Obsurfable"Publish listicles to earn AI citations" and "publish direct-answer pages to earn AI citations" are both common GEO advice. They cannot both be right for every query — and a new nightly tracking study from INSIDEA, with figures computed on 2 October 2026, shows they are not.
Over 28 days, INSIDEA ran 49 unbranded buyer questions through four AI models every night and stored every answer and citation. The result: 2,667 answers carrying 35,314 citations. Question intent — not a universal page-type playbook — determined which sources AI engines retrieved.
What INSIDEA measured
From 5 September to 2 October 2026, INSIDEA asked buyer-style questions across four models with web search enabled:
| Parameter | Value |
|---|---|
| Questions | 49 unbranded (43 for most of the period; 6 added 1 October) |
| Models | GPT-4.1, Claude, Perplexity, Gemini |
| Nightly runs | Every night for 28 days |
| Total answers | 2,667 |
| Total citations | 35,314 |
| Unique domains cited | 623 |
Example prompts: "Which HubSpot partner is best for enterprise implementations?" and "How much does HubSpot onboarding cost?" All figures were computed by a script from raw run files — nothing typed from memory.
Finding 1: Intent splits citation page types
INSIDEA grouped questions by intent and classified every cited URL by page type.
| Question intent | List/comparison page share of citations | Answers citing at least one list |
|---|---|---|
| "Which" / "best" questions | 46.5% | 83.1% |
| "How much" / "how long" questions | 6.6% | Low |
On recommendation questions — "which," "best," "top" — list and comparison pages made up nearly half of all citations. 83.1% of those answers cited at least one list page.
On pricing and timeline questions — "how much," "how long" — lists made up just 6.6% of citations. Those answers cited pages that stated a number or timeline directly.
The same site cannot optimize one page type and win both intent classes. A listicle strategy for "best CRM" prompts will not help for "how much does CRM cost" prompts — and vice versa.
This partially reconciles conflicting industry studies. Evertune found listicles dominate citation formats at scale. Kern Media found company commercial pages dominate on-site citations. INSIDEA shows both are true depending on what the buyer asked.
Finding 2: Citation concentration is extreme
Only 29 of 623 domains were cited in 5% or more of answers. 133 domains were cited exactly once.
| Domain | Share of answers citing it |
|---|---|
| hubspot.com (any page) | 65.9% |
| insidea.com | 50.6% |
| HubSpot partner directory | 48.2% |
In 50.6% of answers, a citation pointed to insidea.com — ahead of HubSpot's own partner directory at 48.2%. A single agency's content was as dominant as the vendor's official directory.
The head is enormous; the long tail is invisible. GEO programs that track "are we cited anywhere?" miss whether they are cited on the prompts that matter — and whether the citations come from the page types those prompts retrieve.
Finding 3: Engines cite at radically different volumes
| Engine | Average citations per answer | Median |
|---|---|---|
| Perplexity | 18.6 | 17 |
| Gemini | 17.3 | — |
| GPT-4.1 | 5.0 | — |
| Claude | 4.6 | — |
Perplexity and Gemini cast a wide net — 17–19 sources per answer. ChatGPT and Claude are selective — 4–5 sources per answer.
This means:
- A Perplexity citation is one of many; a ChatGPT citation is one of few.
- "We got cited by Perplexity" is a different outcome than "we got cited by ChatGPT" in terms of visibility share within the answer.
- Strategies optimized for Perplexity's breadth (directory presence, forum mentions, listicle coverage) differ from strategies for ChatGPT's selectivity (own-site commercial pages, entity authority).
Almost every Reddit, Clutch, and LinkedIn citation in the dataset came from Perplexity. ChatGPT and Claude rarely cited those surfaces.
Finding 4: The same engine disagrees with itself on source selection
Perplexity cited HubSpot's partner directory in 72% of answers. GPT-4.1 cited it in 36.9%. Claude cited it in 3.4%.
| Engine | HubSpot partner directory citation rate |
|---|---|
| Perplexity | 72.0% |
| GPT-4.1 | 36.9% |
| Claude | 3.4% |
For the same question set, three engines produced three different source profiles. Engine-specific optimization is not a nice-to-have — it is the only honest approach.
Finding 5: Vendor pages and partner content compete for the same slots
HubSpot's own pages appeared in 65.9% of answers. The partner directory appeared in 48.2%. A third-party agency (insidea.com) appeared in 50.6%.
Buyers asking about HubSpot partners did not get a clean split between vendor-owned and third-party sources. They got heavy citation of the vendor's own properties and a partner agency's content — competing for the same answer slots.
For agencies and service providers, this is the opportunity: you can out-cite the vendor's own directory on category prompts if your content is more specific, more current, or better structured for extraction. For vendors, it is the threat: your partners' content can displace your official pages in AI answers about your ecosystem.
What this means for content strategy
Map page types to prompt intent, not to a universal playbook
| Buyer intent | Page type to prioritize | Page type to deprioritize |
|---|---|---|
| "Best X," "which X," "top X" | Listicles, comparisons, directories | Pricing pages |
| "How much," "how long," "what does X cost" | Pricing, timeline, spec pages with numbers | Listicles |
| "What is X," "how does X work" | Definitional guides, product pages | Comparison roundups |
Build separate content tracks for separate intent classes. One blog strategy cannot cover all of them.
Measure by intent cluster, not aggregate citation rate
"Our citation rate is 12%" is meaningless without knowing which prompts produced those citations. A brand can dominate "how much" prompts and be invisible on "best" prompts — or the reverse. INSIDEA's HubSpot partner dataset shows a 46.5% vs 6.6% split between intent types.
Obsurfable tracks prompts by intent cluster so you can see which question types cite you and which cite competitors instead.
Weight citations by engine selectivity
A citation in a Perplexity answer (1 of 18 sources) carries different weight than a citation in a ChatGPT answer (1 of 5). Reporting total citations without engine context overstates Perplexity visibility and understates ChatGPT visibility.
Expect partner and competitor content to compete with yours
On ecosystem prompts — "best partner for X," "top agency for Y" — vendor pages, official directories, and third-party agencies fight for the same citation slots. If you are the vendor, monitor whether partner content outranks your directory. If you are the partner, your content can become the authoritative source the engine prefers.
How this connects to other recent findings
Kern Media's study of 5,376 citations found 84% pointed to company websites — but only 6.9% of company-site citations went to blogs. INSIDEA adds the intent layer: on "best" questions, third-party listicles dominate; on "how much" questions, direct-answer vendor pages dominate. The Kern finding is true for ChatGPT-heavy datasets on recommendation prompts; INSIDEA shows it breaks on pricing prompts.
Provena's Copilot study found category guides took 86% of citations while comparison pages got zero — on a single B2B site. INSIDEA's "which/best" questions cited list pages 83.1% of the time. The formats differ (category guides vs listicles) but the intent match is the same: recommendation queries retrieve roundup-style evidence.
How Obsurfable fits
INSIDEA's study covers one vertical's buyer questions over 28 days. Obsurfable lets you run the same nightly tracking for your prompts, your competitors, and your intent clusters — across ChatGPT, Gemini, Perplexity, Claude, and Google AI surfaces.
The Visibility Director identifies which intent classes are failing and drafts content matched to the page types those prompts actually retrieve.
Should we stop writing listicles?
No — for "best" and "which" prompts, listicles and comparison pages are exactly what engines retrieve. Stop writing listicles only for prompts where buyers want a number or a timeline.
Does this mean blogs are useless?
Blogs can serve definitional and thought-leadership intents. They are simply not the primary citation target for commercial recommendation or pricing prompts in this dataset.
How many nightly runs do I need?
INSIDEA ran 28 days. AI citation outcomes are volatile week to week — see AI Citation Volatility. A minimum of two weeks with daily or every-other-day runs gives a stable baseline. Single checks are snapshots, not measurements.
Can one page serve multiple intents?
Rarely well. A page that tries to be a listicle, a pricing sheet, and a product overview usually serves none of them extractably. Match one primary intent per page.
Bottom line
Across 2,667 AI answers and 35,314 citations, question intent — not a universal page-type formula — determined which sources engines retrieved. List pages took 46.5% of citations on "best" questions; 6.6% on "how much" questions. Perplexity averaged 18.6 citations per answer; ChatGPT 5.0. Only 29 of 623 domains appeared in 5%+ of answers.
Build content by intent cluster. Measure by intent cluster. And stop treating "publish listicles" or "publish product pages" as universal advice — the buyer's question decides which one wins.