"What are some universal truths you see from AI search / GEO audits?" I always avoided that question because every clients needs are different. But enough people bent my arm, so here goes:
We've done probably 100+ AI search audits for brands by now. Lost count honestly. But audits from last year look quite different from today given how much the LLMs have evolved and how brands are creating more GEO-specific content.
So I'll just focus on data from the last 90 days across 20+ projects for brands spanning audio, pharma, finance, semiconductors, insurance, energy, telecom, and real estate.
Here are some key takeaways / watch-outs:
Brand’s visibility can vary by 15+ percentage points across ChatGPT, Gemini, Claude, Perplexity for the same prompts in the same week. If you're just looking at blended averages, you're missing the truth. And one strategy for all models won't work. Focus on the prompts and LLMs that your users use most.
Why the huge variance? The models each have their favourite citation sources (or lack of). For example, Reddit was retrieved in 64% of Perplexity answers, 48% of Google AI Overview answers, but just 14% of ChatGPT answers. However, there are some constants in certain categories.
NIH.gov is the goat for many healthcare prompts no matter which global market, for example.
High visibility ≠ always positive. One brand in my dataset shows up in 58% of conversations but with a sentiment score of 45/100. Being talked about and being talked about well are not the same. We often see competitor brands cluster in their sentiment (positive or negative), so context is key as well.
While visibility varies, negative sentiment is surprisingly consistent. For example, a pharma brand scored (out of 100) 44, 45, 45, 45, 45, and 47 across ChatGPT, Perplexity, Google AI Overview, Gemini UI, Google AI Mode, and Claude Sonnet. The internet seems pretty consistent in its dissatisfaction.
Old brand names can refuse to die. A health company that rebranded years ago still has its old brand domain cited about 2x more than the new one. Wikipedia, industry rankings, awards, and long lead editorial coverage move slowly. You need a dedicated strategy to provide LLMs with a surround-sound of trusted signals for the new brand.
Ranking listicles are the top format type regardless of category. “Best [X] of 2026” articles are the single most retrieved URL type in every project I looked at. Headphones, AI chips, pharma treatments, mall directories — the LLMs simply love that format.
#GEO #AISearch #marketing #LLMs #ChatGPT #Claude #Gemini #Perplexity