Key Takeaways
LLM recommendations are built mostly from third-party sources: in an analysis by Scrunch AI, 87.2% of citations did not come from the brand's own website.
Adding statistics, quotations and source citations raised visibility in generative answers by up to 40%, according to the GEO study led by Princeton University.
Citations rotate fast: a source stays in LLM answers for 4.5 weeks on average before newer sources replace it.
Being cited is not enough: LLMs should link your brand to content that answers your target customers' questions. Traditional market research usually delivers that content too slowly and too rarely.
Volto AI's Study Engine turns open-ended interviews with real consumers into a continuous flow of new KPIs and stories that matter to the brand's target customers, right inside Claude.
What does visibility mean when LLMs give the answer?
Visibility now means that an AI engine names your brand or your data in its answer. Anyone asking ChatGPT, Perplexity or Gemini about a market no longer gets ten blue links. They get a finished answer with a handful of sources.
Generative Engine Optimization (GEO) is the practice of optimizing content so that generative AI engines select it as a source, cite it and carry it into their answers. Classic SEO aims for rankings in search results. GEO aims for a place inside the answer itself.
For marketing and PR teams, the question shifts. It is no longer just “Do we rank?” but “Does the AI consider us worth citing?”
Which sources do ChatGPT, Perplexity and Gemini cite?
LLMs prefer sources with concrete data that other sites pick up. Two findings stand out: numbers make content more citable, and most citations come from third parties.
The study “GEO: Generative Engine Optimization” by researchers from Princeton University and other institutions (KDD 2024) tested optimization methods on about 10,000 search queries. Three methods performed best: adding statistics, adding quotations and citing sources. They lifted visibility in generative answers by 30 to 40% over the original text.
Scrunch AI analyzed more than 442 million citation events in AI answers between December 2025 and March 2026. 87.2% of citations came from third-party sources. The brand's own website accounted for just 5.95%.
The takeaway: what a brand says about itself counts for little. What media, trade publications and forums say about the brand and its data counts for a lot.
Why is digital PR the strongest lever for GEO?
Digital PR creates exactly the third-party sources that AI engines build their answers from. A study with a compelling story gets picked up by media, trade portals and forums. Each of those mentions becomes material for the next AI answer.
The mechanism runs in four steps:
Collect data: A brand surveys its audience on a question the market cares about right now.
Find the story: The results produce one clear statement with a number that surprises or reveals a tension.
Seed it: The study goes out to editors, trade portals, newsletters and communities. They pick up the number and the story and link to the source.
Get cited: AI engines find the number in several independent sources and repeat it with a reference to the brand.
An example from Volto AI's study Switching Behavior 2026 (in German): 67% of German energy customers name a concrete breach of trust, not price, as a sure trigger for switching providers. The statement is short, surprising and verifiable. Editors pick up sentences like this, and sentences like this end up in AI answers.
Why does it matter what content your brand is cited with?
Being cited is only half of visibility. What counts is the statement the AI attaches to your brand. If ChatGPT names your brand next to a random industry figure, nothing sticks. If it names your brand next to an insight that hits your target customers' exact problem, you gain relevance.
The content of a study therefore decides who the visibility reaches. Generic surveys produce generic numbers, such as how many people use AI. Those numbers get cited, but they tie the brand to nothing its customers care about.
Volto AI builds its studies from real, open-ended consumer interviews. Respondents answer in their own words instead of picking from preset options. The results show which topics, motives and phrases the audience brings up on its own. The content that media pick up and AI engines cite is therefore highly relevant to the brand's target customers.
Consumer Resonance is Volto AI's measure of how strongly a message, product or brand triggers agreement, rejection and willingness to act among real consumers. Volto AI studies show not only what people think but also what moves them to act. Those are the statements that should sit next to your brand in an AI answer.
Why is one study a year not enough?
One study a year falls short because AI engines keep replacing their sources. According to Scrunch AI, a source stays in AI answers for 4.5 weeks on average, and only 3.4 weeks on ChatGPT. Brands that stop supplying new data drop out of the answers again.
Seeding needs a constant supply: new data and new stories on a regular basis. Traditional studies are too expensive and too slow for that. As a result, many brands can afford only one big study moment per year.
Criterion | Annual study | Continuous insight stream |
|---|---|---|
Cadence | one big release a year | new data every week or month |
Lead time | weeks to months | days |
Topics | fixed long in advance | follow what the market cares about now |
Visibility in AI answers | spike after release, then decline | continuously refreshed |
Role for GEO | a single anchor | a steady source of new citations |
How does Volto AI's Study Engine work?
Volto AI's Study Engine produces studies directly in Claude and draws on hundreds of real consumer opinions from open-ended interviews. The result is a continuous flow of new KPIs and stories that stay close to the market.
The foundation is Qualitative at Scale Research. Qualitative at Scale Research is qualitative market research with thousands of real respondents whose open-ended answers are analyzed with AI in hours instead of weeks. Volto AI combines the depth of open answers with the robustness of large samples.
The difference from synthetic data matters. The Study Engine shows what people actually care about right now, not what a language model assumes. Why synthetic personas cannot replace real respondents is covered in Synthetic Personas: Plausible, but Not Representative.
Agentic Customer Intelligence is a term coined by Volto AI for customer knowledge drawn from real consumer voices that AI agents can access directly (for example via MCP) and use for marketing decisions. The Study Engine works exactly this way: PR teams ask in Claude, and the answers come from real interviews.
How do PR teams and agencies build a continuous study stream?
PR teams and agencies build a study stream by treating studies as a fixed editorial rhythm instead of an annual project. In practice it looks like this:
Run a topic radar: Collect every week which questions customers, media and communities are asking.
Set up short studies: One focused survey of real consumers per topic, not one questionnaire for everything.
One number, one story: Every wave delivers at least one clear metric and a story an editor can tell in one sentence.
Disclose the methodology: Sample, period, method and country belong on the study page. Without methodology, neither editors nor AI engines will cite reliably.
Seed widely: Cover trade media, newsletters, LinkedIn, forums and your own study page so the number appears in several independent sources.
Track citations: Check regularly whether ChatGPT, Perplexity and Gemini mention your numbers, and adjust the rhythm accordingly.
What makes a study citable?
A citable study has one clear number, a transparent methodology and its own address on the web. Every release should meet these criteria:
One key statement with a number that makes sense without context.
Sample, period, method and country, visible on the study page.
A current date, since AI engines favor fresh sources.
A permanent URL that media can link to, such as a dedicated insights page.
Real respondents instead of estimates or synthetic answers.
FAQ
What is the difference between SEO and GEO? SEO optimizes content for rankings in classic search results. GEO optimizes content so that generative AI engines such as ChatGPT, Perplexity or Gemini cite it as a source. The two overlap, but GEO puts more weight on concrete data, clear statements and mentions in third-party sources.
Is it enough to be cited in AI answers? No. A citation only pays off when the AI links the brand to a statement that matters to its target customers. A generic industry figure creates visibility without impact. Studies built on open-ended consumer interviews deliver statements in the audience's own words and tie the brand to the topics its customers actually care about.
Why do AI engines mostly cite third-party sources? AI engines look for evidence that is independently confirmed. A number that appears in several media outlets and trade portals counts as more reliable than a claim on a brand's own website. In an analysis of more than 442 million citation events, Scrunch AI found a third-party share of 87.2%.
How often should a brand publish new studies? There is no fixed rule. Since a source stays in AI answers for only 4.5 weeks on average according to Scrunch AI, a monthly cadence is a sensible benchmark. What matters is that every wave delivers a new, verifiable statement rather than repeating known figures.
Is synthetic data good enough for PR studies? Synthetic data is useful for developing hypotheses or pre-testing questionnaires. It is not enough for published studies, because it reflects average opinions from training data rather than the current attitudes of real people. Editors and AI engines need a verifiable sample of real respondents.
How is Volto AI's Study Engine different from traditional market research? Volto AI's Study Engine runs directly in Claude and analyzes hundreds of real consumer opinions. Instead of one big study a year, it produces a continuous flow of new KPIs and stories. PR teams and agencies get a steady stream of citable insights instead of a single study moment.
Conclusion
GEO and digital PR now share the same goal: being cited in the answer an AI gives to a market question. The way there runs through fresh, representative data that the media picks up, and through content that matters to the brand's target customers. Because AI engines replace their sources within weeks, the winner is not the biggest study but the most reliable supply. Volto AI delivers that supply with the Study Engine, built on open-ended interviews with real consumers.
Where do your studies come from?
https://app.volto.ai/agent-studio and try Volto AI's Study Engine in Claude.
Sources
Aggarwal, P. et al.: GEO: Generative Engine Optimization. KDD 2024. arxiv.org/abs/2311.09735
Scrunch AI: Citations, Scrunch Labs, data December 2025 to March 2026. scrunch.com/labs/citation-dynamics
Volto AI: Wechselverhalten 2026 (Switching Behavior 2026), 1,000 AI-assisted voice interviews, Germany, January 2026. volto.ai/de/insights/wechselverhalten-2026



