We’ve been tracking monthly changes in LLM citation behavior in this blog series for the past six months, providing you with ongoing insight for building brand AI visibility strategies. This month we’re making a significant change to the report; expanding the volume of prompts in our test set from 4,000 to 120,000, which gives us over 88 million citations to analyze, compared to just 6 million in previous months.
This means our analysis can be much broader and more comprehensive than before, giving you more useful insights. However, it also means that in this first edition of the expanded analysis, it would be inaccurate to make month over month comparisons with the earlier data, since they would be based on completely different sample sizes.
So, for this month only, this report compares the first half of October with the second half. In November’s update, we’ll be able to track a full month of data against the previous period.
Contents
How we measured AI citation behavior in October
Earned media slips, with one platform behind most of the decline
October’s most-cited sources
The leading sources differ across AI platforms
Gemini and Perplexity cut the number of sources in each answer
Reddit splits: Gemini pulls back while Perplexity leans in
Perplexity’s shorter source lists change who appears
Google AI Overviews lean further into video and social
Reference sources move in different directions
What PR and marketing should do next
FAQ
How we measured AI citation behavior in October
Using Meltwater GenAI Lens, we analyzed about 120,000 prompts, 12.7 million answers and 88.6 million citations across eight AI platforms: Claude, ChatGPT, Copilot, Google AI Mode, Google AI Overviews, Gemini, Perplexity, and Grok.
The prompt set spans many industries and languages, and some prompts name brands. The two halves of October (1–15 and 16–31) are compared throughout.
There are two useful ways to read citation performance. Volume tells us how often a source was referenced. Share tells us how much of a citation mix it represented. Copilot’s answer volume grew about fourfold between the two halves of the month (from 253,501 to 1,049,611 answers), which lifts the counts of every source it cites. Where that matters, we point to the share within each model rather than the raw count.
We also compare source movements within each model’s citation mix. That shows how widely a change is distributed, so the models that generate the most citations do not decide the whole story. Domains are grouped at the registered domain, so ncbi.nlm.nih.gov counts as NIH.
These findings describe the tracked prompts and reporting period. They do not establish a universal ranking of trusted sources or measure the quality of AI answers.
Earned media slips, with one platform behind most of the decline
Earned/News fell from 31.1% of the citation mix in the first half of October to 26.8% in the second, a decline of 4.3 percentage points. Sources are grouped into categories: Earned/News covers news and editorial outlets such as Yahoo and Forbes; Social covers platforms such as YouTube, Reddit, Facebook, LinkedIn and Instagram; and Other covers everything else, including brand websites, institutional sources such as NIH, and Google’s own pages. Wikipedia is tracked as its own category. The three largest categories moved as follows:
| Source type | 1–15 Oct | 16–31 Oct | Change |
|---|---|---|---|
| Other | 59.6% | 63.8% | +4.2 percentage points |
| Earned/News | 31.1% | 26.8% | −4.3 percentage points |
| Social | 6.4% | 6.2% | −0.2 percentage points |
Category percentages give more influence to the platforms that generate the most citations, so the model view matters here. About 1.7 percentage points of the decline came from the mix of models, as Perplexity’s news-heavy volume shrank after it cut its source lists (covered below) and Copilot’s grew. The other 2.6 points came from changes within models, and mainly within Perplexity.
| Model | Earned/News, 1–15 Oct | Earned/News, 16–31 Oct | Change |
|---|---|---|---|
| Perplexity | 39.2% | 29.5% | −9.7 pp |
| AI Overviews | 23.6% | 22.3% | −1.3 pp |
| Gemini | 24.3% | 23.7% | −0.6 pp |
| Grok | 36.2% | 35.6% | −0.6 pp |
| Copilot | 22.4% | 22.7% | +0.3 pp |
| Google AI Mode | 23.3% | 23.6% | +0.3 pp |
| ChatGPT | 23.5% | 23.9% | +0.4 pp |
| Claude | 29.0% | 29.7% | +0.7 pp |
Perplexity’s Earned/News share fell from 39.2% to 29.5%. No other model moved by more than 1.3 percentage points, and Claude, ChatGPT, Copilot and AI Mode all edged up. Earned media’s position across most platforms was therefore stable through the month.
The Other category needs similar care. It rose to 63.8%, but it is a broad mix, so the rise says little about any one type of source.
The practical response is to keep developing relevant, well-supported earned stories, and to measure whether that coverage appears in answers on the platforms your audience uses.
October’s most-cited sources
YouTube held first place, with Reddit second and Facebook third. Google ranked fourth, and most of those citations are links to Google’s own shopping and search pages from AI Mode and AI Overviews. Wikipedia gained the most within the top 10, up 37.7%, while Yahoo fell the most, down 10.1%.
Source types follow the research’s classifications, so Yahoo and Forbes sit within Earned/News, while NIH and Google are included in Other. The table also shows how differently a single source can behave across models: Facebook’s total rose 14.4%, driven by Google’s AI surfaces, even though its share fell in Gemini and Claude.
Google AI Overviews alone produced 64% of YouTube’s citations and 75% of Facebook’s. Perplexity produced 61% of Yahoo’s, and Claude produced 43% of Wikipedia’s. A ranking that combines all eight models is useful for scale, but it mostly reflects a few platforms at a time.
The leading sources differ across AI platforms
Most individual models have a very different order from the overall ranking. The table below shows the most-cited source for each model across the whole month (1–31 October), ordered by that source’s citation count. Because Gemini’s change came on 15 October, its entry is mostly pre-change data: Reddit leads for the month, even though Reddit’s share of Gemini’s citations fell sharply in the second half.
| AI platform | Leading domain | Citations, 1–31 Oct | Additional context |
|---|---|---|---|
| Google AI Overviews | YouTube | 1,099,761 | Facebook and Google follow |
| Perplexity | Yahoo | 484,272 | Reddit is second, with 304,068 citations |
| Google AI Mode | 463,622 | YouTube is second, with 294,023 citations | |
| Claude | Wikipedia | 270,358 | NIH and SEC follow |
| Gemini | 146,754 | YouTube is second, with 117,226 citations | |
| Copilot | Wikipedia | 104,774 | LinkedIn and Reddit follow |
| ChatGPT | NIH | 71,986 | FDA is second, with 45,307 citations |
| Grok | X | 67,657 | Wikipedia and NIH follow |
Claude and Copilot led with Wikipedia, and ChatGPT with NIH. Google’s AI experiences leaned on YouTube, Facebook and Google’s own pages, Gemini on Reddit, and Perplexity on Yahoo. Grok led with X.
A single AI visibility score is a useful summary for leadership, but GEO planning needs the model breakdown beneath it: where the brand appears, which sources support those mentions, and how the pattern changes over time.
Gemini and Perplexity cut the number of sources in each answer
Two models changed how many sources they cite per answer. Perplexity began listing at most 10 sources around 6 October, down from about 24. From 15 October, Gemini’s answers fell from about 9.8 links to about 3.7, and its total citations dropped 66%.
With fewer slots, sources compete harder to stay in the answer, which is why totals moved so differently from model to model. The next sections show how this played out: Reddit’s split between Gemini and Perplexity, LinkedIn and Yahoo in Perplexity, and shifts in reference sources such as NIH.
One caveat on timing: Perplexity’s change falls inside the first half of October, so first-half versus second-half comparisons for Perplexity mix days from before and after it. The daily chart in the Reddit section shows the timing more precisely.
Reddit splits: Gemini pulls back while Perplexity leans in
Across all eight models, Reddit’s share of citations slipped only from 1.43% to 1.32%, and it remained the second most-cited source. That stable total hides the largest single source movement of the month.
Gemini’s Reddit citations fell 88% between the first and second halves of October. Gemini cut its citations overall from 15 October, but Reddit was cut harder than the average source. Reddit made up 4.0% of Gemini’s citations on 1 October and 1.1% on 31 October.
Perplexity moved in the opposite direction. Reddit’s share of its citations rose around 6 October, when its source lists were cut, and kept climbing through the month. When Perplexity cites fewer sources, Reddit is one it tends to keep.
Google AI Overviews and AI Mode remained Reddit’s steadiest large audiences at around 3%. ChatGPT stayed near 0.4%, and Claude did not cite Reddit in the tracked prompts.
For brands, the useful question is which AI platforms draw on Reddit. Community activity can support visibility in Google AI Overviews, AI Mode and Perplexity, while it currently does little for Gemini and nothing for Claude. Check which of your audience’s questions surface Reddit threads, and on which platforms, before changing community investment. Gemini’s change took hold over a few days around 15 October, so November’s results will show whether it is a new baseline.
Perplexity’s shorter source lists change who appears
Perplexity remains the largest single contributor of citations in the research, and the domains that appear in its answers changed as its source lists shortened.
LinkedIn saw the sharpest change: its citations in Perplexity fell about 99% between the two halves of October. LinkedIn held steady in Google AI Mode and grew in Copilot, so its overall total fell only 7.4%.
Yahoo’s Perplexity citations fell about 30%, and its overall total fell 10.1%. Perplexity still generated 61% of Yahoo’s citations for the month, and Yahoo remained Perplexity’s leading source. Forbes lost about two-thirds of its Perplexity citations, but growth in Copilot more than offset it.
For brands that care about visibility in Perplexity, the next step is to examine which pages still make the shorter lists. Look at the topics, publishers and types of evidence behind those answers, and treat LinkedIn’s near-disappearance as a platform change to monitor before you shift B2B content plans.
Google AI Overviews lean further into video and social
While Gemini reduced its social sources, Google’s search surface added them. Social rose from 19.9% to 22.1% of AI Overviews citations. YouTube’s overall total rose 8.6% and Facebook’s 14.4%, despite Gemini’s cuts.
For marketers, Google AI Overviews is now the main route for video and social content into AI answers in this data. Keep titles, descriptions and transcripts clear and accurate, check which of your videos and posts are cited, and assess performance against your own audience’s prompts.
Reference sources move in different directions
Wikipedia’s citations rose 37.7%, but its share of Claude’s mix held at about 3.6%. Most of the gain reflects more answers being collected, rather than a stronger preference for Wikipedia.
NIH’s citations fell 6.5%. Its share of Claude’s mix dropped from 2.1% to 1.4%, and Gemini’s shorter source lists cut its NIH citations too.
For brands, the lesson is to keep factual information and supporting evidence accurate and easy to find, then examine which reference sources shape answers in your categories. Claude cites Wikipedia more than any other source, so an out-of-date entry can carry directly into its answers.
What PR and marketing should do next
Plan community activity by platform
Reddit now reaches AI answers mainly through Google AI Overviews, AI Mode and Perplexity. Map which platforms matter to your audience before deciding where community and social effort goes.
Prepare for fewer source slots
Perplexity and Gemini cite fewer sources per answer than they did at the start of October. Don’t rebuild strategy around Gemini’s change until November’s results show whether it lasts, but pages that answer a question directly and show their evidence stay useful whichever way the lists go.
Match formats to Google’s AI surfaces
If AI Overviews matters in your category, review your YouTube and Facebook presence alongside your web content, and measure which assets it cites.
Keep reference information accurate
Wikipedia leads Claude’s and Copilot’s citations, and institutional sources lead ChatGPT’s. Check that the facts these sources hold about your brand and category are current and well sourced.
Track volume and share together
A falling citation count and a falling share of a model’s mix mean different things. Examine both with consistent prompts, reporting windows and source definitions, and record changes in the measurement setup so they are not mistaken for changes in citation behavior.
Meltwater GenAI Lens helps PR and marketing teams track brand visibility and citations across AI platforms. Use that view to connect the sources shaping AI answers with your media, social and content priorities.
FAQ
What were the biggest AI citation developments in October 2026?
Gemini cut its Reddit citations by 88% as it began citing fewer sources overall, while Reddit’s share in Perplexity rose and LinkedIn almost disappeared from it as Perplexity’s source lists shortened. Google AI Overviews gave more space to YouTube and Facebook, and Earned/News slipped to 26.8% of the citation mix, mainly because of Perplexity.
Should brands reduce their investment in Reddit?
The findings do not support that. Reddit remains the second most-cited source, and its share rose in Perplexity and held on Google’s search surfaces. Assess Reddit by the AI platforms your audience uses, and treat Gemini’s change as one to monitor.
Why can a source gain citations without gaining share?
Citation volume depends on how many answers a model gives and how many links each one contains. Share measures a source’s proportion of that model’s citations. Wikipedia’s total rose by more than a third in October, while its share of Claude’s citations stayed about level.
Why aren’t Statista and YouGov in this month’s top 10?
October’s research covers a much wider prompt set than September’s, including branded prompts and many more industries. Statista ranks 93rd with 58,840 citations and YouGov 272nd with 27,047. Those positions reflect the different prompt set and cannot be compared with September’s ranking.
Is the decline in earned media a confirmed trend?
No. Most of it comes from one platform within one month, and Earned/News held steady or rose slightly in most other models. November’s results will help show whether Perplexity’s change persists.
