How to tell if ChatGPT recommends your brand (a 5-step method)

Before investing a single peso in AI search visibility, you need a baseline: what does each engine answer today when someone asks about your category? This is the same method ia¹ uses in its assessments, and you can run it yourself in an afternoon.

# Step 1 · Define the critical prompts for your category

Don’t ask about your brand—ask the way your customer does before they know you. Three prompt types: recommendation (“Which X company would you recommend in Mexico?”), comparison (“What are the best X options in Guadalajara?”), and problem (“I need to solve Y. Who can help me?”). Write 10-15 in your buyer’s natural language, including city variations.

# Step 2 · Run the prompts in the 5 engines under clean conditions

ChatGPT, Perplexity, Gemini, Claude, and Google with AI Overviews. Three rules to avoid contaminating the results: use sessions with no history (prior context biases responses), don’t use an account you’re logged into if you’ve discussed your brand with the assistant, and run each prompt twice on different days—responses vary, and you’re interested in the trend, not a single snapshot.

# Step 3 · Record mentions, order, and sentiment

For each prompt and engine, record the following in a spreadsheet: the brands mentioned and their order, whether your brand appears, how it describes you (is the description accurate, outdated, or confused with another brand?), and which attributes it highlights about cited competitors. Order matters: the first mention gets the largest share of the decision.

# Step 4 · Identify the sources behind each answer

Perplexity and AI Overviews cite their sources explicitly; ChatGPT with search also links to sources. That list is strategic gold: these are the pages the engine considers authoritative in your category. If your competitors appear and you don’t, review which sources mention them and which could mention you—that’s your external authority plan.

# Step 5 · Turn the baseline into a monthly series

A one-time measurement isn’t useful for making decisions: models vary across versions and days. Repeat the exercise every month using the same prompts and record the mention rate for each engine. After three months of data, you can distinguish a trend from noise—and attribute the impact of what you implement (schema, content, authority) to changes in the curve.

# Sources cited

• Aggarwal et al. · GEO: Generative Engine Optimization (KDD 2024) · Measurement framework for visibility in generated answers, from which the prompt-level citation metric is derived.