The Format-Intent Grid: which pages AI engines actually cite

The most common content question I get is some version of: 'Should I publish listicles or educational content for AI search visibility?' It's the wrong question. Asked that way, the answer is always 'it depends' — which is true and useless.

The right question has three axes, not one. AI citations follow a grid: five page formats (listicle, comparison, how-to, research essay, glossary) × four query intents (informational, commercial, transactional, navigational) × three platforms (ChatGPT, Google AI Overviews, Perplexity). That's 60 cells. Most content teams optimize on a 1×1 — pick one format, point it at one intent, hope it works on every platform. The published research says it doesn't.

This article does three things. It defines the five formats precisely (because most studies collapse 'comparison page' into 'listicle' and lose the signal). It maps each format-intent-platform combination to the strongest available data — Wix Studio's 75K-answer dataset, Allsopp's 26,283-URL ChatGPT analysis, Profound's 680M citations, Seer's 5.47M-query AIO study, Princeton's GEO paper, and others. And it surfaces the format hiding inside the grid that almost no B2B brand is running: glossary pages.

Most published studies treat content format as 2–3 buckets: 'listicle', 'article', 'product page'. That's why the answers people get are vague. To map the grid you need a finer taxonomy. These five formats cover almost every page that gets cited by an AI engine.

AI engines do not weigh formats uniformly across queries. The intent of the query determines which format wins the citation. The classification below is the standard search-intent taxonomy adapted for generative engines — and the column you sit in matters more than the format you publish.

Five primary studies anchor the grid. Each one isolates a different slice — read them together, not in isolation.

Wix Studio AI Search Lab analysed 75,000 AI answers and over 1M citations across ChatGPT, Google AI Mode, and Perplexity. Their headline split: listicles get 21.9% of citations across all queries, articles 16.7%, product pages 13.7%. Inside that average, intent reverses everything. For informational queries, articles are cited 2.7× more than other formats and reach 45.5% share. For commercial queries, listicles jump to 40%. Format choice is intent-conditional — that's the empirical foundation of this article.

Glen Allsopp's 750-prompt, 26,283-URL ChatGPT study gave the strongest single number on listicles: 'Best X' lists were 43.8% of all cited page types and 51.4% in the software vertical. Allsopp's prompt set was top-of-funnel commercial, which is why this number is so high — not because ChatGPT loves all listicles, but because his prompts triggered the cell where listicles win.

Profound's 680M-citation analysis exposed the platform variance that makes the third axis necessary. Wikipedia is 7.8% of all ChatGPT citations and 47.9% of its top-10 sources. On Perplexity, Reddit is 46.7% of top-10 sources. ChatGPT cites Wikipedia roughly 80× more than Google AI Mode does. The same content does not perform the same way on different engines — the grid has to be 3-dimensional.

Seer Interactive tracked 5.47M queries and reported that AI Overviews triggers on 95.4% of comparison-format queries — the single highest trigger rate of any query class. Their related dataset showed ChatGPT's listicle citations declined 30% month-over-month December 2025 to January 2026, while AI Overviews' listicle citations stayed flat or rose. That's the cell-level resolution this article is about: 'listicle' is not one answer.

Princeton's GEO paper (KDD 2024, Aggarwal et al.) provides the format-agnostic baseline. Adding statistics improved citation visibility by 32.8%, citing sources by 27.7%, and adding quotations by 42.6%. None of those depend on format. Whichever cell you optimize for, these three modifications stack on top.

The matrix below maps each format-intent-platform combination to a 1–5 fit rating, with a one-line note and a confidence flag. Five means the format wins that cell; one means it loses outright. High confidence means at least one named primary study supports the rating directly; medium means it's inferred from related data; low means it's reasoned from first principles or single-source.

Read the grid by row to find your format's best cells, or by column to find which format owns a query intent on a given platform. The largest concentration of fives sits along three diagonals: listicles → commercial, comparison pages → transactional, glossary pages → informational + navigational.

Three patterns are worth calling out. First, the navigational column is mostly twos — every format loses navigational queries to first-party brand pages, which is the correct outcome. Second, comparison pages are the only format with a complete diagonal of fives across all three platforms for transactional intent — they are the cleanest bottom-of-funnel format and the most under-built one. Third, glossary pages quietly score four-and-five across informational and navigational on every platform — and almost no B2B brand is running them.

The most under-deployed AI citation format is the glossary or definition page — the one that owns the 'what is X?' query for a category-defining term. It's the playbook Wikipedia has run for two decades and Investopedia has run for one. Almost no B2B SaaS brand runs it.

The reason it works for AI engines is structural. Generative models need to disambiguate entities before they can answer questions about them. The page that defines the entity authoritatively becomes the page they reach for first — and the brand attached to that page becomes a default reference in answers about the category. That's not a clever optimization. It's how retrieval works.

Audit your existing content against the grid before you write a new piece. Most teams have a stack of pages aimed at the wrong cell — that's the cheapest fix.