Which prompts should you track for AI visibility?

SEO taught us to pick keywords by volume. GEO (Generative Engine Optimization) works differently. You optimise not for isolated search terms, but for full questions, prompts. And you choose those prompts not by volume, but by purchase intent.

Why prompts differ from keywords

In Google someone types "lawyer amsterdam". In ChatGPT that same person types: "I am looking for a lawyer in Amsterdam who specialises in employment law and is also reachable in the evening, who would you recommend?"

Prompts contain the full problem, the context (region, price, sector) and the desired form of answer. Tracking only "lawyer amsterdam" misses 90% of what AI users actually ask.

The 4 prompt categories you should track

1. Recommendation prompts (highest priority)

"Which [service] in [region] would you recommend for [audience]?": directly tied to revenue.

2. Comparison prompts

"What is the difference between [brand A] and [brand B]?": you lose deals here if only your competitor is named.

3. Problem prompts

"My [problem] keeps happening, how do I fix it?": early-funnel intent where being mentioned as a route to the solution is gold.

4. Local prompts

"Best [service] near [city]": AI models weigh local signals differently from Google Maps; track per region.

Which prompts NOT to track

• Just your brand name: you will always show up. No signal.
• Overly broad questions: "What is marketing?", 10,000 brands could answer.
• Pure facts: irrelevant to your business.

How many prompts should you track?

Start with 20–50 prompts spread across the four categories, then expand where the action is. In Briljant's Prompt Tracker we generate a starting set automatically from your website, then test each prompt 100x per cycle across ChatGPT, Gemini, Claude and Perplexity.