Earn the listicles
40% of citations sit with independent roundups. Getting into the lists that already rank for your category is the single highest-leverage earned move.
The Prefer Index · Edition 01 · July 2026
Who actually gets cited in AI search.
We ran the same 47 buyer questions across four major answer engines and recorded every source behind every answer. Edition 1: 1,237 citations, 713 domains, and a clear picture of who AI trusts. Every figure recomputes from the committed dataset.
Tested on
ChatGPT
Perplexity
Gemini
Claude
Key findings
Five things the data says plainly.
Who gets cited most in AI search results?
Independent listicles take 40% of all citations in AI answers, with vendors' own sites second at 34%. Across 47 buyer questions and four engines, the four engines shared only 5 of the 713 domains they collectively cited, and ChatGPT answered 29 of 47 questions from memory without searching at all. For brands, the practical finding is that AI visibility is won on a mix of other people's listicles and your own clearly stated pages, and it is won engine by engine.
- 01Listicles take 40% of citations, vendor sites 34%40% vs 34%
The two biggest slices are independent roundups and the brands' own pages. Both are earnable; only one is under your control.
- 02ChatGPT often does not search at all29 of 47 from memory
ChatGPT answered 29 of 47 questions from memory, with no live search and no citations. What a model remembers about you matters as much as what it can retrieve.
- 03The engines barely agree5 of 713 shared
Of 713 domains cited at least once, 539 (76%) were used by a single engine, and only 5 were cited by all four.
- 04Citation is less concentrated than you think15% from the top 10
The top 10 domains carried 15% of citations and the top 50 carried 31%. The tail is long, which means room to enter it.
- 05Head-to-head questions distrust vendors18% vs 37% vendor share
On X-vs-Y questions, vendor pages took 18% of citations versus 37% on best-of questions, and 41% of vs answers cited no vendor page at all.
The dataset
What we ran, and how much of it.
47buyer questions
Across 8 categories, from category discovery to head-to-head comparisons
4answer engines
ChatGPT, Perplexity, Gemini and Claude, web search on
188answers recorded
Every question on every engine, captured with its sources
1,237citations logged
Each traced to a domain and a classified source type
713distinct domains
Every site any engine used at least once
772brands named
Every product or company recommended in any answer
Run 10 July 2026 via API from a clean US English context: no personalisation, no logged-in history, one pass per question per engine.
01 · Where citations go
Forty per cent of AI's sources are listicles.
Every citation was classified by what kind of site it came from. Independent roundups lead, but the second slice surprises most teams: engines cite the vendors' own pages heavily, when those pages state facts plainly enough to quote.
Share of all citations
Citation share by source type
1,237 citations across four engines, classified by site type.
40%34%15%
- Independent listicles
- 40%
- Vendors' own sites
- 34%
- Media & SaaS blogs
- 15%
- Reddit & forums
- 4%
- Review sites (G2 etc.)
- 3%
- YouTube
- 3%
Per engine
The same split, engine by engine
ChatGPT leans hardest on review sites among the four. Perplexity is the only one that reaches for community and video at scale. Gemini and Claude read almost identically: listicles first, vendors second.
ListiclesVendorMedia/SaaSCommunityVideoReviews
ChatGPT
44231812
Perplexity
283617
Gemini
473413
Claude
473615
02 · The concentration curve
713 domains cited. The head is smaller than you think.
AI citation has a head, but it is flatter than search folklore suggests. The top ten domains carry 15% of citations, and the long tail is genuinely long, which is the opportunity.
Cumulative citation share
How citation share accumulates by domain rank
Domains ranked by total citations, cumulative share of all 1,237.
Top 10 domains
15%
Top 25
23%
Top 50
31%
Top 100
42%
Top 250
63%
All 713
100%
15%of citations from the top 1031%from the top 5058%carried outside the top 10076%of domains cited by one engine only
03 · Engine personalities
The engines have different habits.
Same questions, same day, four very different reading styles. How many sources each engine pulls, how widely it ranges, and how often it searches at all.
EngineCitations per answerDistinct domains usedVendor-cited shareBrands named per answerAnswers with live sources
Answered 29 of 47 questions from memory with no citations at all. When it does search, it cites few sources from the narrowest range, so being in its set matters most.
Searched on every question and cited the most per answer. The only engine reaching for Reddit and YouTube at scale.
Cited the most sources in total across the widest domain range, and leans hardest on third-party listicles.
Searched on nearly every question and attaches citations to specific claims, split between listicles and vendors' own pages.
04 · The agreement problem
Four engines, four reading lists.
We checked how many domains each engine shared with the others. The overlap is far smaller than most teams assume, which is why a single-engine view of your visibility misleads.
Domain overlap across engines
How many domains were cited by how many engines
Of 713 domains cited at least once, by number of engines citing them.
5Cited by all 4 engines
52Cited by 3
117Cited by 2
539Cited by 1 only
5domains cited by all four engines76%of domains cited by one engine only1.3average engines per domain4reading lists to earn separately
05 · Most-cited domains
The sites AI reads before it answers you.
The twenty most-cited domains across the run, with which engines used them. Sorted by total citations.
#DomainTypeEnginesCitationsShare
1
reddit.comCommunity3433.5%
2
youtube.comVideo2322.6%
3
zapier.comMedia / SaaS blog4312.5%
4
g2.comReview platform4191.5%
5
salesforce.comVendor site3131.1%
6
techradar.comMedia / SaaS blog3121%
7
monday.comVendor site3110.9%
8
forbes.comMedia / SaaS blog4100.8%
9
emailvendorselection.comListicle390.7%
10
guideflow.comListicle390.7%
11
hubspot.comMedia / SaaS blog380.6%
12
peoplemanagingpeople.comListicle380.6%
13
technologyadvice.comListicle380.6%
14
emailtooltester.comListicle370.6%
15
marketermilk.comListicle270.6%
16
brex.comVendor site360.5%
17
gusto.comVendor site360.5%
18
klaviyo.comVendor site360.5%
19
nightwatch.ioVendor site360.5%
20
thedigitalprojectmanager.comListicle460.5%
Top 20 of 713 cited domains. Reddit's 43 citations came almost entirely from Perplexity; no single domain reached 4% of the total.
06 · By category
Your category has its own rules.
The overall split hides real variation. In HR, finance and support, engines lean on the vendors' own pages more than on listicles; in project management and SEO tools, the listicle layer decides. Samples per category are small (about 6 questions each), so read these as directional.
Source mix by category
Where citations come from, per category
Eight categories, each with its own retrieval habits. Percentages are of citations within that category.
ListiclesVendorMedia/SaaSReviewsCommunity
SEO & AI searchListicle-led. The roundup layer decides who gets named, which is why conquest listicles work in this category.
473312
Marketing toolsListicle-led with a real review-site presence. Email tooltester-style comparison sites carry weight.
443112
CRM & salesMedia and SaaS blogs punch above average. Zapier's comparison content shows up across engines.
413019
Project managementThe most listicle-led category in the set. Independent roundups decide the shortlist.
492716
HR & recruitingVendor-led. Engines quote the HR platforms' own pages more than anyone's roundups.
324514
Accounting & financeVendor-led. Clear pricing and feature pages on the vendors' own domains get quoted.
334114
Customer supportVendor-led with the highest review-site share. Comparison shoppers get review-platform answers.
333915
EcommerceBalanced, with the strongest media/SaaS blog presence in the set.
402822
07 · The Index
The brands AI recommends most.
How often each brand was named across every answer in the run. This is a measure of recommendation, not market share, and no brand dominated: the most-named brand appeared in 22 of 188 answers.
Mentions across 188 answers
Most-recommended brands across all categories
Of 772 brands named across 188 answers, top 14 by mentions. All 14 were named by all four engines.
772distinct brands named8.8brands named per answer22answers naming the top brand4 of 4engines naming each of the top 14
08 · Comparison questions
X vs Y questions change the sourcing.
We isolated the head-to-head comparison questions in the set. They are the highest-intent questions a buyer asks, and engines source them differently: less trust in vendors, more in third-party comparisons.
18%of citations on vs questions went to vendor pages
Half the 37% vendor share on best-of questions. Engines discount the vendors' own words when two of them are being compared.
54%of vs citations went to independent listicles
The comparison layer belongs to third parties, which is why neutral referee pages earn citations.
41%of vs answers cited no vendor page at all
The engine went looking for an impartial account and built the answer entirely from third parties.
8head-to-head prompts in the set
A small slice, so treat these splits as directional. The direction is consistent across engines.
The vendor pages that did get cited on comparison questions stated facts plainly: pricing, feature scope, honest fit. Pages that only listed their own wins were not the ones quoted.
Methodology
How we ran it.
- 01The question set
47 buyer questions across eight software categories, phrased the way buyers actually ask them, from category discovery ('best X for Y') to head-to-head comparisons ('X vs Y').
- 02The run
Every question was submitted to ChatGPT, Perplexity, Gemini and Claude via API on 10 July 2026, web search enabled, from a clean US English context with no personalisation and no conversation history.
- 03Capture
All 188 answers were recorded in full with every cited URL. Citations were counted per appearance, traced to a domain, and each domain classified by source type.
- 04Classification
Each cited domain was classified into one of seven source types (listicle, vendor site, media or SaaS blog, community, video, review aggregator, other). The classified dataset is committed and every figure on this page recomputes from it.
- 05Brand extraction
Product and company names were extracted from every answer's recommendation text, 772 distinct brands in total.
- 06What we did not do
No paid placements, no logged-in accounts, no regional variation, and one pass per question per engine. Absolute figures will differ on a re-run; the shape is the finding.
Cite this research
Use it, quote it, link back to it.
The Prefer Index is free to cite with attribution. Reproduce any figure with a link back to this page.
Suggested citation
The Prefer Index, Edition 01. Prefer, July 2026. https://tryprefer.com/data/benchmark/
The statistics pageThe headline stats from this study in one quotable, sourced page.See the statistics ↓The narrative studyThe full write-up with figures: who gets cited in AI search and why.Read the study ↓Industry cheat sheetsThe per-industry companion: where citations go in your category.Open the cheat sheets ↓
People also ask
The questions readers ask next, taken from what assistants cluster with this one.
In our run, reddit.com led with 43 citations (3.5% of the total), then youtube.com (32) and zapier.com (31). Only five domains were cited by all four engines: zapier.com, g2.com, forbes.com, thedigitalprojectmanager.com and getaleph.com.
Questions
Asked plainly.
What is The Prefer Index?
A benchmark of how AI answer engines cite sources and recommend brands. Edition 1 ran 47 buyer questions across ChatGPT, Perplexity, Gemini and Claude with web search on, recorded all 188 answers, and classified all 1,237 citations by source type.
Why does source type matter more than ranking?
Because AI answers are assembled from a handful of retrieved sources rather than ranked from an index. Which kind of site an engine reaches for determines who gets named, so understanding the mix tells you where to invest.
Is the data available?
The classified dataset is committed alongside our published methodology, and every figure on this page recomputes from it. The headline numbers are collected on the statistics page for quoting.
Can I cite this research?
Please do. Cite it as The Prefer Index, Edition 01, Prefer, July 2026, and link to this page.
Why do the numbers differ from other studies?
Question set, engine mix and classification method vary between studies, and answer engines are non-deterministic. Absolute percentages are approximate; direction and relative ordering are the durable findings.
How is this different from a keyword ranking study?
A ranking study measures positions in a list of links. This measures which sources an engine chose to build an answer from, and which brands it named as a result.
Will you add more engines?
Google AI Overviews is measured separately in our Reddit citation study, and Citation Watch adds engines as editions accrue. The core four stay constant so the series remains comparable.
Can I get this run for my own category?
Yes, that is what the product does. A Prefer workspace runs your own prompt set continuously rather than once, sliced by your competitors, markets and buyer personas.
What it means
You win AI search in two places, and only one is your website.
Forty per cent of citations sit with independent listicles you do not control, and 34% with pages you do. The work is earning the first and sharpening the second, engine by engine.
Make your own pages quotable
34% of citations went to vendors' own sites. Plain stated pricing, feature scope and comparison facts get quoted; marketing language does not.
Write the honest comparison
Vendor share halves on X-vs-Y questions. The vendor pages that still get cited are the ones that state facts a referee could quote.
Win each engine separately
76% of cited domains were used by one engine only. ChatGPT's memory, Perplexity's Reddit habit and Gemini's wide net each need their own play.
Get your free AI visibility report
in about 10 minutes.
See how answer engines describe your brand today, and where the openings are to outpace the competition.