Where Can I Find Real-World GEO Case Studies, Especially for Small Sites?
Real-world GEO case studies are hard to find, and many that exist come from vendors or agencies promoting their services. The most reliable evidence comes from four places. There are large-scale studies of what AI answers cite, published by analytics and SEO companies. A few businesses have published their own data. Academic research tests techniques under controlled conditions. And you can run a small, controlled test on your own site. For small sites specifically, the encouraging evidence is that AI answers often cite pages outside the top search results. So a small site with a clear, specific page can be cited without dominating rankings.
Where can you find credible evidence?
| Source | What it offers | Watch out for |
|---|---|---|
| Large-scale citation studies | Patterns across millions of answers, such as which sources are cited and how often | Usually published by tool vendors; patterns, not playbooks |
| Businesses publishing their own data | Specific numbers from one site, such as referral traffic and conversions | Small samples, one market |
| Academic research | Controlled tests of content changes | Benchmarks may not match live engines |
| Industry conferences and newsletters | Practitioners sharing tests and results | Varies widely in rigour |
| Platform documentation | What Google, Bing and OpenAI say about their systems | Describes principles, not outcomes |
| Your own tests | Evidence for your market and site | Takes time and discipline |
What does the published evidence show?
- Academic research. The 2023 paper that named "generative engine optimization", later accepted at the KDD 2024 conference, tested content changes on a benchmark of queries. Some changes raised visibility in generative answers by up to 40%. It is a controlled benchmark rather than live engines, but it is the most-cited starting point.
- Citation patterns. Ahrefs' 2026 analysis of AI Overviews found only about 38% of citations came from the top 10 organic results. About 31% came from pages ranked 11 to 100, and about 31% from pages beyond the top 100. That leaves room for pages that do not rank highly.
- Earned media. Muck Rack's analysis of more than 25 million cited links found most came from earned sources, such as journalism, reference sites and third-party pages, rather than paid content.
- Business data. Ahrefs has reported that AI search sent about 0.5% of its visitors but 12.1% of its sign-ups. That is a single site, but a real one.
What are real examples of websites getting cited by AI search engines?
- A small local florist. A Portland florist published its own test of 208 recommendations from ChatGPT, Gemini and Claude across four US cities. It found local florists, including itself, ranked first in every answer that named a city. National order-gathering sites dominated only when no city was given. It is one of the clearest small-business examples published so far.
- Specialist pages below the top results. The Ahrefs figures above show that many AI Overview citations go to pages outside the top 10, and often outside the top 100. They are usually specific pages answering a narrow question well.
- Video. Ahrefs' 2026 analysis found YouTube accounted for a notable share of citations beyond the top 100 results. Small creators with useful videos can be cited too.
- Review and community pages feature heavily in many answers. That is why small businesses with well-described reviews often appear for local and niche questions.
Where do practitioners share results?
Beyond published studies, practitioners share tests in a few places worth following:
- Search conferences such as BrightonSEO, SMX and MozCon, where talks increasingly cover AI search tests, often with slides that include methods and data.
- Research newsletters and analyst blogs, some of which publish joint studies with tool vendors, such as the ghost citation analysis run with Semrush.
- Platform blogs: Google Search Central and Bing Webmaster Tools publish guidance and occasionally data about their AI features. Bing's AI Performance report announcements, for example, describe what site owners can now measure.
- Trade publications such as Search Engine Land and Search Engine Journal, which summarise new studies and sometimes run their own tests.
Treat talks and posts like any other case study. Look for the method, not just the headline.
What results should small sites expect?
Realistic expectations help. Small sites rarely win broad category questions. They are more likely to win specific questions that name a place, a niche or a precise need, where their page is one of the few that answers clearly. Gains are usually measured in a handful of extra mentions across dozens of runs, and in a few enquiries that mention AI. Traffic spikes are unusual.
How should you judge a case study?
Ask these questions before trusting a result:
- Which questions were tested, and were most of them unbranded? Branded questions inflate results.
- How many runs per question and engine? Single runs are unreliable.
- What was the baseline, and how long was the period?
- Was there a control, such as unchanged pages or unrelated questions?
- Which engines, locations and settings were used?
- What exactly changed, and was anything else changed at the same time?
- Did business outcomes move, such as leads, sales or sign-ups, or only visibility?
- Who published it, and do they sell a related service?
Spotting selective prompt sets is covered in how to detect cherry-picked prompts in an AI search case study.
How can a small site run its own case study?
A simple, credible design:
- Choose 20 questions your buyers ask, mostly unbranded, and split them into two matched groups.
- Take a baseline: run each question three times in three assistants for two weeks, recording mentions and cited URLs.
- Change one thing for the first group only, such as rewriting the relevant pages with direct answers and specific facts, or earning a listing in a cited directory.
- Leave the second group unchanged as a control.
- Measure for six to eight weeks with the same method.
- Compare the change in each group, and record business signals such as enquiries that mention AI.
Publishing the method and results, including null results, adds to the evidence that small sites need. Whether GEO is worth the effort at all is covered in whether GEO works or is a waste of time and money.
What might an illustrative small-site test look like?
Consider an illustrative independent bike shop with a small website. It tracks 20 questions, and rewrites its e-bike servicing and repair pages to state prices, turnaround times and the brands it services. It also adds its listing to a regional cycling directory that answers cite. The ten questions tied to those pages move from 2 mentions in 30 runs to 11 in 30 over eight weeks. The ten control questions stay flat. The shop also notices several customers mentioning ChatGPT at the counter. It is a small test, but the method makes it credible.
What do people ask about GEO case studies?
Why are there so few public GEO case studies?
AI answers change quickly and are hard to measure, attribution is difficult, and many results are kept private by agencies and clients.
Are vendor case studies worthless?
No, but read them critically. Check the questions, runs, baseline and control, and look for independent confirmation.
Can small sites really be cited?
Yes, especially for specific questions. Citation data shows many cited pages rank outside the top search results.
How long should a small-site test run?
At least six to eight weeks after the change, with a two-week baseline before it.
Published by The Kovmere Team · updated Sep 29, 2026 · More resources