How to Check If AI Recommends Your Brand (Free Baseline Test)

You cannot improve what you have not measured, and one question in ChatGPT is not a measurement. Answers vary nearly every run, so a baseline needs repetition. This takes about an hour and costs nothing.

Step 1: Pick prompts across the buyer journey#

• Problem-aware (3–5 prompts): the pain, no product named. “How do small agencies stop chasing late invoices?”
• Solution-aware (3–5): the category. “Best invoicing software for design agencies.”
• Comparison (2–4): you vs a rival, or alternatives. “FreshBooks alternatives for agencies that bill in EUR.”
• Branded (2–3): your name. “Is LedgerLoop good for a five-person agency?” This checks accuracy, not discovery.
• Constraint-heavy (2–3): real context. “Invoicing tool under $30 a month that integrates with Xero and handles retainers.”

Write prompts the way people talk to assistants: full sentences, with their situation. Keyword fragments like “invoicing software agency” test something no one types into ChatGPT.

Step 2: Control the test conditions#

AI answers are personalized more than search results. OpenAI's help page says ChatGPT collects general location information based on your IP address, and may use relevant memories to refine a search query if Memory is on. Your own history with your brand can make you look more visible than you are.

• Use clean sessions. Run logged-out where the engine allows it, or in a fresh private window. If you must stay logged in, turn off memory and personalization and use a temporary chat.
• Start a new chat for every run. Earlier turns in a thread steer later answers.
• Keep location constant. Run from the market you sell to. If you sell in several countries, treat each as its own baseline.
• Note the mode. Record whether web search was on. A ChatGPT answer with search and one without come from different paths (live retrieval vs training data).
• Record the date. Engines update models and indexes often. A baseline without dates cannot be compared later.

Step 3: Run and log#

Run each prompt at least three times in at least two engines. Good pairs: ChatGPT plus Google AI Mode or AI Overviews (the biggest reach), or ChatGPT plus Perplexity (citation-heavy). Twenty prompts × three runs × two engines gives you 120 answers. Log each one as a row.

Step 4: Calculate your baseline#

• Appearance rate = runs where you were mentioned ÷ total runs.
• Citation rate = runs where your domain was linked ÷ total runs.
• Top competitor rate = the same calculation for your strongest rival.

Do not over-read small numbers. With 120 answers, a swing of a few points is noise. Module 2 shows how to size samples and read the margin of error. For now, the baseline answers three questions: are you in the conversation at all, who owns it, and which sources feed it.