AI search visibility report for September 2026: YouTube falls and YouGov enters the top 10

A source can lose thousands of AI citations and still become more prominent in some models’ answers, while another can climb the overall ranking because one platform suddenly cites it much more often. September’s AI search visibility research highlights both patterns, alongside two clearer developments: YouTube declined across the models that meaningfully cited it, and YouGov gained ground across all eight platforms. 

Following August’s LinkedIn-led results, this month gives PR and marketing professionals fresh reasons to examine where citations come from before deciding where to invest.

Contents

• How we measured AI citation behavior in September

• September’s most-cited sources

• YouTube declines across six AI models

• YouGov enters the top 10 across all eight models

• Earned media recovers after four monthly declines

• Yahoo’s growth is driven by Perplexity

• LinkedIn and Reddit tell different stories by model

• Reference sources move in different directions

• The leading sources differ across AI platforms

• What PR and marketing should do next

• FAQ

How we measured AI citation behavior in September

Using Meltwater GenAI Lens, we analyzed more than 4,000 prompts and approximately 6.1 million citations across eight AI platforms: Claude Sonnet 4, ChatGPT using GPT-5, Copilot, Google AI Mode, Google AI Overviews, Gemini 2.5 Pro, Perplexity Sonar Pro, and Grok 4.

The prompts covered a balanced mix of high-intent consumer categories, including AI, retail banking, automotive, fast food, streaming services, apparel, and fashion. Prompts were distributed equally across categories and contained no brand mentions. September is the reporting period, with August as the main comparison baseline.

This update builds on our reports covering March to April, April to May, June, July, and August.

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. When the overall pool contracts, a source can receive fewer citations while accounting for a larger proportion of the remaining total.

September’s research also compares source movements within each model’s citation mix. That gives us a way to assess how widely a change is distributed, rather than allowing the models that generate the most citations to determine the whole story.

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.

September’s most-cited sources

YouTube retained first place despite a steep decline, and Yahoo moved ahead of Reddit by just 79 citations, while LinkedIn ranked fourth. Market research firm, YouGov entered the top 10 for the first time.

Source types follow the research’s classifications, so YouGov, for example, sits within Earned/News, while NIH and Statista are included in Other. Those distinctions help explain why a category’s performance can differ from the fortunes of a familiar publisher or social platform.

YouTube declines across six AI models

YouTube generated 92,855 citations in September, a fall of more than half over the previous month. Our research shows that its citation share declined across all six models with meaningful YouTube activity: Perplexity, Copilot, Gemini, Google AI Overviews, Google AI Mode, and Grok.

That breadth makes this a stronger signal than a decline confined to one platform, and also answers a question raised in August’s report, when YouTube’s softer month followed a strong July rebound. September adds another decline, with a much larger drop in volume, but even so, YouTube remains the leading individual source. 

Gemini generated 33,939 YouTube citations, Google AI Overviews 23,588, and Google AI Mode 23,298. Video still has a substantial place in the source mix.

For marketers, the immediate task is to review reliance on YouTube. Check which questions your videos help answer, which AI platforms cite them, and whether other sources can support the same information. Keep titles, descriptions, and transcripts clear and accurate, then assess performance against your own audience’s prompts.

YouGov enters the top 10 across all eight models

YouGov recorded 21,558 citations and entered the top 10 for the first time and, more strikingly, the research records an increase in its citation share across every model studied.

Perplexity generated 4,915 YouGov citations, Gemini 4,227, ChatGPT 4,208, and Copilot 2,267, while the remaining four models also cited it. YouGov had citations in August, but they were too small to reach the top 10. September therefore marks a substantial increase in visibility for an existing source.

For brands, this offers a useful hypothesis to test: original polling and clearly presented research may create more opportunities to appear in AI answers. Publish findings that address questions your audience actually asks, explain the methodology, and make the underlying evidence easy to locate. Give each important finding enough context to be understood on its own.

YouGov’s performance is evidence about one research source during one month, so it cannot establish that all similar survey content will gain citations. The breadth of its rise makes research-led content worth testing, with results measured across individual AI platforms.

Earned media recovers after four monthly declines

Earned/News rose from 29.4% of the citation mix in August to 35.6% in September, an increase of 6.2 percentage points. It is the first recovery after four consecutive declines from April through August.

The three largest source categories moved as follows:

This is encouraging for PR, though the category remains below April’s 39.5%, so October’s results will help establish whether September begins a sustained recovery.

The distinction between share and volume is particularly useful here. Forbes generated 46,513 citations in September, fewer than in August, even as the wider Earned/News category gained share, showing that a category can strengthen relative to the total without every publisher receiving more citations.

Category percentages also give greater influence to platforms that generate more citations. The research identifies Perplexity and Gemini as contributing the majority of tracked citations, so this aggregate recovery should be read alongside individual model results.

The Other category (which contains the long-tail of sources including corporate, government, and academic websites) requires similar care, because while it has remained above 50% for six months, it combines brand websites and blogs with institutional and data sources such as NIH and Statista. Its size does not establish that brand-owned content is gaining visibility. Teams need to track citations to their own domains directly.

The practical response is to keep developing relevant, well-supported earned stories and measure whether that coverage appears in AI answers.

Yahoo’s growth is driven by Perplexity

Yahoo recorded 80,575 citations, continuing its growth and reaching second place overall, and Perplexity generated 59,980 of those citations, approximately 74% of the total.

The research’s model comparison shows Yahoo gaining considerable share within Perplexity, while its share declined in Claude, Google AI Mode, and Google AI Overviews. Its overall growth therefore tells us much more about Perplexity than about a common preference across AI platforms.

There is another reason to avoid treating this as a settled hierarchy. Among the leading domains tracked, Perplexity’s primary source changed from Facebook in July to LinkedIn in August and Yahoo in September.

For brands that care about visibility in Perplexity, the next step is to examine the specific Yahoo pages being cited. Identify the topics, original publishers, and types of evidence behind those answers before deciding whether the pattern creates a relevant media opportunity.

A domain’s rising citation count alone cannot tell you which story to pitch or whether its audience aligns with yours.

LinkedIn and Reddit tell different stories by model

LinkedIn generated 73,264 citations, down from roughly 100,200 in August, and Perplexity remained its largest contributor with 53,947 citations, approximately 74% of LinkedIn’s total.

The research reports only a small change in LinkedIn’s share within Perplexity’s source mix. It also records a gain in Google AI Overviews, even as most other active models reduced LinkedIn’s share. That combination shows why a lower overall count can overstate the extent of a source’s retreat within a particular platform.

LinkedIn still provides a substantial route into Perplexity’s answers in this sample. Executive expertise, company research, and useful explanations remain sensible content priorities for B2B, with their AI contribution assessed alongside their wider marketing value.

Reddit’s volume also fell sharply, by 38%. Its model-level picture was mixed: the research records share declines in Grok and Perplexity, alongside gains in Google AI Mode, Google AI Overviews, and Gemini.

Facebook shows an even sharper split, with the September export records meaningful activity on Google’s AI platforms, while Facebook citations in Perplexity and Grok fell below the export’s reporting threshold. 

Measure each social source by AI platform. A blended social total can hide gains on one surface and losses on another, leading teams to make decisions that do not fit the platforms their audiences use.

Reference sources move in different directions

NIH, Wikipedia, and Statista all recorded lower citation volumes in September - their relative positions within individual models were much less uniform.

The research records gains in NIH’s citation share across five of eight models, despite its lower overall count. Statista gained considerable share within Grok while losing share within ChatGPT. Wikipedia became more prominent within Claude while declining within Grok and ChatGPT.

The raw September counts show that these sources still play a substantial role. NIH generated 51,230 citations, Wikipedia 38,480, and Statista 37,225. All three remained in the top 10.

For brands, the useful lesson is to keep factual information and supporting evidence accurate and accessible, then examine which reference sources influence answers in the categories you operate in. The data supports a more specific assessment of source influence than an overall claim that institutional authority is either rising or fading.

The leading sources differ across AI platforms

The overall ranking begins with YouTube, Yahoo, and Reddit. Several models have a very different order.

The table below identifies the leading source for each model among September’s overall top 10 domains. It does not represent a ranking of every source that model cited.

ChatGPT and Claude continued to draw heavily from NIH within these leading domains. Google’s AI experiences retained YouTube as their leading source, while Perplexity’s September mix was led by Yahoo and LinkedIn. Statista led for both Copilot and Grok.

A single AI visibility score can give leadership a useful overview, but GEO planning needs the model breakdown underneath it: where the brand appears, which sources support those mentions, and how the pattern changes over time.

What PR and marketing should do next

Review concentration in your source mix

Find out how much of your brand’s AI visibility depends on one domain or one model. YouTube’s decline and Perplexity’s changing leaders show why that concentration deserves attention. Use your brand’s actual citation pathways to identify gaps and opportunities.

Test original research that answers useful questions

YouGov’s rise gives research-led content a clear reason to feature in the next round of GEO experiments. Choose a relevant audience question, publish credible findings with a transparent methodology, and provide a concise explanation alongside the full analysis. Measure which platforms cite the work and which pages they use.

Keep earned media connected to supporting evidence

Continue supplying journalists with useful data, knowledgeable spokespeople, and clear context. When coverage appears, connect it to relevant owned explanations and research with accurate attribution. Monitor whether those sources contribute to the brand’s presence in AI answers.

Track volume and share together

A falling citation count and a falling proportion of a model’s source mix mean different things. Examine both, using consistent prompts, reporting windows, and source definitions. Make sure changes in the measurement setup are recorded so they can be distinguished from changes in citation behavior.

Review results monthly

September brought a broad YouTube decline, a new top-10 research source, and another change in Perplexity’s leading domain. A monthly review gives you a practical way to notice movements, test responses, and build enough history to assess whether they persist.

Meltwater GenAI Lens helps PR and marketing 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 September 2026?

YouTube’s citation volume more than halved, with the research recording a share decline across all six models with meaningful YouTube activity. YouGov entered the top 10 and gained citation share across all eight models. Earned/News also recovered to 35.6% of the overall citation mix.

Should brands reduce their investment in YouTube?

The findings support reviewing how much AI visibility depends on YouTube. It remains the most-cited individual source, and video can serve marketing goals beyond AI discovery. Assess its performance for your brand, audience, and target AI platforms before changing investment.

Why can a source lose citations while gaining share?

Citation volume measures the number of references. Share measures a source’s proportion of a defined citation pool. If that pool shrinks faster than the source’s own count, the source can receive fewer citations while representing a larger share. The denominator must be consistent for comparisons to be useful.

Does YouGov’s rise mean AI platforms prefer survey content?

It shows that one established research source gained citation share across all eight tracked models in September. That supports testing relevant, clearly documented research content. It does not establish a universal preference for surveys or guarantee that a brand’s research will be cited.

Is earned media’s recovery a confirmed trend?

Earned/News increased from 29.4% to 35.6% after four monthly declines. One recovery month is not enough to establish a sustained trend. October’s results will help show whether the improvement continues.