logo
Free tool · AI citation analyzer

What do AI answers actually cite?

We asked ChatGPT, Perplexity, Gemini and Claude real questions and sorted every source they cited by type. Ask your own question to see what they cite today.

Try:

4 AIs · 96 questions

0%

of the sources AI cited are blog content: listicles, articles and comparisons.

Blog content 74%
Everything else 26%

Based on 2,671 cited sources. Hover or tap a segment for details.

Every content type, ranked

Click a type to filter the sources. Violet types are blog content.

View the numbers as a table
Content typeShareSources
Listicles30%810
Blog articles & guides29%785
Comparisons & reviews14%380
Wikipedia & research9%230
Product & landing pages7%193
Docs & help centers4%119
Forums & social3%78
Videos1%35
Review platforms<1%24
Other<1%13
News<1%4

By AI assistant

Click an assistant to see only its citations.

  • 17%blog content
    Why so low? See the ChatGPT deep dive below
  • 85%blog content
  • 74%blog content
  • 94%blog content

By type of question

Blog content leads for buying questions, how-tos and comparisons alike.

  • "Best X" questions

    77%blog content

    Most cited: Listicles 73%

  • "How to" questions

    71%blog content

    Most cited: Blog articles & guides 55%

  • "X vs Y" questions

    79%blog content

    Most cited: Comparisons & reviews 75%

  • "What is" questions

    64%blog content

    Most cited: Blog articles & guides 63%

  • Advice & decisions

    80%blog content

    Most cited: Blog articles & guides 56%

  • "Alternatives to X" questions

    72%blog content

    Most cited: Listicles 71%

ChatGPT deep dive

Why ChatGPT cites so few blogs

We ran the same 48 questions through ChatGPT again and logged every page it opened, not only the ones it linked to. It reads widely, then cites the brands' own pages.

36

pages read per answer, 4 cited

485

Reddit pages read, 0 cited

85%

of its searches target one brand's own site (site:)

9/16

"best X" and "alternatives" questions where its first search already names the brands

What it read vs what it cited

  • Pages ChatGPT read

    1,732 pages across 48 answers

    15%blog content

  • Pages ChatGPT cited

    194 pages across 48 answers

    15%blog content

How it searches

  • Question: best CRM for small businesses

    • best CRM small business 2026 Zoho Bigin HubSpot Pipedrive pricing official
    • site:zoho.com/crm zohocrm pricing USD standard 14 free three users

  • Question: best email marketing tools

    • best email marketing software Mailchimp Brevo Kit Klaviyo ActiveCampaign MailerLite pricing official
    • site:kit.com pricing newsletter plan 10000 subscribers

  • Question: best noise cancelling headphones

    • best noise cancelling headphones 2026 Sony WH-1000XM6 Bose QuietComfort Ultra 2nd generation
    • site:sony.com WH-1000XM6 $449.99

  • Question: best VPN for privacy

    • site:mullvad.net audit 2025 2026 no logging cash account
    • site:mullvad.net pricing 5 euro month no email

Is it our prompt or the model?

Same 48 questions, different setups.

  • GPT-6.1 Sol

    Same setup as the benchmark

    15%blog content

  • GPT-6.1 Sol

    No instruction, just the question · 8 answers without a web search

    11%blog content

  • GPT-6 Luna

    Same setup, lighter model

    18%blog content

  • GPT-5.4

    Same setup, previous generation

    34%blog content

ChatGPT builds its shortlist from what it already knows, then checks each brand's pricing and docs pages. To show up there, your brand has to be known before the question is asked, and your own pages have to answer it plainly.

Examples of cited sources

Real pages the assistants linked to.

Benchmark run on October 20, 2026.

How we measured it

  • Real questions, four assistants

    We asked 96 everyday questions, 16 for each type: "best X", "how to", "X vs Y", "what is", advice and "alternatives to X". Each one went to ChatGPT, Perplexity, Gemini and Claude through their APIs with web search turned on.

  • Only the sources they actually cited

    We count the pages each assistant linked to in its answer, not every page it read. Duplicate links in one answer count once.

  • Every page classified by format

    Known platforms (Reddit, YouTube, G2, Wikipedia, docs sites) are classified by rule. Everything else is classified by Claude from the URL, title and snippet. Blog content means listicles, blog articles and comparisons.

  • A second look at ChatGPT

    For ChatGPT we re-ran the first 48 questions and logged every page its web search opened, not only the pages it linked to. We also ran them without our instruction, and on GPT-6 Luna and GPT-5.4, to check whether the result came from our setup or from the model.

  • Limits

    API answers can differ from the consumer apps, and results change from one day to the next. Run your own question to see today’s answer.

AI cites blog content. Outrank writes it for you.

Keyword research, listicles, guides and comparisons published to your site every day, so ChatGPT and Google have something of yours to cite.

Get Started for Free