If you have ever tried to check how your brand shows up in ChatGPT or Perplexity by typing in one quick question, you already know the problem. One prompt tells you almost nothing. AI answers change depending on how a question is phrased, who is asking, and what stage of the buying journey they are in. If you want a real picture of your brand's AI visibility, you need a real set of prompts, written with intention.
Below is a practical, step by step guide to writing prompts that actually tell you something useful, along with examples you can copy and adapt for your own brand.
1. Start with your brand name, not your category
Before you test anything else, ask the AI directly about your brand. This tells you what the model already "knows" and whether that information is accurate.
Try prompts like:
- "What is [Brand]?"
- "What does [Brand] do?"
- "Is [Brand] a good brand?"
- "Who founded [Brand] and when?"
This step matters because it catches the easy mistakes first. Wrong founding year, outdated pricing, a headquarters that moved three years ago. If the model gets basic facts wrong here, nothing downstream will look right either. Think of this as your accuracy baseline before you move on to anything about market position or competitors.
2. Write prompts the way a real buyer would type them, not the way a marketer would
This is the single biggest mistake people make when testing AI visibility. They write formal, keyword-stuffed prompts because that is what they are used to from SEO. But nobody actually types "best enterprise HVAC repair service provider near me for commercial buildings" into ChatGPT. They type something closer to "my AC broke, who do I call near me."
Before you write a single prompt, picture an actual customer sitting at their kitchen table, phone in hand, half frustrated. What would they type? Short, casual, sometimes with a typo, sometimes with only half the context spelled out. That is the version worth testing.
3. Split your prompts by who is asking, not just what they're asking
Annapurna Influence breaks this down well in their piece on AI visibility strategy. Not every buyer is at the same stage, so a single style of prompt will never capture the full picture. Their framework groups people into three types, and it is a genuinely useful lens for building out a prompt list:
The brand aware researcher(Type A). They already know your name and want reassurance before they commit. Example prompt: "Is Brand A a good brand?" or "Brand A reviews."
The solution seeker(Type B). They know the fix they need, just not who to call. Example: "HVAC near me" or "best project management software for a 10 person team."
The problem aware, solution unaware user(Type C). They know something is wrong but haven't connected it to a category yet. Example: "why does my kitchen sink smell bad" instead of "plumber near me."
Write at least a handful of prompts for each of these three groups. If your whole prompt list is made up of branded questions, you are only measuring accuracy, not real visibility. The problem aware prompts are usually where brands get caught most off guard, because almost nobody tests for them.
4. Add location when location actually changes the answer
If your business serves a specific city, region, or service area, your prompt list needs to reflect that. A national brand and a local plumber are competing for completely different prompts, and an AI assistant handling a "near me" style question will lean heavily toward locally relevant answers.
Test both versions of the same question:
- Generic: "best marketing agency for small businesses"
- Localized: "best marketing agency for small businesses in Austin"
You will often see different brands surface in each one. If you only test the generic version, you might think you're invisible when really you're just missing from the geo-specific results that matter most to your actual customers.
5. Include comparison prompts, even the uncomfortable ones
Most brands are happy to see how they answer "tell me about Brand A" but avoid asking "Brand A vs Competitor B." That avoidance is exactly why this step matters. If buyers are asking that question anyway, you need to know what the AI says, whether you're mentioned at all, and how you're framed when you are.
Prompts to try:
- "[Your Brand] vs [Main Competitor]"
- "Alternatives to [Main Competitor]"
- "Best [your category] competitors"
Don't skip the ones where you're not the subject. Testing "Competitor A vs Competitor B" tells you whether your brand even enters the conversation when you're not directly named. If it doesn't, that's useful information too.
6. Ask the problem first, before the product
This one connects back to Annapurna Influence's framework, and it deserves its own step because it's so often overlooked. Before someone knows they need a plumber, they know their sink smells bad. Before someone knows they need project management software, they know their team keeps missing deadlines.
Write prompts that start from a symptom rather than a solution:
- "Why is my dishwasher making a grinding noise"
- "My team keeps missing project deadlines, what's going wrong"
- "My car pulls to the left when I brake"
If your brand publishes content that answers these problem stage questions well, you have a real chance of getting cited before a competitor's name ever enters the conversation. Testing these prompts tells you whether that content exists and whether it's working.
7. Run the same prompt across more than one AI engine
An answer in ChatGPT is not the same as an answer in Perplexity, Gemini, or Google's AI Overviews. Each one pulls from different sources and weighs them differently. Perplexity tends to lean heavily on citations. Gemini leans on Google's own index. If you only check one engine, you're getting a partial view at best.
At minimum, test the same prompt in ChatGPT, Perplexity, and one Google AI surface (AI Mode or AI Overviews). If your buyers skew technical, add Claude to the mix. You might find you're doing well in one engine and completely absent in another, and that gap is exactly the kind of thing a single-engine check would never catch.
8. Run every prompt twice before you trust the answer
AI answers are not fixed. Ask the same question twice, even minutes apart, and you can get a noticeably different response. One good answer might be a fluke. A pattern across two or three runs, on different days, tells you something real.
Keep it simple. Log the date, the exact prompt wording, the engine, and whether your brand showed up, and where in the answer it appeared. A basic spreadsheet works fine for this. The goal isn't to overbuild the tracking system, it's to make sure you're not drawing conclusions from a single lucky or unlucky response.
9. Track where you appear, not just whether you appear
Being mentioned third in a list of five is a very different result than being named first, or being the only brand mentioned at all. When you log your results, note the position, not just a yes or no.
A simple scale works well here:
- First or only brand mentioned
- Mentioned, but after competitors
- Mentioned briefly, buried near the end
- Not mentioned at all
Over time, this position tracking tells you a lot more than a raw visibility count. A brand mentioned in every answer but always last is a very different problem than a brand missing from half its prompts entirely.
10. Build a reusable prompt list, and revisit it on a schedule
The point of all this isn't a one time check. AI models update, competitors publish new content, and your own pricing or positioning changes. A prompt list you build once and never touch again goes stale fast.
Save your prompt list somewhere you'll actually reopen it. Tag each prompt by category (branded, comparison, problem aware, and so on) so you can rerun the same set every month or quarter and compare results over time. The value here comes from watching the trend, not from any single snapshot.
A quick example, put together
Say you run a regional home services company. A well built prompt list might look like this:
- "What is [Brand]?" (accuracy check)
- "Is [Brand] a good HVAC company?" (brand aware)
- "HVAC repair near [City]" (solution seeker, localized)
- "My AC is blowing warm air, what's wrong?" (problem aware)
- "[Brand] vs [Local Competitor]" (comparison)
- "Best HVAC companies in [City] 2026" (transactional)
Run each of these across two or three AI engines, twice, on different days. That's a small set, twelve to eighteen data points, but it already tells you far more than typing your brand name into ChatGPT once and calling it a day.
The bigger picture
Writing better prompts isn't really about the prompts themselves. It's about thinking like your buyer at every stage of their decision, from the moment they notice a problem to the moment they're comparing you against a competitor by name. Brands that only check what AI says about them by name are missing most of the picture. The real opportunity, the one companies like Annapurna Influence build their entire content strategy around, is showing up earlier, at the problem stage, before a buyer even knows what to search for.
Start small. Pick ten to fifteen prompts across the categories above, run them consistently, and build from there. The habit of testing regularly will teach you more than any single audit ever could.