What do AI engines actually cite when they answer a buying question?
AI engines cite listicles most, then brand-owned pages and paid publishers, and the mix shifts with the question and the engine. Otterly1's October 2026 study of 617,985 citations put listicles at 66.6% of classified citations. A separate academic study found every cited page was first retrieved by the engine's own search.
What do AI engines cite when a buyer asks for a recommendation?
By the New Map team. Published 16 October 2026. Updated 16 October 2026.
New Map is an AI visibility platform. It measures what the major AI engines answer about a brand, in every language, separates discovery from trust, drafts the fixes and places the brand where the engines look.
Otterly analysed 617,985 citations and published the results in October 2026. Of 519,140 classified citations, listicles were 66.6% and comparison articles 7.9% (Otterly, October 2026). Ranking prompts drew 73.2% listicles, while comparison prompts drew 62.3% comparison articles (Otterly, October 2026).
The next chart shows who owns the cited pages.
Brand-published listicles were 35.6% of all listicle citations (Otterly, October 2026). In that dataset, the idea that engines shun brand-written pages does not hold.
Intent changes the mix. For factual questions on pricing, features or company details, 84.4% of ChatGPT's citations across 20,000 US prompts included at least one official source (Ahrefs2, 3, October 2026). For comparison questions, the share ran from 63.4% to 46.4% (Ahrefs, October 2026).
These are single-vendor studies with their own prompts and labels, so each is one view. See also which sources AI engines cite most.
Does a page need to be retrieved before it can be cited?
In one academic study, yes. Every page ChatGPT cited had come back from the engine's own search, run on sub-questions the model rewrote for each answer (Chen et al.4, arXiv, October 2026). Retrieval rank explained 32% of within-site citation variation, and all page features combined explained 2.8% (Chen et al., arXiv, October 2026).
Jiang et al.5 reported that candidates retrieved more often across repeated runs were also cited more often (Jiang et al., arXiv, October 2026). Google Search and the OpenAI Responses API each recovered only about one quarter of the URLs ChatGPT Web cited (Jiang et al., arXiv, October 2026).
The same preprint found that two traits of a page predicted citation: how closely its title matched the question, and how closely a paragraph did. Adding FAQ blocks, structured data, summary boxes or faster loading made no difference inside a single site (Chen et al., arXiv, October 2026). From that, write pages that answer the rewritten sub-question directly, not only the headline keyword.
How much do the engines differ from each other?
A lot, on format, on source count and on platform. In New Map's measurement of 16 October 2026 (how New Map measures), 30 discovery questions went to each of five engines. Perplexity cited 18 sources per answer, Grok 9.1, Gemini 7.3, Claude 3.6 and ChatGPT 3.2.
In the same measurement, reddit.com was cited in 7 of 150 answers (4.7%) and youtube.com in 6 of 150 (4%), from 30 questions on 5 engines (New Map's measurement of 16 October 2026). Otterly found that ChatGPT's citation share for reddit.com dropped from 4.01% to 0.44% on 8 August 2026 (Otterly, October 2026).
| Finding | Source and date |
|---|---|
| Claude cites listicles 84.6% of the time on buying questions, ChatGPT 52.8% | Otterly, October 2026 |
| Forums and video are 25% of Google AI Overviews and AI Mode citations, under 3% elsewhere | Otterly, October 2026 |
| ChatGPT triggered a web search in 14.4% of 66.6 million responses | Ahrefs, August to September 2026 |
| Sonnet 5.5 runs 1.17 query fan-outs per searched answer, against 1.83 for Sonnet 5 | Profound6, 7, October 2026 |
Models within one vendor change too: Sonnet 5.5's top 10% of cited pages have a median age of 107 days against 164 for Sonnet 5 (Profound, October 2026, 28,440 runs).
Can I see where competitors are cited and I am not?
Platforms are starting to show this. Bing's Topics Portfolio groups grounding queries into topic clusters and shows topic size, sites cited and queries (Microsoft Bing8, October 2026). Its Top Opportunity Queries list queries where other sites appear but yours does not.
Bing says all figures are approximate, directional and shown as ranges, that topic labels are AI-generated and imperfect, and that smaller sites may not see the feature (Microsoft Bing, October 2026). That is a platform statement about its own tool, not proof of what lifts citations.
Attribution is shaky as well. Tagged source links in ChatGPT answers with citations fell from 98% on 8 October to 66% by 14 October (Profound, October 2026, sample size not stated).
How should a newer brand get cited?
Start with what engines look up when a buyer checks a brand by name. Of ChatGPT's site: searches, 75.9% targeted brand sites (Ahrefs, August to September 2026). Fan-outs containing the word official rose from 10.1% in August to 13.1% in September (Ahrefs, August to September 2026).
AI Overviews appeared on 82.91% of US desktop branded SERPs by 29 September 2026, and Wikipedia was 71.9% of their external citations (Ahrefs, October 2026).
A citation is not a recommendation. Ahrefs found that 43% of answers citing a brand's event page never named the brand (Ahrefs, July 2026, 9,886 answers). Scrunch found a brand's own number one listicle lifted its recommendation rate from about 4% to 7% when cited (Scrunch, September 2026, placebo-tested).
Third-party lists seem to carry more. Rank 1 in a frequently cited third-party listicle went with a rise of 13 to 17 points in mention probability (Peec AI, May 2026, observational). Our read: pair each owned page with a third-party companion. Pursue only mentions you can authentically earn.
What to do
- Write pages whose titles and opening paragraphs answer the sub-questions engines rewrite buyer questions into, with a specific price, a current date and evidence.
- Keep official pages for pricing, features and company details accurate, since 84.4% of ChatGPT's citations on factual questions included an official source (Ahrefs, October 2026).
- Track citations per engine and per question type, repeated over time. See plans and pricing for what New Map offers.
Frequently asked
Which content format do AI engines cite most?
Listicles led in Otterly's October 2026 analysis, at 66.6% of 519,140 classified citations, with comparison articles at 7.9% (Otterly, October 2026). The mix differs by question type and engine, so check the formats cited for your own buying questions.
Do brands get cited from their own sites?
Often. Brand-owned pages were 38.9% of 450,473 citations (Otterly, October 2026). For factual questions, 84.4% of ChatGPT's citations included at least one official source (Ahrefs, October 2026). Comparison questions showed lower shares, from 63.4% to 46.4% (Ahrefs, October 2026).
Does schema markup or an FAQ block increase citations?
Not within sites, in one preprint. Chen et al. found no within-site effect from FAQ blocks, structured data or summary boxes, while retrieval rank explained 32% of within-site citation variation (Chen et al., arXiv, October 2026). Treat markup as hygiene, not a lever.
Sources
- Otterly, What AI Engines Actually Cite: 617,985 Citations Analyzed, otterly.ai, 14 October 2026
- Ahrefs, How ChatGPT Does Query Fan-Out (15 Million Queries Studied), ahrefs.com, 12 October 2026
- Ahrefs, AI Overviews Currently Appear for 83% of Branded Searches, ahrefs.com, 2 October 2026
- Chen et al. (arXiv), What Drives AI Citations? Content, Retrieval and Model Knowledge in Generative Engine Optimization (GEO), arxiv.org, 10 October 2026
- Jiang et al. (arXiv), Understanding ChatGPT Citations: Evidence from Production Web Search Across Model Generations, arxiv.org, 11 October 2026
- Profound, ChatGPT’s Missing UTM Tags: What Our Research Shows, tryprofound.com, 14 October 2026
- Profound, What’s changed with Sonnet 5.5, tryprofound.com, 13 October 2026
- Microsoft Bing, Introducing Topics Portfolio (Beta) in AI Performance, blogs.bing.com, 13 October 2026
See where you stand
Book a call and find out how visible your brand is across the AI engines your customers already ask.
Book a callFor agencies: See the agency package