Which Sources Answer Engines Cite for Association Questions: 1,363 Answers from ChatGPT, Claude, Perplexity and Gemini (Sep 30 to Oct 7, 2026)
In 1,363 answers to 163 association questions, docs pages were 26.8% of ChatGPT's citations and listicles 0.2%; other engines cited listicles 7% to 22% of the time.
Published October 20, 2026. Data collected September 30 to October 7, 2026. Download the data: data.csv (every number on this page) and questions.csv (the 163 questions).
When association buyers ask AI answer engines about member-facing AI, ChatGPT and the other three engines cite different kinds of sources. In 1,363 answers to 163 association questions, docs and help-center pages made up 26.8% of ChatGPT's citations and listicles 0.2%. Claude, Perplexity and Gemini cited listicles for 7.1% to 22.0% of their sources and docs pages for 2.6% to 4.3%.
In 673 ChatGPT answers to association AI questions in October 2026, the products named most often were Higher Logic (23.9% of answers), Betty (16.9%), CustomGPT.ai (11.3%) and ChatGPT (11.0%). In 230 Claude answers, Higher Logic (44.3%) and CustomGPT.ai (36.1%) were named most.
This study asked ChatGPT, Claude, Perplexity and Gemini the questions association executives, membership directors and IT leads ask when they evaluate AI assistants. It then recorded every source each engine cited and every product it named, and compared the cited pages with Google's top 10 results for the same question. The full question list, the per-engine tables and the method are below.
30-second summary: ChatGPT cited docs pages 26.8% of the time and listicles 0.2%; the other engines cited listicles 7.1% to 22.0%. Source: this study, 1,363 answers, Sep 30 to Oct 7, 2026.
Findings
- ChatGPT cites primary sources; the other engines cite roundups. 26.8% of ChatGPT's 2,410 citations went to docs and help-center pages and 10.7% to government and standards pages. Listicles and comparison pages made up 0.2%. For Claude the listicle share was 22.0%, for Perplexity 17.0% and for Gemini 7.1%.
- ChatGPT is the engine that cites regulators and standards bodies. 20.7% of ChatGPT answers cited at least one government or standards page, such as the FTC or the NIST AI Risk Management Framework. For Claude the figure was 2.2%, and for Perplexity and Gemini 0.9% each.
- The association community's own publisher is ChatGPT's most-cited domain. asaecenter.org received 258 ChatGPT citations, 10.7% of the total, ahead of learn.microsoft.com (182) and higherlogic.com (174). Pages from association bodies appeared in 21.8% of ChatGPT answers, 22.6% of Perplexity answers, 19.1% of Gemini answers and 9.6% of Claude answers.
- ChatGPT rarely cites what ranks on Google. Only 6.3% of ChatGPT's citations were URLs in Google's top 10 for the same question, against 18.8% for Claude, 15.6% for Perplexity and 12.8% for Gemini. At the domain level the figures were 21.6%, 35.5%, 32.3% and 28.5%.
- Gemini and Claude name products; ChatGPT and Perplexity mostly do not. Gemini named at least one product in 94.3% of answers (median 5), Claude in 81.3% (median 3), ChatGPT in 53.8% (median 1) and Perplexity in 50.4% (median 1).
- The engines rarely agree on sources. For the same question, the average overlap (Jaccard) between two engines' sets of cited domains ranged from 0.094 (Perplexity and Gemini) to 0.193 (Claude and Perplexity).
- Platform-choice questions draw the most product names; security questions the fewest. On questions such as "Which AI chatbot should an association use?", Claude named 9.68 products per answer, Gemini 7.82, Perplexity 4.89 and ChatGPT 3.17. On security and governance questions ChatGPT named 0.59 and Perplexity 0.32.
Which association AI products answer engines name
| Engine | Answers naming at least one product | Median named | Mean named | Answers naming a vendor other than a general AI assistant |
|---|---|---|---|---|
| ChatGPT | 53.8% | 1 | 1.49 | 48.3% |
| Claude | 81.3% | 3 | 4.72 | 77.4% |
| Perplexity | 50.4% | 1 | 2.06 | 43.9% |
| Gemini | 94.3% | 5 | 4.88 | 83.5% |
The most-named products, as a share of each engine's answers:
- ChatGPT: Higher Logic 23.9%, Betty 16.9%, CustomGPT.ai 11.3%, ChatGPT 11.0%, iMIS 8.2%, Intercom 7.0%, Microsoft Copilot 6.4%, Chatbase 5.3%.
- Claude: Higher Logic 44.3%, CustomGPT.ai 36.1%, Nimble AMS 24.8%, Salesforce 24.3%, Fonteva 22.6%, Intercom 21.7%, Betty 21.3%, iMIS 20.9%.
- Perplexity: Higher Logic 20.0%, CustomGPT.ai 18.7%, ChatGPT 14.8%, Salesforce 13.0%, Microsoft Copilot 11.7%, Betty 10.0%, iMIS 8.7%, Agentforce 8.3%.
- Gemini: ChatGPT 39.6%, Betty 36.5%, Higher Logic 33.9%, CustomGPT.ai 33.5%, iMIS 31.3%, Salesforce 25.7%, Claude 24.3%, Fonteva 23.5%.
| Question family | Questions | ChatGPT | Claude | Perplexity | Gemini |
|---|---|---|---|---|---|
| Content search | 25 | 1.30 | 5.11 | 2.35 | 5.57 |
| Competitor alternatives | 21 | 2.02 | 5.32 | 3.39 | 4.36 |
| Member services | 20 | 0.65 | 4.69 | 0.73 | 4.62 |
| Platform choice | 19 | 3.17 | 9.68 | 4.89 | 7.82 |
| Pricing | 14 | 1.68 | 4.06 | 1.61 | 5.00 |
| Security and governance | 14 | 0.59 | 1.86 | 0.32 | 3.50 |
| Adoption | 12 | 1.61 | 3.06 | 0.75 | 3.31 |
| AMS integration | 12 | 1.82 | 6.25 | 2.31 | 7.00 |
| Trust and accuracy | 8 | 1.18 | 2.23 | 1.85 | 3.00 |
| Revenue and retention | 7 | 1.11 | 2.64 | 1.73 | 3.00 |
| Use cases | 6 | 0.17 | 2.38 | 0.00 | 4.88 |
| Education | 5 | 0.50 | 2.14 | 1.14 | 1.71 |
Where each engine's citations point
| Page type | ChatGPT | Claude | Perplexity | Gemini |
|---|---|---|---|---|
| Vendor product or home page | 32.6% | 37.7% | 43.8% | 44.6% |
| Docs or help center | 26.8% | 2.6% | 3.8% | 4.3% |
| Association body or its publication | 13.3% | 2.2% | 5.7% | 5.4% |
| Government or standards body | 10.7% | 1.0% | 0.5% | 1.0% |
| Vendor pricing, policy or about page | 8.2% | 1.6% | 1.6% | 3.0% |
| Blog or guide article | 2.7% | 24.6% | 23.5% | 25.8% |
| News or press release | 4.1% | 6.2% | 1.7% | 2.7% |
| Listicle or comparison | 0.2% | 22.0% | 17.0% | 7.1% |
| Original research or report | 1.4% | 1.9% | 2.3% | 2.0% |
| Reference or encyclopedia | 0.0% | 0.2% | 0.0% | 0.1% |
| Unresolved redirect | 0.0% | 0.0% | 0.0% | 3.9% |
| Citations | 2,410 | 1,216 | 1,363 | 1,026 |
How often an answer cites each kind of source
| Answers citing at least one | ChatGPT (673) | Claude (230) | Perplexity (230) | Gemini (230) |
|---|---|---|---|---|
| Association body page | 21.8% | 9.6% | 22.6% | 19.1% |
| Government or standards page | 20.7% | 2.2% | 0.9% | 0.9% |
| Docs or help-center page | 43.4% | 9.1% | 14.3% | 11.3% |
| Listicle or comparison page | 0.9% | 41.7% | 40.4% | 19.1% |
| Page in Google's top 10 for the same question | 19.9% | 53.9% | 61.3% | 29.1% |
The ten most-cited domains per engine
| # | ChatGPT | Claude | Perplexity | Gemini |
|---|---|---|---|---|
| 1 | asaecenter.org 10.7% | higherlogic.com 6.7% | customgpt.ai 13.3% | meetbetty.ai 6.1% |
| 2 | learn.microsoft.com 7.6% | sortresume.ai 5.9% | chitika.com 7.3% | asaecenter.org 4.2% |
| 3 | higherlogic.com 7.2% | memberlounge.app 5.1% | higherlogic.com 6.4% | higherlogic.com 3.4% |
| 4 | meetbetty.ai 5.6% | chitika.com 5.0% | glueup.com 3.7% | sidecar.ai 3.4% |
| 5 | help.openai.com 3.6% | customgpt.ai 4.1% | asaecenter.org 2.6% | burstingsilver.com 3.3% |
| 6 | ftc.gov 3.2% | addsearch.com 3.5% | addsearch.com 2.6% | youtube.com 3.2% |
| 7 | customgpt.ai 3.0% | meetbetty.ai 2.7% | i4a.com 2.3% | propfuel.com 2.2% |
| 8 | airc.nist.gov 2.2% | pollthepeople.app 2.6% | smartthoughts.net 2.1% | customgpt.ai 2.0% |
| 9 | propfuel.com 2.0% | sidecar.ai 2.6% | sidecar.ai 1.8% | sortresume.ai 1.7% |
| 10 | docs.aws.amazon.com 1.9% | smartthoughts.net 2.1% | meetbetty.ai 1.6% | momentivesoftware.com 1.7% |
Shares are of each engine's resolved citations. Some of the domains above, including chitika.com, are publishers whose own pages were cited; see Limits.
Overlap with Google's top 10
| Engine | Citations compared | Cited URL is in Google's top 10 | Cited domain is in Google's top 10 |
|---|---|---|---|
| ChatGPT | 2,410 | 6.3% | 21.6% |
| Claude | 1,216 | 18.8% | 35.5% |
| Perplexity | 1,363 | 15.6% | 32.3% |
| Gemini | 986 | 12.8% | 28.5% |
Ranking on page one of Google for an association question is a weak predictor of a ChatGPT citation for it. Fewer than one ChatGPT citation in fifteen was a page-one URL for the same question.
How much the engines agree
| Engine pair | Mean Jaccard overlap of cited domains per question | Questions compared |
|---|---|---|
| Claude and Perplexity | 0.193 | 138 |
| ChatGPT and Perplexity | 0.122 | 160 |
| ChatGPT and Gemini | 0.109 | 157 |
| Claude and Gemini | 0.107 | 136 |
| ChatGPT and Claude | 0.102 | 138 |
| Perplexity and Gemini | 0.094 | 160 |
A Jaccard overlap of 0.1 means that, of all the domains either engine cited for a question, about one in ten was cited by both. A question counts only when both engines cited at least one source for it.
What this means for association content
For associations that want their own material to appear in AI answers, the ChatGPT pattern is the one to study. It cites the page that is the primary source for a fact: the vendor's own documentation, the regulator's guidance, ASAE's dated articles. It does not cite a page that summarizes other pages. An association's standards, research reports, policy statements and credential requirements are primary sources by definition. The open questions are whether those pages can be fetched without a login, and whether they state their facts in plain sentences with dates.
The other three engines reward a different shape, and the difference is large enough that one page rarely serves all four. A buyer who asks the same question in ChatGPT and in Perplexity is, in this data, usually reading different sources.
Methodology
Questions
163 questions in English, in 12 families: content search (25), competitor alternatives (21), member services (20), platform choice (19), pricing (14), security and governance (14), adoption (12), AMS integration (12), trust and accuracy (8), revenue and retention (7), use cases (6) and education (5).
- Seed questions. 43 were written by hand to mirror questions association staff and buyers ask when they evaluate AI assistants.
- Added questions. 120 more were added in three rounds of 40. They came from Google People Also Ask and related searches, the sub-queries the engines themselves issued, autocomplete, and wording variants that swap the organization noun (association, professional society, trade association, chamber of commerce, credentialing body).
- Wording. Each question was sent exactly as written, with no persona or system prompt.
- List. The full list is in questions.csv.
Engines and models
| Engine | Model and access | Answers | Runs per question | Collected |
|---|---|---|---|---|
| ChatGPT | gpt-6-sol through the OpenAI Responses API with the web_search tool: 652 answers at default search context, 21 at high search context with a US user location | 673 | 4 (5 for 21 questions) | Sep 30 to Oct 7, 2026 |
| Claude | claude-sonnet-5-5 through the Anthropic API with web search | 230 | 2 for 67 questions, 1 for 96 | Sep 30 to Oct 1, 2026 |
| Perplexity | sonar through the Perplexity API | 230 | 2 for 67 questions, 1 for 96 | Sep 30 to Oct 1, 2026 |
| Gemini | gemini-3-flash-preview through the Gemini API with Google Search grounding | 230 | 2 for 67 questions, 1 for 96 | Sep 30 to Oct 1, 2026 |
Google's top 10 organic results for each question were collected on September 30 and October 1, 2026 through the Serper API (United States, English).
Measures
- Citation. One URL in one answer's list of sources, after removing tracking parameters. Gemini returns redirect links; these were resolved to the destination page where the page could be fetched. 40 Gemini citations could not be resolved and are reported as "unresolved redirect".
- Page type. Assigned from the URL pattern and the page's title, with fixed rules for domains whose type is unambiguous. For example, ftc.gov and nist.gov are "government or standards", and arxiv.org is "original research". The classification is approximate: a vendor blog post on a generic URL is counted as a vendor page.
- Association body. A hand-curated list of 20 member associations and their own publications that appeared among cited domains. It includes asaecenter.org, associationsnow.com, americanbar.org, credentialingexcellence.org and uschamber.com. The full list is in data.csv.
- Product named. A whole-word match in the answer text against a fixed list of 63 product and company names (listed in data.csv): association-specific vendors, association management systems, help-desk and chatbot platforms, and general AI assistants.
- Google overlap. Whether a cited URL, or its domain, is among Google's top 10 results for the same question.
Limits
- APIs, not the consumer apps. These are API answers. The consumer apps at chatgpt.com, claude.ai, perplexity.ai and gemini.google.com can retrieve and display sources differently, and logged-in users may see personalized results.
- Citation is not retrieval. The APIs return the sources an answer cites, not every page the engine read.
- Who chose the questions. The 43 seed questions were chosen by hand, not sampled from search logs, so they lean toward buying questions rather than everything associations ask. The 120 added questions reduce, but do not remove, that selection effect.
- Unequal runs and dates. Runs per question differ by engine. ChatGPT was sampled again on October 7; the other engines were sampled once, on September 30 and October 1.
- Pages from this publisher and its category peers. Pages on chitika.com, pollthepeople.app and sortresume.ai received 165 of Claude's 1,216 citations, 135 of Perplexity's 1,363 citations, 21 of Gemini's 1,026 citations and 0 of ChatGPT's 2,410 citations. They are counted like any other source in every table above, and data.csv has their per-domain rows.
- A snapshot. Answer engines change their retrieval often, and these results describe one week.
Data and reuse
data.csv has one row per published number, with its numerator and denominator, so every figure on this page can be checked. questions.csv lists each question with its family and the number of runs per engine. Both may be reused with a link to this page.
How to cite this study: Chitika. "Which Sources Answer Engines Cite for Association Questions: 1,363 Answers from ChatGPT, Claude, Perplexity and Gemini." October 20, 2026. https://www.chitika.com/answer-engines-association-questions-study-2026/
