How Claude, ChatGPT and Gemini Cite Social Listening Tools: Nine Weeks of Data
Between July 7 and August 30, 2026, we asked Claude, ChatGPT and Gemini the same 37 buyer questions about social listening tools, social media APIs and monitoring every week, and logged every source their answers cited: 1,645 answers and 9,458 citations. The three engines build their answers from different webs. ChatGPT draws 38% of its citations from the enterprise listening vendors' own sites and 15% from official platform documentation; Claude and Gemini draw about three quarters of theirs from small vendors and independent blogs, and Claude never cited reddit.com once in 3,818 citations.
This post reports what we measured: which surfaces each engine cites, which domains and pages come up most, how often an engine answers without searching at all, and how the picture differs by question type. The measurement was built to steer our own content, so the panel reflects our category. The method and the open-source agent that ran it are described at the end, so anyone can repeat it for theirs.
Headline Findings
- Each engine has its own source mix. ChatGPT leans on incumbents and official docs; Claude and Gemini assemble answers from a long tail of small vendor pages and independent blogs.
- Reddit is a source for ChatGPT and Gemini, not for Claude. ChatGPT's share of reddit.com citations rose from 3% in early July to 7% at the end of August; reddit.com is Gemini's second most-cited domain; Claude cited it zero times.
- ChatGPT answers without searching one time in four. 24% of ChatGPT answers ran no web search and carried no citations, against 13% for Claude and 0% for Gemini.
- Gemini spreads citations widest. Gemini cited 844 distinct domains, 457 of them exactly once; ChatGPT cited 254 domains and a median of 3 per answer.
- Question type changes the winners. "Best tools" questions are answered from Hootsuite's and Sprout Social's blogs and GitHub lists; pricing questions from small explainer pages rather than official docs; "how do I" questions from Reddit threads.
The Panel and the Numbers Behind It
The panel holds 37 questions a buyer of social listening or social data tools would ask an AI assistant, in five groups: branded (2), category ("what are the best social listening tools", 12), comparison ("best Brandwatch alternatives", 8), pricing ("how much does the Reddit API cost", 2) and problem-solution ("how do I monitor Reddit for mentions of my brand", 13). Thirty-two questions ran from July 7; five were added on August 26.
Each question was asked twice per engine per weekly run, with web search enabled, on Claude Sonnet, GPT-5.4 and Gemini 3 Flash through their APIs. A citation is a URL the engine returned in its answer's source list; we count each domain once per answer. Nine runs produced 596 Claude answers, 532 ChatGPT answers and 517 Gemini answers (Gemini joined in the second week).
| Engine | Answers | Citations | Distinct domains | Median domains per answer | Answers with no search |
|---|---|---|---|---|---|
| Claude | 596 | 3,818 | 424 | 7 | 13% |
| ChatGPT | 532 | 1,998 | 254 | 3 | 24% |
| Gemini | 517 | 3,642 | 844 | 6 | 0% |
Where Each Engine's Citations Come From
We sorted every cited domain into a surface type by hand: the engine's answer is assembled from these, and a vendor can only influence the surfaces it can reach.
| Surface type | Claude | ChatGPT | Gemini |
|---|---|---|---|
| Small vendors and independent blogs | 75% | 29% | 74% |
| Enterprise listening vendors' own sites (Brand24, Sprout Social, Brandwatch, Meltwater, Hootsuite, Talkwalker, Mention and peers) | 12% | 38% | 7% |
| Official platform docs (X, Reddit, TikTok, Meta, Google) | 1% | 15% | 1% |
| Review sites (G2, Capterra, Gartner) | 2% | 6% | 2% |
| reddit.com | 0% | 4% | 4% |
| github.com | 1% | 4% | 2% |
| Publisher platforms (Medium, YouTube, Dev.to) | 1% | 0% | 3% |
| AI tool directories | 2% | 0% | 2% |
| xpoz.ai (our own domain) | 8% | 3% | 6% |
The ChatGPT column explains a pattern anyone who has tried to get a small vendor cited by ChatGPT will recognize. Its retrieval queries, visible in the API's search logs, frequently pin themselves to specific domains ("site:brandwatch.com social listening official 2026", "site:sproutsocial.com social listening official 2026"), so its answers resolve to the incumbents' own pages and to first-party documentation. More than half of ChatGPT's citations (53%) go to enterprise vendors and official docs combined, against 12% on Claude and 8% on Gemini.
Claude and Gemini behave like search engines with a long tail. Three quarters of their citations go to pages outside the enterprise vendors, review sites and platforms: dedicated "best X in 2026" pages from small tools, independent explainers on API pricing, and personal blogs. Gemini's tail is the longest in the study: more than half of the 844 domains it cited appeared in exactly one answer.
Our own domain sits in the last row and needs a caveat. The panel was chosen because these are the questions we want to be cited for, and we have published content against them all summer, so xpoz.ai's share here is a measurement of our own program and not evidence that a different vendor's panel would show the same thing. We report it because leaving it out would misstate the totals.
Try this with Xpoz
No API keys needed. Query Twitter, Reddit, Instagram & TikTok with natural language.
The Most-Cited Domains, by Engine
| Rank | Claude | ChatGPT | Gemini |
|---|---|---|---|
| 1 | xpoz.ai (299) | brand24.com (116) | xpoz.ai (197) |
| 2 | blog.hootsuite.com (90) | sproutsocial.com (93) | reddit.com (136) |
| 3 | socialcrawl.dev (90) | brandwatch.com (85) | twitterapi.io (71) |
| 4 | octolens.com (78) | reddit.com (84) | octolens.com (69) |
| 5 | api.sorsa.io (73) | meltwater.com (80) | github.com (59) |
| 6 | sproutsocial.com (66) | github.com (76) | socialcrawl.dev (53) |
| 7 | twitterapi.io (65) | blog.hootsuite.com (64) | zernio.com (51) |
| 8 | brand24.com (56) | developer.x.com (64) | brand24.com (46) |
| 9 | syncly.app (54) | xpoz.ai (60) | blotato.com (44) |
| 10 | aimultiple.com (51) | talkwalker.com (49) | sociavault.com (44) |
Counts are citations across all nine runs. The Claude and Gemini lists are dominated by companies most marketers have never heard of: SocialCrawl, Octolens, Sorsa, TwitterAPI.io, Zernio, Blotato and SociaVault are small data-API or monitoring vendors that publish one dated, intent-matching page per question ("best social media scraping APIs 2026", "X API pricing 2026"). ChatGPT's list is the enterprise category as a buyer would recognize it, plus Reddit and GitHub.
Reddit's Rising Share on ChatGPT
reddit.com is the one third-party surface whose share moved during the study. On ChatGPT it went from 3% of citations in the first week to 7% in the last, and in the final run reddit.com was ChatGPT's single most-cited domain (19 citations) and Gemini's second (25). In that final run alone, the two engines cited 38 distinct Reddit threads in answers and retrieved more than 200 into their search results.
| Run | ChatGPT reddit.com share | Gemini reddit.com share |
|---|---|---|
| July 7 | 3% | not measured |
| July 13 | 2% | 5% |
| July 21 | 3% | 5% |
| August 2 | 5% | 3% |
| August 10 | 3% | 3% |
| August 17 | not measured | 4% |
| August 23 | 5% | 2% |
| August 26 | 5% | 4% |
| August 30 | 7% | 5% |
The threads that get cited are practitioner questions that match the panel prompt almost word for word ("what's the best way to track brand mentions on reddit without going insane?", "What are you using to monitor Reddit mentions?") and event threads (Reddit's August 5, 2026 announcement on its public Data API was cited by ChatGPT on three different questions in one run). Claude, which cited reddit.com zero times in 3,818 citations, treats the same threads as retrieval results it reads but does not cite.
Which Pages Get Cited: The Latest Run
The pages an engine cites are more instructive than the domains. In the August 30 run, the most-cited third-party pages across the three engines were:
| Citations | Page |
|---|---|
| 8 | octolens.com/blog/best-social-listening-tools |
| 7 | syften.com/blog/best-reddit-monitoring-tools |
| 6 | postproxy.dev/blog/x-api-pricing-2026 |
| 5 | blog.hootsuite.com/social-listening-tools |
| 5 | communitytracker.ai/blog/best-free-social-listening-tools |
| 5 | outx.ai/blog/the-ultimate-guide-to-social-listening-tools-2025 |
| 5 | api.sorsa.io/blog/twitter-api-pricing-2026 |
| 5 | socialcrawl.dev/blog/x-twitter-api-2026 |
| 5 | socialcrawl.dev/blog/reddit-data-api-2026 |
| 5 | therankmasters.com/insights/brand-monitoring/best-reddit-brand-monitoring-tools |
Every page in the table has the same shape: a dated title that restates the question ("best social listening tools", "X API pricing 2026"), an answer in the opening paragraph, and a comparison table. Vendor size does not appear to matter for Claude and Gemini; page shape and freshness do. Over the study, the fastest first citations we observed for a new page were four days after publication, on comparison pages with intent-exact titles; refreshed pages that were already indexed were re-cited within about five days; new tutorials typically took two weeks or more.
Question Type Changes Who Wins
Grouping the citations by the five question types shows that the same engine trusts different surfaces for different intents.
| Question type | Most-cited domains (all engines) |
|---|---|
| Category ("best social listening tools") | blog.hootsuite.com (148), sproutsocial.com (105), github.com (101), octolens.com (81), brandwatch.com (69) |
| Comparison ("best Brandwatch alternatives") | brand24.com (115), syncly.app (60), g2.com (60), hikerapi.com (58), meltwater.com (54) |
| Pricing ("how much does the X API cost") | postproxy.dev (33), redditapis.com (33), socialcrawl.dev (32), twitterapi.io (32), data365.co (31), docs.x.com (28) |
| Problem-solution ("how do I monitor Reddit for my brand") | reddit.com (114), api.sorsa.io (59), twitterapi.io (53), developer.x.com (52), scrapfly.io (45) |
Two details stand out. For pricing questions, five small third-party explainers each out-cite the official X documentation, even though the official page is the primary source; engines appear to prefer a page that states the price in the first sentence over a page that requires navigation. For problem-solution questions, reddit.com is the top source by a wide margin, which is the surface no vendor page can enter.
What This Means for Anyone Trying to Be Cited
- There is no single "AI search" to optimize for. A page that ChatGPT will cite (first-party, official, or on an incumbent's domain) and a page that Claude or Gemini will cite (a dated, intent-exact explainer or comparison) are different pages on different surfaces.
- On Claude and Gemini, a small vendor with the right page beats a large vendor without it. The domains at the top of those lists are companies with a few employees and one page per buyer question.
- Reddit participation is the only route into ChatGPT's and Gemini's community citations. A useful answer in a thread the engines already cite inherits that thread's citation.
- A quarter of ChatGPT answers cannot be influenced by any page at all, because no search ran. For those, the only lever is what the model already knows about a brand: directories, documentation and community mentions that made it into training data.
- Freshness is event-driven. The pages that held citations through the summer were re-dated when the platform they covered changed (Reddit's August 5 announcement moved the Reddit pricing field within days); a changed date with no changed content did nothing in our record.
How We Measured, and How to Repeat It
The measurement ran on an open-source agent, geo-seo-agent, which holds the prompt panel, calls the three engines' APIs with web search enabled, records every answer with its search queries and cited URLs, and writes a weekly summary. Engines: Claude Sonnet (claude-sonnet-5), GPT-5.4, Gemini 3 Flash (gemini-3-flash-preview). Two samples per question per engine per run; nine runs between July 7 and August 30, 2026; 32 questions until August 26 and 37 after. A citation is a URL in the answer's citation list, counted once per domain per answer. Surface types were assigned by hand to every domain. Answers that returned an API error (about 5% of attempts, concentrated in one degraded run in mid-August) were excluded.
Two caveats. Two samples per engine per week means single-week movements are noise; the findings above rest on patterns that held across runs and across questions. And the panel is ours: it over-represents Reddit, X and developer intents relative to a marketer-only panel, which is where several of these findings should be re-tested.
Frequently Asked Questions
Which AI engine cites Reddit the most?
Gemini, by share: reddit.com was 4% of its citations across the study and its second most-cited domain overall. ChatGPT's Reddit share rose fastest, from 3% to 7% over nine weeks. Claude did not cite reddit.com at all, although it retrieved Reddit threads into its search results.
Does ChatGPT cite vendor websites?
Yes, more than the other engines: 38% of ChatGPT's citations in this study went to the enterprise listening vendors' own sites, and its retrieval queries often pin themselves to a vendor's domain with a site: operator. Claude and Gemini gave those vendors 12% and 7%.
How quickly can a new page get cited by an AI engine?
In our record, the fastest was four days from publication to a citation, on comparison pages whose titles matched the question exactly. Refreshed pages that were already indexed were re-cited in about five days. New tutorials and guides typically took two weeks or longer, and some were never retrieved.
Do AI engines cite the same sources as Google ranks?
Partly. The enterprise vendors' blogs that rank on Google for "best social listening tools" are cited on ChatGPT, but on Claude and Gemini the most-cited pages come from small vendors whose Google rankings are modest. Page shape and a current date in the title were better predictors of a citation than domain authority in our data.
