---
title: "Original research: What actually gets cited in AI search [Updated for 2026]"
description: Original research earns AI citations and builds search visibility in ways optimised content can't. Here's why it matters — and how small teams can do it.
image: https://amplistory.com/hubfs/Blog%20images/Blog%20banners%20(1)-2.png
---
[.png)](https://amplistory.com)
- [Services](https://amplistory.com/#services)
- [Case studies](https://amplistory.com/portfolio)
- [Blog](https://amplistory.com/blog)
- [Messaging templates](https://amplistory.com/messaging-framework-template)
- [About me](https://amplistory.com/about-us)
[](https://amplistory.com/hs/cta/wi/redirect?encryptedPayload=AVxigLJ6f3vacLa924TcaZ%2BxDmk%2Frrszv5ZTKZjSooraEqLYf5AgaUEJvdcOXU9hKCgww%2BxA8SG%2BOFApoJyfC2iOLq75oQMTYK3iUamn%2B0Abnaf%2F82fzo6adGdDuDsO3ya7t%2FSQj4qXl7yc0dH3FfWwkadxkgQKvMlz%2BkHuG2sN2M0hDoWm5ExrkS8ZozOAEZR45&webInteractiveContentId=437682054388&portalId=144074470)
Marketing Strategy
# Original research: What actually gets cited in AI search \[Updated for 2026\]
[Grace Windsor](https://amplistory.com/blog/author/grace-windsor)
Jul 23, 2025
---
Originally published 23 July 2025. Updated 23 September 2026.
When I first published this article in July 2025, AI search was already moving quickly. It has not slowed down. New research, tools, and terminology are appearing all the time, and some of what looked fairly settled a year ago now needs more qualification.
This version reflects what we know in September 2026. If you want to follow the field as it changes, [Ahrefs](https://ahrefs.com/blog/how-to-rank-in-ai-overviews/) and [Searchable](https://www.searchable.com/blog/geo-vs-seo-vs-aeo) are both worth keeping an eye on for current research, tools, and reports.
My own view has evolved with the evidence too. But one thing has held up: whatever happens to AEO, GEO, and the rest of the terminology, original research still gives you something genuinely new to contribute.
It gives people, search engines, and AI systems something that is not simply another synthesis of material that already exists.
## **Search is splintering, even when the foundations overlap**
Not long ago, ranking on Google was the main search objective. Now people move between traditional search results, AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini, and Copilot. Those surfaces do not all retrieve or present information in the same way, and visibility in one does not guarantee visibility in another.
[Ahrefs compared 730,000 paired AI Mode and AI Overview responses](https://ahrefs.com/blog/ai-overviews-vs-ai-mode/) and found only 13.7% citation overlap, even though the answers were semantically similar most of the time. So even inside Google, “AI search” is not one surface.
It is worth distinguishing AI Overviews from AI Mode. [AI Overviews appear inside standard Google search results](https://search.google/pdf/google-about-AI-overviews-AI-Mode.pdf) and are designed to help people get the gist of more complex queries quickly. AI Mode is a separate conversational experience designed for deeper exploration and follow-up questions, using query fan-out to run multiple related searches.
[Since January 2026, Google has let users move directly from an AI Overview into AI Mode](https://blog.google/products-and-platforms/products/search/ai-mode-ai-overviews-updates/) while carrying the context with them. The experiences are getting closer for users, even though they can still draw on different sources.
This is where the SEO, AEO, and GEO labels can still be useful, as shorthand rather than hard categories:
- SEO = Helping pages get discovered and ranked in traditional search results.
- AEO = Making content clear enough to be extracted or presented as a direct answer in search and answer surfaces, such as featured snippets and AI Overviews.
- GEO = Improving the chances that your content or brand is cited, mentioned, or synthesised in generative AI responses, including through third-party sources.
[Searchable’s recent comparison](https://www.searchable.com/blog/geo-vs-seo-vs-aeo) puts the practical overlap between the three at roughly 80%. [Google itself describes AEO and GEO as industry terms](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) and says the core SEO fundamentals still apply to its generative AI features.
That distinction matters because the job is now broader than ranking. You also need to understand whether your content is being cited, whether your brand is being mentioned accurately, which third-party sources are shaping the answer, and whether the same thing happens across different platforms.
I do not think that means building three separate content strategies. It means keeping the SEO foundation, while paying more attention to whether your content is clear enough to retrieve and cite, and what exists about your brand beyond your own website.
## SEO still matters, but it is not a direct map to AI citations
The earlier version of this article said that the majority of AI Overview citations came from pages already ranking in Google’s top ten.
I would not state that as a general rule now.
[A later Ahrefs analysis](https://ahrefs.com/blog/ai-overview-citations-top-10/) published in 2026 looked at 863,000 keyword search results and around four million AI Overview URLs. It found that 37.9% of URLs cited in AI Overviews also appeared within the first ten result blocks for the same query.
That is still meaningful overlap, and [Google itself says the SEO fundamentals remain relevant](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) because AI Overviews and AI Mode are built on its Search systems. To appear as a supporting link, a page still needs to be indexed and eligible to appear in Search with a snippet.
That tells us SEO still matters, but an AI Overview does not simply cite the pages ranking highest in ordinary search. Some cited pages are already in the top results; many are not.
Different studies put the overlap at different levels, partly because the datasets and systems keep changing. The consistent point is that SEO and AI visibility overlap, but they are not the same thing. Ranking well can make content easier to discover, but it does not determine whether an AI system cites it.
## AI search also changes how topics are retrieved
One part of the earlier AEO discussion that does still look useful is query fan-out.
[Google describes query fan-out](https://developers.google.com/search/docs/appearance/ai-features) as the system running several related searches behind a user’s original question. That means an AI-generated answer may retrieve passages from pages that answer different parts of the wider topic, not just pages targeting the exact wording of the original query.
[Ahrefs points to research](https://ahrefs.com/blog/how-to-rank-in-ai-overviews/) showing a strong relationship between appearing across those fan-out queries and being cited in the final AI Overview.
I would not turn that into another content hack. The practical implication is much more ordinary: understand the question properly, cover the parts that genuinely belong to it, and build enough depth around the subject that your useful pages can be found from more than one route.
[The same Ahrefs analysis](https://ahrefs.com/blog/how-to-rank-in-ai-overviews/) found almost no correlation between page word count and AI citations. Longer is not automatically better.
That matters because AI optimisation advice can easily become an excuse to add more sections, more FAQs, more definitions, and more words. If they dilute the actual answer, they are not helping.
## **Original research does seem to punch above its weight, with a caveat**
There is now better evidence for the argument that made me write this article in the first place.
[A July 2026 Search Engine Land analysis of Gauge citation data](https://searchengineland.com/why-most-original-data-never-gets-cited-481676) looked at 301 live pages that had been cited across 316 prompts in seven verticals, carrying 1,075 citations in total.
Only eight of those 301 pages qualified as primary research, where the original data and methodology were published on the page. Those eight pages received 90 citations. On average, the primary-research pages received 11.3 citations each, compared with 3.4 for the other pages in the sample.
That sounds like a very strong case for original research, suggesting original data can punch above its weight when it answers a specific question that other people need a source for.
[Google’s own guidance](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) arrives at a similar point from a different direction. It recommends creating unique, expert-led, non-commodity content rather than recycling what is already on the web or could easily be generated by an AI model.
Original research is one way to do that. It is not the only one. First-hand experience, genuinely useful analysis, and a distinctive expert point of view can also add information that was not already there.
## **Your own website is only part of the citation story**
Publishing good research on your own site matters, but AI systems do not build their answers only from the pages you control.
[Searchable analysed 19.2 million citations](https://www.searchable.com/blog/improve-ai-brand-visibility) across 4.6 million AI answers for 7,768 brands in the 30 days to 17 September 2026. For unbranded prompts, where the user did not name the company, only about 5% of citations pointed to the brand’s own website. When the prompt named the brand, that rose to about 44%.
The exact percentages will move, but the difference is useful.
If somebody already knows your company and asks about it directly, your own site can be an important source.
If they are asking a category question before they know you, the answer is much more likely to be built from competitors, publishers, review sites, forums, institutions, and other third parties.
[State of Brand makes a similar point using AirOps data](https://www.thestateofbrand.com/news/ai-search-third-party-citations-strategy), which found that 85% of brand mentions in AI answers came from third-party pages rather than owned domains.
I would be careful about jumping from that to “your website no longer matters”. It does.
Your own site is still where you explain the work properly, host the original evidence, convert interest, and give people somewhere authoritative to check what is true.
But it does mean distribution matters.
If you publish useful original research and it gets referenced by journalists, trade publications, partners, analysts, newsletters, communities, or other credible sites, the research starts to travel beyond the page where you first published it.
That was always valuable. AI search makes the value easier to see.
## **What counts as original research?**
Original research means you are the source of the evidence, rather than summarising somebody else’s.
For a B2B business, that could include:
- A customer or buyer survey designed around a specific question
- Analysis of first-party product, usage, CRM, pricing, or support data
- A benchmark or comparison where you publish the method as well as the result
- Structured customer or expert interviews where you explain who you spoke to, what you were trying to understand, and how you analysed the patterns.
Original does not have to mean a huge dataset, but the method has to match the claim.
Five customer interviews can tell you a great deal about how a particular group made a buying decision. They cannot tell you that 60% of the market behaves that way. A LinkedIn poll can be a useful pulse check. It should not suddenly become a representative industry statistic because the number looks good in a headline.
The bar is not academic rigour. It is being clear about what you actually did, what the evidence supports, and where the limits are.
If customer interviews are part of the research, I have a separate [guide to running B2B customer interviews](https://amplistory.com/blog/guide-customer-interviews) to help you get started.
## **What makes research useful as a source?**
The temptation with AI search is still to start formatting the research for the machine.
I think that is backwards.
A source becomes useful because the underlying work answers something, and because another person can understand where the answer came from.
### 1. Start with a question that deserves research
“Let’s publish some original data for AI search” is not a research question.
Start with something customers, buyers, journalists, partners, or your own team genuinely do not know. A useful benchmark. A change in buyer behaviour. A pattern in your own product data. A question that keeps coming up in sales conversations but is poorly answered elsewhere.
The research should still have a reason to exist if no AI system ever cites it.
### 2. Show your workings
If you publish a statistic, give people enough information to understand it.
Who or what was included? How large was the sample? When was the research carried out? How did you define the thing being measured? What are the obvious limitations?
This is useful for human readers and for anyone deciding whether your finding is safe to repeat.
### 3. Make the important findings easy to find
You do not need to turn an article into a collection of tiny answer blocks or force every heading into a question.
But if a section answers a clear question, answer it clearly. Do not bury the useful finding under three paragraphs of scene-setting.
[Google says there is no requirement to “chunk” content](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) for its generative AI systems. [Ahrefs also found no meaningful relationship](https://ahrefs.com/blog/how-to-rank-in-ai-overviews/) between simply making a page longer and earning more AI citations.
Clear structure is useful because it helps people and systems understand what the page contains. That is different from writing in a special machine-first style.
If you publish a report, put the main findings, context, and methodology on a stable, crawlable web page rather than hiding everything inside a gated download.
The PDF can still exist. It just should not be the only place the evidence lives.
### 4. Keep the evidence current when it actually changes
There is evidence that recently maintained content is common among pages cited by AI systems, but I would still be careful about turning that into a fixed refresh schedule.
[Seer Interactive’s July 2026 study](https://www.seerinteractive.com/insights/study-content-recencys-impact-on-ai-visibility-in-2026) analysed 7,683 dated pages carrying 47,097 citations across ChatGPT, Gemini, and Perplexity. Seventy-five per cent of those pages had been updated within the previous year.
The interesting part was that many were older pages that had been maintained, rather than brand-new pages. The pattern also varied by platform, industry, and content type.
That supports maintaining useful research. It does not mean changing a date every three months will make a page more citable.
Refresh the work when the underlying data, market, examples, or interpretation have changed. If nothing substantive has changed, leave it alone.
### 5. Do not create research just so AI has something to cite
“Original research gets cited” can become another content tactic very quickly.
A badly designed survey can generate a unique statistic and still be bad research. A proprietary dataset can be genuinely exclusive and still answer a question nobody cares about. Ten interviews can produce a lovely report that confirms exactly what the team already believed because the questions were designed that way.
The research should first help the business understand something it did not know, reduce uncertainty around a decision, or challenge an assumption that matters.
If it then becomes a source that customers, journalists, other writers, or AI systems want to cite, that is a valuable second benefit.
That is why I still think the research question and what you plan to do with the answer matter more than the eventual content format. My [guide to making customer research actionable](https://amplistory.com/blog/customer-research-actionable) goes into that part in more detail
## How do you measure AI visibility now?
AI answers are not fixed search rankings.
[Searchable’s visibility work](https://www.searchable.com/blog/ai-brand-visibility-tracking) makes the problem quite clear: the same prompt can produce different answers on different runs, and different AI platforms can mention or cite different sources for the same question.
So a single manual search is interesting, but it is not much of a measurement system.
For a small B2B business, I would keep this practical.
Choose a small, stable set of questions that genuinely matter to your buyers. Include unbranded questions, where the user is looking for an answer or category rather than asking about you by name. Run the same set over time and across the AI platforms that matter to your audience.
Then look at a few different things:
- Mention rate: How often the brand appears at all
- Citation rate: How often the answer actually links to your site or research
- Share of voice: Whether you appear alongside or instead of the competitors you care about
- Accuracy: Whether the answer describes you correctly
- Referral quality: Whether identifiable AI traffic actually does anything useful once it arrives.
[Searchable also makes an important distinction between branded and unbranded prompts](https://www.searchable.com/blog/improve-ai-brand-visibility). Mixing the two can make visibility look much stronger than it really is. In its September 2026 dataset, the median brand appeared in 93% of answers to branded prompts but only 7% of answers to unbranded prompts.
For discovery, that second number is usually the more interesting one.
You can then add the first-party data you already have. [Google introduced dedicated generative-AI performance reporting in Search Console](https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports) in 2026, and [ChatGPT Search referral links include the parameter utm\_source=chatgpt.com](https://help.openai.com/en/articles/12627856-publishers-and-developers-faq).
Third-party monitoring tools can make repeated tracking easier, particularly across several platforms. I would still treat any single visibility score as directional rather than absolute. There is no universal AI rank position in the way there is a traditional search result.
## Wrapping up
Original research is not a guarantee of AI visibility. Different systems retrieve and cite information differently, and the evidence is still developing.
But research can create something generic synthesis cannot: evidence that originated with you.
If your page says essentially the same thing as 30 other pages, adding an FAQ block, making it 500 words longer, or giving it a new piece of schema is not much of a defence.
If you have data, experience, or customer evidence that only you can provide, you have something harder to replace.
Then the work is fairly recognisable: publish it properly, make the evidence understandable, keep it current, and give other people a reason to reference it.
The AI citation is one possible distribution benefit.
The more important asset is owning evidence that your customers, your team, and other people in the market have a reason to use.
If you are planning a research project, start with how to make customer research actionable: what decision should the research change before you collect anything?
---
*Not sure whether you're facing a marketing problem or something deeper? Bring one business challenge to [a free Alignment Call,](https://amplistory.com/story-alignment-check) and we'll work through where the real issue might be.*
---
[Marketing Strategy](https://amplistory.com/blog/tag/marketing-strategy) [AI in research and messaging](https://amplistory.com/blog/tag/ai-in-research-and-messaging)
## Similar posts
Customer research
### [How to prove the ROI of customer research](https://amplistory.com/blog/prove-roi-customer-research)
Customer research is an easy budget to cut — until something goes wrong. Here's how to build the case for it before that happens.
Grace Windsor Sep 10, 2025
Positioning and messaging
### [B2B Rebrand Strategy: Why Customer Research Comes First](https://amplistory.com/blog/rebrand-customer-research-messaging)
Two rebrands. One built on internal opinion, one built on customer research. Here's what the difference looked like in practice.
Grace Windsor Nov 5, 2025
Customer research
### [How to make customer research actionable](https://amplistory.com/blog/customer-research-actionable)
Most customer research doesn't fail in the field. It fails when the findings land in a document nobody acts on. Here's how to close that gap.
Grace Windsor Oct 9, 2025
[.png)](https://amplistory.com)
Supported by
[](https://udaras.ie/en/)
### Work
- [Research & strategy services](https://amplistory.com/#services)
- [Case studies](https://amplistory.com/portfolio)
- [Story alignment call](https://amplistory.com/story-alignment-check)
### Resources
- [Blog](https://amplistory.com/blog)
- [A Guide to B2B Messaging Alignment](https://amplistory.com/b2b-messaging-alignment)
- [One-page messaging framework](https://amplistory.com/messaging-framework-template)
- [When to work with an external parter](https://amplistory.com/blog/marketing-agency-consultant-fractional-cmo)
### Company
- [About me](https://amplistory.com/about-us)
- [Fúinn](https://amplistory.com/fúinn)
- [Get in touch](https://amplistory.com/contact-us)
© 2026 AmpliStory All rights reserved [Privacy Policy](https://amplistory.com/privacy-policy)
[Powered by Atlas - a B2B SaaS HubSpot theme](https://www.kalungi.com/atlas-hubspot-theme-for-b2b-saas-software)
```json
{
"@context" : "https://schema.org",
"@type" : "BlogPosting",
"author" : {
"@type" : "Person",
"name" : "Grace Windsor",
"url" : "https://amplistory.com/blog/author/grace-windsor"
},
"dateModified" : "2026-09-23T09:34:59.653Z",
"datePublished" : "2025-07-23T09:33:00.000Z",
"headline" : "Original research: What actually gets cited in AI search [Updated for 2026]",
"image" : [ "https://amplistory.com/hubfs/Blog%20images/Blog%20banners%20(1)-2.png" ],
"mainEntityOfPage" : {
"@id" : "https://amplistory.com/blog/original-research-seo-ai",
"@type" : "WebPage"
},
"publisher" : {
"@type" : "Organization",
"logo" : {
"@type" : "ImageObject",
"url" : "https://amplistory.com/hubfs/Logo%20files/AmpliStory%20Logo%20(9).png"
},
"name" : "AmpliStory"
}
}
```
```json
{
"@context" : "https://schema.org",
"@id" : "https://amplistory.com/#website",
"@type" : "WebSite",
"name" : "AmpliStory",
"url" : "https://amplistory.com/"
}
```
```json
{
"@context" : "https://schema.org",
"@id" : "https://amplistory.com/#organisation",
"@type" : "Organization",
"name" : "AmpliStory",
"url" : "https://amplistory.com/"
}
```
```json
{
"@context" : "https://schema.org",
"@id" : "https://amplistory.com/about-us#grace-windsor",
"@type" : "Person",
"name" : "Grace Windsor",
"url" : "https://amplistory.com/about-us"
}
```
```json
{
"@context" : "https://schema.org",
"@id" : "https://amplistory.com/blog/original-research-seo-ai#webpage",
"@type" : "WebPage",
"description" : "What does current evidence say about original research, AI citations, SEO, freshness and AEO/GEO? An updated 2026 guide to content worth citing.",
"isPartOf" : {
"@id" : "https://amplistory.com/#website"
},
"name" : "Original research and AI search: what actually gets cited in 2026",
"primaryImageOfPage" : {
"@id" : "https://amplistory.com/blog/original-research-seo-ai#primaryimage"
},
"url" : "https://amplistory.com/blog/original-research-seo-ai"
}
```
```json
{
"@context" : "https://schema.org",
"@id" : "https://amplistory.com/blog/original-research-seo-ai#primaryimage",
"@type" : "ImageObject",
"url" : "https://144074470.fs1.hubspotusercontent-eu1.net/hubfs/144074470/Blog%20images/Blog%20banners%20(1)-2.png"
}
```
```json
{
"@context" : "https://schema.org",
"@id" : "https://amplistory.com/blog/original-research-seo-ai#article",
"@type" : "Article",
"articleSection" : "Marketing Strategy",
"author" : {
"@id" : "https://amplistory.com/about-us#grace-windsor"
},
"dateModified" : "2026-09-23",
"datePublished" : "2025-07-23",
"description" : "What does current evidence say about original research, AI citations, SEO, freshness and AEO/GEO? An updated 2026 guide to content worth citing.",
"headline" : "Original research and AI search: what actually gets cited in 2026",
"image" : {
"@id" : "https://amplistory.com/blog/original-research-seo-ai#primaryimage"
},
"inLanguage" : "en-IE",
"mainEntityOfPage" : {
"@id" : "https://amplistory.com/blog/original-research-seo-ai#webpage"
},
"publisher" : {
"@id" : "https://amplistory.com/#organisation"
},
"url" : "https://amplistory.com/blog/original-research-seo-ai"
}
```