cohort of past guests, not the full catalog: about seven queries per guest, run in incognito windows between August 10 and 25, 2026, with a guest counted as present on a surface if any query surfaced them. The 77-post observation [4] is a single case seen by TwoSetAI between September 11 and 15, 2026, on one query in Google AI Mode, and the company is unnamed. The Alexander Berger results [23] were checked in August 2026 and may differ on your run. “5-Category Architecture,” “Fan-Out Coverage,” and the other capitalized terms are TwoSetAI’s own terms, not industry standards. Nothing here guarantees a citation, a ranking, or a business outcome.