I tracked what ChatGPT, Perplexity, and Gemini cite for AEO queries. Small sites showed up more than expected.

To get cited by AI answer engines, publish a focused page that directly matches the query, answers the question upfront, and is structured for extraction: standalone sentences, tables, lists, and clear comparisons.

That was the strongest pattern from this first small AEO/GEO citation test.

The surprising part: in this sample, small specialist sites appeared next to major domains such as HubSpot, Conductor, Search Engine Land, and Coursera. Domain authority may still matter, but it was not the only thing deciding who showed up.

Study details

This is an early exploratory test, not a definitive ranking study.

I kept reading AEO advice that felt like theory. So I ran real queries, recorded every visible source, and looked for patterns in what the engines actually cited.

What I tested

The test used queries people might actually search while trying to understand AI visibility:

• "best AEO tools 2026"
• "how to get cited by ChatGPT"
• "AEO vs SEO 2026"
• "AEO vs GEO 2026"

The engines and modes mattered. Perplexity cited sources consistently. ChatGPT only surfaced sources when search or deeper reasoning was involved. Gemini's faster model returned a rich answer in this test, but did not surface visible citations.

The citation data

The first thing that stood out: small specialist sites appeared as often as large, familiar domains in this sample.

This does not prove that domain authority does not matter. The sample is too small for that. But it does suggest that AI citation visibility is not reserved only for giant websites. Focus, structure, and query match appear to matter a lot.

Evidence example

In one Perplexity result for "AEO vs SEO 2026," the answer cited small specialist sites inline, including digitalotters, lasso-up, width.ai, and devexhub.

One nuance: the footer showed that Perplexity drew on more sources than it cited beside each claim. That means the visible citation is not always the full retrieval set. A page can influence an answer without being shown beside every sentence.

Finding 1: The cited pages matched the query closely

The cited pages were usually dedicated pages built around the exact query or intent.

They were not broad marketing pages that happened to mention AEO once. They were pages with titles, headings, and URLs that clearly matched the user's question:

• best AEO tools
• how to get cited by ChatGPT
• AEO vs SEO
• AEO vs GEO

This matters because AI answer engines need extractable, query-specific material. A focused page gives the engine a clean answer to retrieve, summarize, and cite.

What to do

Create one focused page for each query cluster you want to win. Do not rely on one broad guide to rank for every related AI visibility question.

For example:

• /best-aeo-tools
• /how-to-get-cited-by-chatgpt
• /aeo-vs-seo
• /aeo-vs-geo

Each page should answer one search intent directly.

Finding 2: Small sites can get cited beside major domains

In this sample, small and niche sites appeared beside HubSpot, Conductor, Search Engine Land, Coursera, Google's developer documentation, and the original arXiv GEO paper.

Examples of smaller or specialist sources included:

• cite.sh
• getairefs.com
• aiclicks.io
• digitalotters.com
• lasso-up.com
• width.ai
• devexhub.com

For small brands, this is the most useful signal from the test: AI citations may be more page-level than brand-level, especially for emerging topics where the web corpus is still thin.

That does not mean authority is irrelevant. It means a small site with the clearest answer may still be eligible for citation.

What to do

If your site is small, do not try to win with generic thought leadership. Build the most specific, useful, extractable page for a query the engine needs to answer.

Finding 3: Format made pages easier to extract

The cited pages tended to use structures that AI engines can reuse easily:

• direct answers near the top
• comparison tables
• short lists
• clear definitions
• question-shaped headings
• standalone sentences
• examples that can be quoted without extra context

This is important because answer engines do not just need "good writing." They need passages that can be pulled into an answer with minimal cleanup.

Dense paragraph walls are harder to extract. Clear, modular sections are easier.

What to do

Structure pages so each section can stand alone.

Instead of writing:

AEO is an emerging discipline with several overlapping definitions across search, content, and AI interfaces...

Write:

AEO is the practice of structuring content so AI answer engines can find it, understand it, and cite it in response to user questions.

Then support that answer with a table, examples, and FAQs.

Finding 4: Citation behavior depends on the model and mode

This was one of the clearest practical lessons.

The same general topic produced different citation behavior depending on the engine, model, and mode.

In this test:

• Perplexity surfaced citations on every query.
• ChatGPT without search surfaced no visible sources.
• ChatGPT with search or deeper reasoning surfaced citations.
• Gemini 3.5 Flash produced a rich answer but did not surface visible sources.

So "ChatGPT did not cite me" is not enough information. The better question is:

Which model, mode, query, region, and search behavior were involved?

An answer generated from live retrieval is a different citation environment from an answer generated from model memory.

What to do

When testing AI visibility, record the exact engine and mode:

• engine
• model
• search on/off
• date
• query wording
• visible citations
• screenshots

Do not treat a no-citation answer from a non-search mode as proof that your page cannot be cited.

Citation vs. memory: the distinction most people miss

There are two different AI visibility games.

Citation is retrieval-based. The engine searches or retrieves sources, writes an answer from those sources, and shows links. This is the easiest layer to measure.

Memory is model-based. The model mentions a brand or concept because it has learned about it from the training corpus. It may not show a source, because it is not retrieving one at answer time.

If you are starting out, play the citation game first.

Citation is page-level, measurable, and achievable now. Memory is a slower brand-ubiquity game. It comes from being mentioned consistently across the web over time.

One nuance: visible citation is not the same as total influence. A page can shape an answer without being shown as a visible citation. It may have been retrieved but not surfaced, or absorbed into model behavior during training.

Still, visible citations are the best measurable starting point.

Finding 5: Deeper models may prefer primary sources

The ChatGPT thinking-mode result cited a different kind of source:

• Coursera
• the original arXiv GEO paper
• Google's developer documentation

That is different from the pattern in Perplexity, where small specialist blogs appeared more often in this sample.

This is only one data point, so it should be treated as a hypothesis, not a conclusion. But it suggests a useful distinction:

• focused specialist content may help with retrieval-based answers today
• original research and primary sources may be more attractive to deeper reasoning models

For creators and brands, the implication is good: original research can serve both layers. It gives retrieval engines something specific to cite, and it gives deeper models a primary source to prefer.

The playbook: how to make a page easier for AI engines to cite

Based on this small test, the practical playbook is:

1. Build one focused page per query or query cluster.
2. Put the direct answer at the top.
3. Use tables, lists, definitions, and comparison sections.
4. Write standalone sentences that can be quoted without extra context.
5. Add original data, examples, or screenshots where possible.
6. Include FAQs that match real user questions.
7. Add structured data when appropriate.
8. Keep the page updated with a visible date.
9. Test in engines that show citations, especially Perplexity and ChatGPT with search.
10. Record the query, model, date, answer, citations, and screenshots.

The goal is not to "trick" an AI engine. The goal is to make your page the clearest, most useful source for a specific question.

Limits of this test

This is a small exploratory sample.

The findings may change with:

• more queries
• different industries
• different locations
• personalization
• model updates
• search availability
• browser/session state
• query wording
• paid vs. free model tiers

The findings should be read as early signals, not universal rules.

The stronger claim is not "small sites always win." The stronger claim is:

In this sample, AI answer engines cited small specialist pages when those pages matched the query closely and were structured for extraction.

That is worth testing at larger scale.

How to replicate this test

If you want to test your own category, use the same basic process:

1. Pick 5 to 10 commercial or informational queries in your niche.
2. Run each query in Perplexity, ChatGPT with search, and Gemini.
3. Save the answer and screenshot.
4. Record every visible citation.
5. Classify each cited source by page type, domain type, format, and intent match.
6. Compare cited pages against pages that did not appear.
7. Repeat after publishing or updating your own pages.

The useful part is not one isolated answer. The useful part is the pattern across repeated queries.

FAQ

Does ChatGPT cite sources?

ChatGPT cites sources when it retrieves information at answer time, such as through live search or connected documents. When it answers from model memory, it may not show sources.

No visible citation does not mean your content had no influence. It only means the answer did not surface your page as a visible source.

Why does Perplexity cite so often?

Perplexity is designed as an answer engine that retrieves web sources and writes answers from them. Because retrieval is central to the product, visible citations appear more consistently.

Do you need a big domain to get cited by AI?

Not always. In this sample, small specialist sites appeared beside major domains. A focused, well-structured page that matches the query can be competitive, especially in newer topics where the corpus is still forming.

Is AEO the same as GEO?

AEO, or Answer Engine Optimization, and GEO, or Generative Engine Optimization, are often used to describe the same practical goal: making content easier for AI answer engines to find, understand, cite, and recommend.

The terms differ, but the work overlaps heavily.

What should a brand do first?

Start with a citation baseline.

Pick the queries where you want your brand to appear. Test those queries across AI answer engines. Record who gets cited, what page type wins, and what format the engines reuse.

Then build or improve pages that match those patterns.

---

Testing your own category? Get in touch — or follow along on X @eg21127b.