The Category Citation Gap: Why “Best X Software” Prompts Rarely Cite Vendors

Ask an AI assistant “what is Zian AI” and it will usually reach for the company’s own site. Ask it “best AI appointment setting software” and the vendor pages mostly vanish, replaced by listicles, review aggregators, publisher round-ups and Reddit threads. Same engine, same brand, same week. The difference is not authority. It is intent.

This gap is the single most misunderstood thing in answer engine optimisation, because teams read a healthy brand-prompt result as evidence the strategy is working, then cannot explain why the category prompt that actually generates demand never mentions them.

What the citation data actually shows

The clearest public dataset on this is the Wix Studio AI Search Lab study by Tom Wells, published 23 March 2026, which classified 1,056,727 citations across 75,000 AI answers from ChatGPT, Google AI Mode and Perplexity, segmented by prompt intent. Its headline conclusion is blunt: “Query intent was more predictive of content type citation than both industry and model choice.”

The intent-level breakdown is where the category gap becomes visible:

• Navigational/local prompts (the closest proxy in the study to brand lookups): product pages 21.95%, category pages 18.31%, homepages 13.56%.
• Commercial prompts (the “best X software” family): listicles 40.86%, category pages 12.42%, discussions 11.44%.
• Informational prompts: articles 45.48%, listicles 21.68%, how-to guides 9.21%.
• Transactional prompts: product pages 24.88%, category pages 14.97%, homepages 7.38%.

Homepages were cited for 13.56% of navigational prompts against a 5.26% average across all intents. Read those two rows together and you have the whole phenomenon: the page types a vendor fully controls peak on brand-shaped prompts and collapse on category-shaped ones, where a multi-vendor document takes over.

A second study points the same way from a different angle. AirOps, in a report published 17 October 2025 covering 21,311 brand mentions across more than 500 commercial-intent queries in ChatGPT, Claude and Perplexity, found that “85% of brand mentions came from external domains, while only 13.2% of mentions came directly from the brands domain” — brands being, in their phrasing, “6.5x more likely to be mentioned through third-party sources than their own domains”. That study covers commercial intent specifically, which is exactly the shape of a category prompt.

Why retrieval behaves this way

There is no anti-vendor rule in any of these systems. The behaviour falls out of how the prompt is interpreted.

“Best AI appointment setting software” is a request for a ranked comparison of several named options with reasons. To answer it from vendor pages alone, a model would need to retrieve five or six separate single-vendor documents, normalise their marketing language, and construct a comparison the sources never made. To answer it from one listicle, it retrieves a document that already contains the comparison, in comparative form, with the vendors named alongside each other.

Retrieval optimises for passages that answer the question as asked. A vendor page answers “what does this product do”. It never answers “how do these six products compare”, because a vendor page that genuinely compared its product unfavourably to five rivals would not survive its own marketing review. The document class is structurally wrong for the intent, and no amount of schema markup changes that.

Brand prompts invert the logic. “What is Zian AI” has exactly one authoritative source, and it is the company. The engine has no comparative work to do, so the owned page wins by default. This is also why the brand floor is the easiest thing to secure and the least valuable thing to celebrate.

The trap: being cited is not being recommended

The obvious counter-move is to publish your own “best X software” listicle with yourself at number one. There is now direct evidence on how that performs.

Lily Ray, in an analysis published 17 June 2026, tracked 100 B2B “best [category] software” queries in Google AI Overviews at three checkpoints (15 April, 15 May and 8 June 2026), covering 184 self-promoting listicle pages across 146 brands. Her finding, across the 80 of those prompts that actually returned an AI Overview: “when a brand’s own self-promotional listicle got cited as a source, that brand was left out of the actual recommendation 69% of the time” — 224 of the 323 self-promotional listicles cited. In her dataset, 74 of the 100 prompts returned an AI Overview that cited a self-promoter’s own listicle but left that brand out. She also reports that Forbes, Reddit and YouTube were among the most-cited domains for “best” queries.

In other words, the self-serving listicle can work as a document and fail as a pitch. The engine harvests the comparison, discounts the ranking, and recommends whoever else is in the table.

Scale data suggests these pages are a small slice of the pool to begin with. Peec AI, in research by Tom Wells published 3 March 2026, ran a fixed set of non-branded software-review prompts across six platforms and analysed 232,000 citations across 13,000 unique listicles between December 2025 and February 2026. It reports that “approximately 1 in 10 citations in AI search results come from self-promotional listicles”. Read that denominator carefully: the sample is listicles surfaced by those prompts, so the honest reading is roughly one cited listicle in ten being self-promotional, not one in ten of everything AI search cites. The rate varied sharply by platform — ChatGPT averaged 3.6% against 10.3% on Google AI Mode and 10.4% on Perplexity — which Peec summarises as “ChatGPT stays away from self-promo listicles 3x more than competitors”. Lower is the better number here. Either way, the overwhelming majority of cited comparisons belong to somebody else.

One caveat worth stating, because nobody else will: the Wix Studio and Peec AI studies share an author and a data platform — Wix says its dataset was “created and retrieved in the Peec AI platform” — so they are not independent confirmations of each other. The AirOps report and Lily Ray’s analysis are separate work with separate methods, and they point the same way.

Prompt type, who gets cited, and what you can actually influence

Prompt type	Example	Who typically gets cited	What a vendor can realistically influence
Brand	“What is Zian AI?”	The brand’s own homepage, product and FAQ pages; secondary corroboration from directories	High. Almost fully controllable. Keep facts identical everywhere, answer the question in the first 60 words, publish a real FAQ.
Category	“Best AI appointment setting software”	Third-party listicles and round-ups, review aggregators, publisher comparisons, forum threads	Low, and indirect. You influence whether you appear inside other people’s tables. You do not influence the ranking they publish.
Problem	“How do I stop inbound leads going cold?”	Explanatory articles and how-to guides, including vendor blogs when genuinely instructional (practitioner judgement — not measured)	Medium-high. The most winnable owned surface. Solve the problem properly; the product mention is a footnote, not the point.
Comparison	“Zian vs [competitor]”	Third-party comparisons, review sites, community discussion; sometimes both vendors’ own pages (practitioner judgement — not measured)	Medium. An honest, specific comparison page can be retrieved. A one-sided one gets cited and discounted.
Local	“AI voice agent providers in Australia”	Local directories, regional publisher round-ups, map and business listings, geo-specific listicles (practitioner judgement — not measured)	Medium. Directory and listing presence is cheap and largely in your control; editorial round-ups are not.

Source-ownership note on this table: only the brand and category rows are anchored to measured data — they map onto the navigational and commercial intent splits in the Wix Studio study cited above, with the category row corroborated by the AirOps and Lily Ray findings. The problem, comparison and local rows are practitioner judgement, inferred from the same intent logic; no study cited here reports figures for them, and they are labelled as such in the table. The “what a vendor can realistically influence” column is judgement in every row, including the two measured ones.

The four plays, and where each one runs out

Earn inclusion in other people’s comparisons. The highest-leverage move, because it targets the document class that actually gets retrieved. It is also the slowest: it means pitching editors and analysts, responding to round-up requests, and being findable enough that a writer building a list of ten includes you. The limit is that you control inclusion at best, never position.

Publish genuinely multi-vendor comparison content. Worth doing if it is honest — real alternatives, real trade-offs, real cases where a competitor is the better fit. The Lily Ray data suggests the payoff is not self-recommendation but retrievability and category association. The limit is that a self-ranking list mostly promotes the field.

Review-site presence. Aggregators are heavily represented in commercial-intent retrieval, and profiles are relatively controllable. The limit bites hardest for early-stage products, which often cannot list at all; we covered the workarounds in earning AI citations without a review-site profile.

Community and sentiment surfaces. Reddit and YouTube appear repeatedly in the “best” query citation sets. The limit is that these surfaces punish manufactured participation more reliably than any algorithm punishes thin content, and the failure mode is reputational rather than merely ineffective. Our note on what B2B SaaS should and shouldn’t do about Reddit citations covers the line.

What this means if you are small or in beta

Zian AI is a waitlist and partnership beta, so we will be direct about the expectation rather than dress it up. For a brand at this stage the honest forecast is a reliable brand-prompt floor — an engine asked directly about you will find your site and describe you accurately, provided your facts are consistent — plus slow, uneven category progress that depends mostly on third-party surfaces you do not own. That is not a strategy failure. It is the shape of the data above applied to a young domain.

The practical consequence is a sequencing decision. Secure the brand floor first because it is cheap and fully in your control. Invest next in problem-shaped content, which is the largest winnable owned surface. Treat category prompts as a long campaign of earning third-party inclusion, and measure them separately, because averaging a strong brand result with a weak category result produces a number that tells you nothing. If you are starting from scratch, our practical guide to answer engine optimisation for SaaS covers the groundwork.

Zian’s own products sit on the other side of this funnel: SmartReach AI™ and PrecisionPitch AI™ work the conversations that AI answers eventually send you, across phone, SMS, email and WhatsApp. Getting cited is a separate discipline from converting what the citation delivers, and conflating the two is how teams end up optimising the wrong metric.

Frequently asked questions

Why does ChatGPT cite my website for brand questions but never for category questions?

Because the two prompts require different document types. A brand question has one authoritative source and your page satisfies it. A category question asks for a comparison across several vendors, which no single-vendor page contains. The Wix Studio AI Search Lab study of 1,056,727 citations, published 23 March 2026, found listicles took 40.86% of commercial-intent citations while homepages peaked on navigational prompts at 13.56%.

Should we publish our own “best [category] software” listicle?

Only if it is genuinely even-handed. Lily Ray’s June 2026 analysis of 100 B2B “best software” queries found that when a brand’s own self-promotional listicle was cited, that brand was left out of the recommendation 69% of the time. The page can still earn retrieval and category association, but do not expect it to rank you.

Is this a Google AI Overviews problem or does it affect all engines?

It appears across engines, though the degree varies. The Wix Studio dataset covers ChatGPT, Google AI Mode and Perplexity; AirOps covered ChatGPT, Claude and Perplexity. Peec AI’s March 2026 listicle analysis found self-promotional citation rates differed by platform, reporting a lower 3.6% average on ChatGPT against 10.3% on Google AI Mode and 10.4% on Perplexity.

How long does it take to move a category prompt?

Longer than any vendor should promise, and it is not a function of your publishing cadence. Category movement generally requires new third-party documents to exist and then to be retrieved, which is outside your control. Measure it monthly, expect volatility, and do not treat a single good week as a trend.

Does any of this replace normal SEO?

No. The listicles and review pages that dominate category citations are themselves discovered through conventional search infrastructure, and being present in those documents still depends on being visible to the writers and editors who compile them.

What is the one thing a small vendor should do first?

Make the brand prompt bulletproof: identical facts across your site, directories and profiles, a clear front-loaded description of what you do and who you serve, and a real FAQ. It is the cheapest win available and it is a precondition for anyone else writing about you accurately.

Sources

• Wix Studio AI Search Lab (Tom Wells), “The content types most cited by LLMs”, 23 March 2026 — 75,000 AI answers, 1,056,727 citations across ChatGPT, Google AI Mode and Perplexity. wix.com/studio/ai-search-lab
• Lily Ray, “Why Calling Yourself the ‘Best’ Could Be Helping Your Competitors”, 17 June 2026 — 100 B2B “best [category] software” queries in Google AI Overviews, three checkpoints, 184 self-promoting listicles across 146 brands. lilyraynyc.substack.com
• AirOps (Oshen Davidson), “Third-Party Sources Drive 85% of Brand Discovery”, 17 October 2025 — 21,311 brand mentions across 500+ commercial-intent queries in ChatGPT, Claude and Perplexity. airops.com
• Peec AI (Tom Wells), “Self-promotional listicles analysis: Data from 232,000 citations”, 3 March 2026 (updated 27 March 2026) — 232,000 citations across 13,000 listicles, December 2025 to February 2026, six platforms. peec.ai

Every figure above was checked against the publishing organisation’s own page on 26 August 2026. Where a study did not measure something, we have said so rather than filled the gap.

Apply for partnership

Zian AI builds autonomous sales agents for phone, SMS, email and WhatsApp, and is currently in waitlist and partnership beta. If you want the conversations that AI answers send you handled properly, Apply For Partnership.