'%20fill='%238B75F3'%20fill-opacity='0.5'/%3e%3crect%20x='4.70581'%20y='10.6467'%20width='15.0568'%20height='15.0568'%20rx='3'%20transform='rotate(-45%204.70581%2010.6467)'%20fill='%238B75F3'%20fill-opacity='0.5'/%3e%3c/svg%3e)
Self-Promotional Content Data Study
Method
From brands to labelled pages
We turned a list of software brands into the kind of questions buyers ask, put each question to six platforms, and collected every page they pointed to. Then we read those pages and labelled each one: who wrote it and what kind of page it is.
Brands per category
+ 189 more
Questions per type
ex. “What are the best alternatives to Samsara for Fleet Management?”
ex. “How do Fleetio and Samsara compare for Fleet Management?”
ex. “What are the best tools for Fleet Management?”
Every question, asked on 6 platforms
ex. “What are the best alternatives to Guardz for Cloud Security Posture Management?”
- Perplexity10 pages ★
- Google search9 pages ★
- ChatGPT8 pages
- AI Overviews7 pages ★
- Claude4 pages ★
- Geminino pages
★ accuknox.com/blog/cspm-tools · self-promotional, cited by 4 of the 6
From answers to pages
Source: citations.csv, answers.csv · Self-Promotional Content Data StudyFrom pages to labels
Source: page-types.csv · Self-Promotional Content Data StudyLabelling each page
Publisher: Software vendor
Page type: Listicle
Sells in this category: Yes
Own tool on the list: First
+ 8 more AI labels, like tone
Results
How many questions get an answer that searched and cited the web
Every chart below counts cited pages, so it only sees the answers that searched and cited at least one. Here is how many do. ChatGPT was made to search on every question when the data was collected, so its 100% is a setting, not a habit. Claude cites nothing for 21–27% of the questions of every kind; Gemini for 45% of “Alternatives to X” but 1% of “X vs Y”; AI Overviews for 40% of “X vs Y”, a results page without an Overview counting as citing nothing. Those answers are missing from the charts below. “How the answers are written” is the exception: it reads every answer, cited or not.
- Answer searched and cited the web
- Answer cites nothing
Σ 11,525 questions
Which kinds of pages get cited
For each kind of question, every bar splits one platform’s citations by type of page.
- Self-promo listicle
- Self-promo comparison
- Vendor's own site
- Vendor's other pages
- Listicle, not self-promo
- Comparison, not self-promo
- Review site
- Video
- Other
Σ 436,290 citations
How many cited listicles are self-promotional
The listicles from the charts above on their own: for each platform, the share of its listicle citations that are self-promotional.
- Self-promo listicle
- Listicle, not self-promo
Σ 278,636 listicle citations
What self-promotional listicles do
A self-promotional listicle is a software vendor’s list of the best tools in its own category that includes its own product. The platforms cited 39,214 of them, 137,105 times in all, and we checked every one for eight patterns.
- Yes
- No
- Each square is 1% of citations
Σ 118,196 citations of 36,037 pages
Deep diveWhere vendors’ own lists dominate, and whose they are
In which software categories
For each software category: of the listicles the five AIs cite when asked about it, the share that are a vendor’s own list (43% over all categories). The 15 highest and the 15 lowest of the 177 categories with at least 300 listicle citations.
Most self-promotional
Least self-promotional
Σ 245,332 listicle citations in 177 categories
The vendors whose own lists the AIs cite most
Citations of each vendor’s self-promotional listicles, the five AIs together; after them, how many of its lists were cited and how often those citations go to a list that puts the vendor first. The top 20 of 9,091 vendors hold 6% of these citations.
Σ 106,745 citations of 9,091 vendors' own lists
How the answers are written
Each cell is the share of an AI’s answers showing the pattern. The first three are read by the answer classifier, the other four straight from the text. AI Overviews text was captured without its formatting, so its tables cannot be seen.
| ChatGPT | AI Overviews | Gemini | Perplexity | Claude | |
|---|---|---|---|---|---|
| Match tools to needs | 100% | 94% | 99% | 99% | 58% |
| Pick one winner | 84% | 32% | 12% | 54% | 26% |
| Ask the user back | 82% | 95% | 91% | 88% | 20% |
| Have a table | 99% | n/a | 58% | 55% | 0% |
| Mention pricing | 62% | 46% | 41% | 28% | 32% |
| Have a “best for” table | 91% | n/a | 44% | 37% | 0% |
| List pros and cons | 8% | 3% | 1% | <1% | <1% |
Σ 57,365 answers
How many pages an answer cites
For each platform, how many pages its answers cite, counting the answers that cite at least one. Google search is left out: its results page always lists about ten.
– – Average pages cited per answer
Σ 49,626 answers
How often an answer cites the same page
The other charts count a page once per answer; here, how many times the answer cites it. Each AI is counted on its own way of citing: Perplexity by its repeated [n] markers, ChatGPT by its links in the answer text, Gemini within one list of sources, Claude and AI Overviews by their citation links. Google search lists a page once, so it is left out. The last point adds up every page cited 10 times or more.
Median times an answer cites a page
Σ 338,261 citations
Counting notes
Do the AIs cite the same pages as Google, and as each other?
For the same question, each AI’s citations split three ways: also in Google’s top 10, not in Google but cited by another AI, or cited by this AI alone.
- Also in Google's top 10
- Not in Google, but cited by another AI
- Cited by this AI alone
Σ 338,261 citations
The most cited domains
Each cell is the share of a platform’s answers that cite the domain at least once, out of its answers that cite any page. The rows are each platform’s five most-cited domains, topped up with the next most cited overall so the tabs have the same number of rows.
| Google search | AI Overviews | Perplexity | Claude | ChatGPT | Gemini | |
|---|---|---|---|---|---|---|
| g2.com | 34% | 42% | 56% | 36% | 32% | 25% |
| reddit.com | 71% | 38% | 14% | 0% | 24% | 2% |
| gartner.com | 30% | 20% | 6% | 5% | 11% | 7% |
| youtube.com | 35% | 35% | <1% | <1% | 1% | 1% |
| capterra.com | 10% | 5% | 7% | 8% | 3% | 2% |
| sourceforge.net | 5% | 3% | 6% | 18% | <1% | 2% |
| cbinsights.com | 9% | 3% | 7% | 6% | 2% | 0% |
| softwareworld.co | 2% | 3% | 4% | 14% | <1% | 1% |
| alternativeto.net | 1% | 1% | 6% | 13% | 2% | <1% |
| checkthat.ai | 4% | 6% | 7% | 3% | <1% | 2% |
| medium.com | 6% | 8% | 2% | 1% | 0% | 2% |
| slashdot.org | 2% | 1% | 2% | 10% | <1% | 2% |
| facebook.com | 13% | 4% | <1% | 0% | 0% | 0% |
| techradar.com | <1% | <1% | 2% | <1% | 13% | <1% |
| microsoft.com | 1% | <1% | <1% | <1% | 5% | <1% |
Σ 61,151 answers
Deep diveHow big are the sites that get cited?
Each cited site’s size, read in October 2026 as DataForSEO’s domain rank: 0 to 100, from the sites linking to it, like Ahrefs’ Domain Rating. A hosted page counts as its host: a Reddit thread carries Reddit’s rank.
How often sites of each domain rank appear in answers
Each site counted once per answer that cites it, as in the most cited domains table, for all 43,034 cited sites. The bumps are single big domains, such as YouTube, LinkedIn and Reddit.
– – Median domain rank
Σ 394,702 sites in answers
How often each kind of page gets cited, by domain rank
Every citation by kind of page, all platforms and kinds of question together. Reddit is left out: it is a single site. Video pages are 97% YouTube, rank 100, so their peak runs off the top.
– – Median domain rank
Σ 660,818 citations
How many products ChatGPT’s searches name
The web searches ChatGPT runs before it answers, by how many products each names: all of them together, then by their place in the answer’s list of searches. The last bar adds up every search from the 5th on. The “X vs Y” searches come from an October re-run; the other two are from August, and ChatGPT’s searches shifted in between, so compare that tab with care (Disclaimers shows how much).
- Exactly one product
- Two or more products
- No product (a generic search)
Σ 52,293 searches · August 2026
The words ChatGPT adds to its searches
Words not already in the question, found in searches across at least half of the 196 categories, so product and category names drop out. The bigger the word, the more searches it is in.
Σ 52,293 searches · August 2026
Where ChatGPT points its “site:” searches
The website a “site:” search is limited to, by kind.
Σ 17,277 searches limited to one website · August 2026
What Perplexity cites out of what it finds
Perplexity is the one platform whose data shows every page its searches found, not only the ones it cites: a typical answer finds 20 pages and cites 9 of them with an [n] marker. For each kind of page, the bar is the pages found, the dark part the ones cited, and the number the share cited.
- Cited
- Found, not cited
Σ 227,584 pages found, 101,643 cited
Deep diveWhat gets a listicle cited once Perplexity finds it
The listicles among the pages Perplexity’s searches found, and how often its answer then cites them.
By place in the search results
The share of found listicles the answer cites, by their place in Perplexity’s search results.
- Self-promo listicle
- Listicle, not self-promo
Σ 192,754 listicles found
With and without each trait
The same share for listicles with and without each trait, compared at the same place in the search results.
- Has it
- Doesn’t