--- title: "How to pick your first prompts to track for AI visibility" description: "Choose the right prompts to track in ChatGPT, Google AI Overviews and Perplexity: five prompt types, where to find real buyer questions, and a 50-prompt starter set." url: https://getnodded.com/blog/how-to-choose-prompts-to-track/ category: "Getting Started" author: "Get Nodded Team" published: 2026-10-01 updated: 2026-10-12 image: https://getnodded.com/blog/covers/how-to-choose-prompts-to-track.png --- # How to pick your first prompts to track for AI visibility Prompts are the questions you track across AI engines, and they decide what your AI visibility numbers mean. Track the wrong ones and you'll get a flattering score that doesn't match revenue. Here's how to pick a first set that reflects how your buyers actually use AI. ## What makes a good prompt A prompt worth tracking passes three tests: - A real buyer would ask it, in their words, not yours. - The answer could plausibly name you or a competitor. - Winning it would matter: it sits close to a buying decision or a high-value problem. ## The five prompt types to cover A balanced first set mixes these types, so one strong area doesn't hide a weak one. | Type | Example | Share of your set | | --- | --- | --- | | Category | “Best project management tool for agencies” | About a third | | Problem | “How do I stop clients missing approval deadlines?” | About a quarter | | Comparison | “Which AI visibility tool is best for a 20-person team?” | About a fifth | | Alternatives | “Cheaper alternatives to Salesforce” | About a tenth | | Brand | “Is [your brand] good for enterprise?” | A few, to check how AI describes you | ## Where to find real buyer questions - Sales call notes and demo requests: the exact phrases prospects use. - Support tickets and onboarding questions. - Search Console queries, especially long, question-shaped ones. - Reddit threads and review sites in your category. - Your competitors' comparison pages, which show the matchups buyers search for. ## Turn one use case into a prompt cluster The fastest way to build a set is to take one use case and write a prompt of every type around it. Say you sell scheduling software for dental clinics. One use case becomes a small cluster: - Category: “What's the best appointment scheduling software for a small dental practice?” - Problem: “How can my dental clinic reduce no-shows without calling every patient?” - Comparison: “Which dental scheduling tools work with practice management software?” - Alternatives: “What are cheaper alternatives to [market leader] for dental clinics?” - Brand: “Is [your brand] a good fit for a three-chair dental practice?” ## A 50-prompt starter set Fifty prompts is enough for a reliable baseline and fits Get Nodded's Starter and Basic plans. A practical split: 15 category prompts across your main use cases, 12 problem prompts, 10 comparisons against your top three or four competitors, 5 alternatives prompts, 5 brand prompts, and 3 prompts for your most important integration or niche. Write them the way people talk to AI: full questions with context, such as team size, industry or budget, not two-word keywords. ## How many prompts each plan covers Repeat the cluster for your three or four main use cases and you already have 20 to 25 prompts that cover the buying journey, from the first problem to the final shortlist. How far you can take it depends on your plan: | Plan | Prompts | AI engines | Checks a day | | --- | --- | --- | --- | | Starter | 50 | 2: ChatGPT, Google AI Overviews | 1 | | Basic | 50 | 5: adds Google AI Mode, Perplexity, Gemini | 1 | | Pro | 100, across 2 brands | 6: adds Copilot | 2 | | Business | 150, across 5 brands | 6 | 4 | See current plans: https://getnodded.com/pricing/ ## Organize with tags Tag every prompt by type, product line and funnel stage. Tags let you filter visibility and share of voice, so you can say “we win category prompts but lose comparisons” instead of reporting one blended number. ## Tags that make your reports useful A handful of consistent tags turns a list of prompts into a report you can act on. Most teams start with four kinds: - Prompt type: category, problem, comparison, alternatives and brand, so you can see which kind of question you lose. - Product or use case: one tag per product line, so a strong product doesn't hide a weak one. - Funnel stage: early research, shortlist and final decision, to see whether AI drops you just before buyers choose. - Audience or region: team size, industry or market, when you sell to very different buyers. ## Common mistakes when choosing prompts - Tracking keywords instead of questions. “CRM software” tells you little; “best CRM for a 10-person real estate team” mirrors how people ask ChatGPT and Perplexity. - Too many brand prompts. AI almost always mentions a brand you name, so they inflate your visibility score. - Only tracking prompts you already win. A good set shows gaps, because that's where the work is. - Mixing two intents in one prompt, such as “best and cheapest CRM with good support”. Split it so you know which part you win. - Rotating prompts every week. Constant changes break your trend lines, so add new prompts but keep a stable core. - Ignoring differences between engines. A prompt you win in ChatGPT can still lose in Google AI Overviews or Copilot, so check results per engine. ## Review and refine after two weeks After a baseline period, drop prompts no engine answers with brand names, split prompts that mix two intents, and add the questions you found in citations and ChatGPT query fan-out. Keep the core set stable so trends stay comparable. ## FAQ ### How many prompts should I track? Start with 50 for one brand. That's enough to see patterns by type and engine without spreading attention thin. Grow to 100 or 150 as you add products, regions or brands. ### Should I include my brand name in prompts? Only in a few brand prompts that check how AI describes you. Most prompts should be unbranded, since that's where buyers discover options and where AI chooses who to recommend. ### How often should I change my tracked prompts? Keep a stable core for at least a quarter so your trends mean something. Add prompts as you find new questions in sales calls, citations and query fan-out, and retire only the ones that never produce brand names. ### Should prompts be long or short? Write them the way people talk to AI assistants: full questions with context such as team size, industry or budget. Two-word keywords are a search engine habit and produce answers that don't match real conversations. ### Do I need different prompts for each AI engine? No. Track the same prompts across every engine so the results are comparable. The differences between ChatGPT, Google AI Overviews, AI Mode, Perplexity, Gemini and Copilot are part of what you're measuring.