Do ChatGPT, Claude and Perplexity cite the same kinds of pages? (2026)

In our Nov 2026 checks, 71% of Claude's cited links and 50% of Perplexity's were lists or comparisons, against 10% of ChatGPT's. Not equal tests.

Illustration: three bridges of different shapes cross one river toward a rising sun on the horizon.
Key points 5 sections · 5 min read
  1. What did we check?
  2. What kinds of pages did each assistant cite?
  3. Whose pages did ChatGPT link?
  4. What does this mean for a business?
  5. What can’t this tell you?

No. In our November 2026 checks, 71% of Claude’s cited links and 50% of Perplexity’s were lists or comparison pages, against 10% of ChatGPT’s. The assistants weren’t tested on equal terms (see below), and two of them have small samples, so treat this as a pointer from one sample, not a rule.

What did we check?

We asked 18 buyer questions: 10 about choosing an AI visibility tool and 8 about hiring a service for AI search. ChatGPT and Claude answered each question once, on November 4, 2026, 18 answers each, with web search capped at 5 results per search and a low search-context setting. Perplexity answered each question 4 times on November 4 and 4 times on November 12, 144 answers in all, with its own built-in search. So the differences can come from these settings and the repeat runs, not only from the assistant, and they aren’t stable preferences.

We asked through each assistant’s API. Every question also carried a short instruction we wrote. It described the asker as the owner or marketing lead of a small or mid-sized US business who wants to be recommended by AI assistants, and asked the assistant to name specific companies, tools or services where relevant, which likely raises how often they are named. Gemini gave domain names only, so it is left out.

Diagram: one question card on the left, with three arrows to three differently shaped pages on the right: stacked rows, a single plain page and a page of mixed blocks.
Conceptual diagram: one question, and different mixes of cited pages in this test.

We sorted each cited link with a rule. In summary: a link counts as a list or comparison page when its web address or title says “best”, “top-rated”, “top” with a number, “alternatives”, “vs”, “compared” or “ranked”, or has a number followed by tools, agencies, platforms or software (plus a few similar patterns). It doesn’t count when the address is a product, pricing, plan, feature, FAQ, help, docs or free-tool page, unless the address or title also says “best”, “alternatives”, “vs”, “compared” or “top” with a number. Review sites, directories and social sites are kept out of this group. We then reviewed 20 random links per assistant by hand, from their web addresses and titles only, not the full pages. Our reading matched the rule for 56 of 60: the rule counted 1 explainer guide as a comparison page and missed 3 lists, so treat each share as approximate. A looser keyword rule, used on our statistics page, gives 20% for ChatGPT, 74% for Claude and 54% for Perplexity. We counted a link once per answer, so a page cited in several answers counts each time. These are citations, not unique pages.

What kinds of pages did each assistant cite?

Lists and comparison pages, share of cited links

ChatGPT and Claude: Nov 4, 2026. Perplexity: Nov 4 and Nov 12, 2026.

Claude (API)
71% (121 of 171)
Perplexity (API)
50% (1,597 of 3,169)
ChatGPT (API)
10% (12 of 118)
Share of each assistant's cited links that were lists or comparison pages, sorted with the hand-checked rule above. Method: what we checked. Not like for like: the search settings and the number of runs differed.
AssistantAnswersDatesLinksLinks per answerLists and comparisonsHomepages
ChatGPT (API)18Nov 41186.612 (10%)13 (11%)
Claude (API)18Nov 41719.5121 (71%)1 (1%)
Perplexity (API)144Nov 4 and Nov 123,16922.01,597 (50%)24 (1%)
Cited links with a full web address, counted once per answer, Nov 4 and Nov 12, 2026. Links per answer depend partly on search settings: ChatGPT and Claude were capped at 5 results per search.

Three things stand out. Claude’s links were mostly lists and comparisons. Perplexity’s were about half. ChatGPT’s were about 1 in 10, and 11% of them were homepages, against 1% for Perplexity and Claude. Perplexity also gave far more links per answer, but it searched without the 5-result cap. The different settings and repeat runs mean we cannot attribute that gap to assistant identity alone.

Across the 10 questions about choosing an AI visibility tool, 30 of ChatGPT’s 62 links (48%) went to the websites of companies on a list we keep of 25 AI visibility tool and agency companies (27 web addresses, ours included), against 15 of 96 for Claude (16%) and 252 of 1,736 for Perplexity (15%). The list doesn’t cover every company website, so these counts are a floor. This is a small sample, 10 ChatGPT answers, and company websites were cited in our earlier test too.

What does this mean for a business?

  • Check each assistant separately. They didn’t cite the same kinds of pages, just as they don’t name the same businesses.
  • Comparison pages made up more of some assistants’ citations than others in this test. Where they do, the pages AI already cites are where to look.
  • Company websites were cited too. ChatGPT linked them often in this sample. Plain, accurate facts about what you do, for whom and at what price are worth having on your own site.

These counts show which pages were cited, not that changing them gets a business named. A citation shows that a page was linked. It doesn’t show that the page decided who was named.

What can’t this tell you?

  • Not equal tests. The search settings and the number of runs differed, so differences aren’t only the assistant.
  • Small samples. ChatGPT and Claude answered each question once. Another run could look different.
  • Citations, not unique pages. A page cited in several answers counts each time.
  • API, not the apps. The ChatGPT and Perplexity apps can answer differently, and answers might change from one ask to the next.
  • One topic. Every question was about AI visibility tools and services. Other markets can cite other pages.
  • Sorting by rule. A hand review of addresses and titles matched the rule for 56 of 60 links (1 wrongly counted as a list, 3 lists missed), so some pages will be in the wrong group.
  • Our own count. It is two dated checks from our saved answers, not a study of the market.

You can see which sites an assistant cites for your own buyers’ questions with our free AI visibility checker.

See how it works and pricing.

Keep reading

All posts and guides
  • Which websites does AI cite when buyers ask?

    Post 6 min read

    Which websites does AI cite when buyers ask?

    ChatGPT and Perplexity cited 325 websites for our buyers' questions, not ours. A grouped selection from our Sep 30, 2026 scan, and how to find yours.

Find out if AI names your business

Type your website to start. The form takes about a minute.

Step 1 of 2
  • Free
  • 6 questions on ChatGPT + Perplexity
  • Results on the scan page when ready, plus an emailed copy; queued or review requests arrive by email within 1 business day.

Use the name customers know you by.

Free: 12 answers, named or not. Full Report, $30 once: 108 answers on 4 AIs, a score and your top 5 fixes. Compare

By requesting, you confirm you may ask for a scan of this website. We use your website and email to run and send your scan, and for up to 2 follow-up emails about it. Privacy

Skip ahead: Get the Full Report, $30

See pricing