
How ChatGPT Chooses Which Brands to Recommend
Reverse engineering AI citations: what data reveals about LLM selection criteria. Reddit, Wikipedia, answer capsules, and proprietary data.
You've been optimizing your site for Google for years. Backlinks, keywords, meta tags — you know the recipe. But when a user asks ChatGPT "What's the best project management tool?", your brand appears nowhere.
Why?
Because generative search engines don't work like Google. They don't rank pages. They cite sources. And the selection criteria are radically different.
“LLMs don't look for the best keyword-optimized content. They look for the most credible and easily extractable sources.”
Most Cited Sources by LLMs: The Reign of Reddit and Wikipedia
A Semrush analysis of over 150,000 LLM citations reveals a surprising hierarchy:
| Source | % of Citations |
|---|---|
| 40.1% | |
| Wikipedia | 26.3% |
| .gov/.edu Sites | ~15% |
| Other Sources | ~18% |
This data is striking. LLMs don't necessarily favor the best-optimized corporate sites. They favor sources perceived as authentic, community-driven, and verifiable.
What if you can't have a Wikipedia page?
Let's be realistic: Wikipedia's notability criteria are strict, and 95% of companies don't qualify. But there are viable alternatives:
- Wikidata - Less restrictive criteria, used by LLMs to enrich their entity understanding
- Sources cited by Wikipedia - Get mentioned in publications Wikipedia already cites
- Google Knowledge Graph - Optimize via Google Business and structured data
- Crunchbase, LinkedIn - These entity signals help LLMs "recognize" you
Factor #1: Content Freshness
“Content less than 3 months old is 3 times more likely to be cited by LLMs.”
LLMs systematically favor fresh information. Unlike traditional SEO where an evergreen article can perform for years, generative AI optimization requires a continuous update strategy.
Practically, this means:
- Update your strategic pages at least quarterly
- Add recent data, current statistics
- Clearly display the last update date
⚠️ A 2022 article, even if it ranks #1 on Google, becomes progressively invisible to LLMs. Freshness is no longer an advantage — it's a condition for existence.
Answer Capsules: The Secret of the Most Cited Pages
A Search Engine Land analysis of sites generating nearly 2 million monthly organic sessions identified the structural factor most correlated with ChatGPT citations: answer capsules.
What is an Answer Capsule?
A concise text block (40-60 words) placed directly after an H2 heading phrased as a question. This block provides a direct, self-sufficient answer that the LLM can extract and cite easily.
H2
What is GEO?
GEO (Generative Engine Optimization) is the practice of optimizing content to be cited by AI search engines like ChatGPT, Claude, and Perplexity. Unlike traditional SEO, GEO aims for citation, not ranking.
Characteristics of high-performing answer capsules
- Optimal length - 40 to 60 words maximum
- No links - over 90% of cited capsules have no links
- Direct answer - no contextual introduction
ℹ️ Important: The absence of links in the capsule doesn't mean abandoning internal linking. Place your links in the following paragraphs, not in the initial answer block.
Proprietary Data: The Second Most Powerful Factor
After answer capsules, the factor most correlated with AI citations is the presence of original or proprietary data.
LLMs value what they can't find elsewhere:
- Original studies and research
- Statistics from your own data
- Exclusive benchmarks
- Documented field insights
This is a paradigm shift. In traditional SEO, you could perform by aggregating and rephrasing existing information. In GEO, LLMs seek uniqueness. They want to cite the primary source, not the summary.
Content Structure: Lists, Tables, and Clear Hierarchy
LLMs don't "read" like humans. They parse content through pattern matching and semantic extraction. Structure directly determines citation probability.
| Format | Citation Impact |
|---|---|
| Listicles | 50% of top AI citations |
| Content with tables | 2.5x more citations |
| Long-form (2000+ words) | 3x more citations |
| Content with FAQ Schema | +140% ChatGPT citations |
Structural elements that improve extraction
- Bullet and numbered lists
- Data tables
- Clear heading hierarchy (H2, H3)
- Structured FAQs
- Numbered steps for tutorials
Mentions and Citations: The New Currency of Authority
“Mentions and citations matter more than link type.”
In traditional SEO, the follow/nofollow distinction and anchor text were critically important. In GEO, it's the context of the mention that matters most.
What LLMs value
- Mentions in respected publications
- Citations by third-party sources (review sites, media)
- Presence in "best of" lists
- Recommendations on Reddit and specialized forums
From link building to mention building
Traditional link building strategy must evolve into a mention building strategy:
- Press relations to generate media mentions
- Presence on review sites (G2, Capterra, Trustpilot)
- Authentic participation on Reddit and communities
- Publishing original research that will be naturally cited
Each LLM Has Its Preferences: Understanding the Differences
Not all LLMs cite the same way. Understanding their "personalities" allows you to adapt your strategy.
| LLM | Citation Preferences |
|---|---|
| ChatGPT |
|
| Google AI Overviews |
|
| Perplexity |
|
| Claude |
|
This divergence means optimizing for a single LLM isn't enough. A robust GEO strategy must cover common fundamentals while accounting for each platform's specificities.
Community Signals: The New Backlink
“Mentions on Reddit and YouTube now rival traditional backlinks in influencing AI citations.”
LLMs are trained on enormous corpora including Reddit discussions, YouTube comments, and specialized forums. When your brand is mentioned positively in these spaces, it gains "LLM credibility".
Community amplification strategies
- Authentic participation on Reddit (no promotional spam)
- Encouraging customer reviews on relevant platforms
- Creating content that naturally generates discussions
- Monitoring brand sentiment on community platforms
What You Can Do Right Now
In summary, here are the priority actions to improve your chances of being cited by LLMs:
- 1Audit freshness
Identify your strategic pages and establish a quarterly update schedule.
- 2Create answer capsules
For each key question, write a 40-60 word answer, without links, placed after the H2.
- 3Invest in proprietary data
Publish studies, benchmarks, exclusive insights.
- 4Structure for extraction
Use lists, tables, FAQ schema, clear heading hierarchy.
- 5Build your entity presence
If Wikipedia isn't accessible, target Wikidata, Knowledge Graph, authority sites.
- 6Activate Reddit and communities
Authentically participate in discussions in your industry.
- 7Develop a mention strategy
Target third-party publications, review sites, "best of" lists.
- 8Break down silos
Create bridges between SEO, PR, social, and brand.
Conclusion: The Game Has Changed
The question is no longer "Do I rank on Google?" It has become: "Does AI cite me?"
Brands that understand this evolution and adapt their strategy now will be 2 to 3 years ahead of their competitors. GEO (Generative Engine Optimization) is not a passing trend — it's the new reality of brand discovery.
The only question that remains: will your brand be visible when they search?
Frequently Asked Questions
Sources and References
- Semrush - Analysis of 150,000+ LLM citations — semrush.com
- Search Engine Land - Content traits LLMs quote most — searchengineland.com
- AirOps Webinar - Kevin Indig, State of AI Search 2026 — airops.com
- Wellows - 7K Queries, 485K Citations Analysis — wellows.com
- Cornell University / Ethinos - GEO Research — ethinos.com
Methodology
This article is based on analysis of hundreds of AI responses and data from our GEO audit tool.
Ready to improve your AI visibility?
Let's discuss your GEO strategy
