AEO ScannerHow ChatGPT decides which brands to recommend
ChatGPT's brand-mention logic is downstream of three things: training data exposure, live web search citations, and prompt-time retrieval. Here's how each works.
When ChatGPT writes "the best project management tools are Asana, ClickUp, and Notion," the model didn't really pick those names. A retrieval system did. Understanding the layers underneath the answer is the difference between guessing at AEO and engineering it.
Three layers, in order
- Training data exposure. Whatever was on the open web before the model's cutoff. This is the base prior for "what brands exist."
- Live web search at prompt time. For most current ChatGPT models with browsing/search enabled, the model issues 1–4 web searches and incorporates results into the answer. This is where most of the actual brand selection happens.
- Reinforcement learning from feedback. Over time, RLHF nudges the model toward answers that get positive user feedback, which usually means well-known, well-reviewed brands.
The retrieval matters more than the training
It's tempting to think "I need to be in the training data." For most B2B categories you don't — your brand probably is in the training data already, just at low density. The decisive factor is what shows up in the live web search when a buyer asks ChatGPT a question.
That means: if a buyer asks "best CRM for SaaS startups," ChatGPT issues a search, pulls in 5–10 results, and synthesizes. Whichever brands appear in those 5–10 retrieval hits get mentioned in the answer.
Which sources show up in retrieval
Roughly in order of frequency across our scans:
- Authoritative "best of" listicles on high-DR sites (Forbes Advisor, TechRadar, G2)
- Vendor comparison pages (clickup.com/blog/X-vs-Y, asana.com/comparisons)
- Software directories (Capterra, G2, SaaS Genius, Product Hunt)
- Reddit threads and forum discussions
- The vendor's own homepage and product pages
The implication
If you want to appear in ChatGPT's recommendations, you need to be in those sources. Not just any sources — these specific sources. AEO Scanner exposes this for you: every scan reports the top domains LLMs actually cited for your category, so you know exactly which placements move the needle.
What about Claude and Perplexity?
Same architecture, slightly different retrieval defaults. Claude tends to favor authoritative reviews. Perplexity is the most aggressive at citing — every answer includes named URLs. The brands that win across all three usually have one thing in common: they're cited in the canonical listicle for their category.