What this log covers
When a business asks whether AI tools "know about them," the honest answer is: it depends entirely on what you ask the tool. A clinic might be named instantly for one phrasing and invisible for another. So before we report any progress, we fix the question set. This log explains the prompt structure Seenu Tech uses to monitor AI visibility for clients across commercial real estate, dental and med-spa, education, and catering in the New York and New Jersey market. We run the same prompts on a weekly cadence so week-over-week changes mean something rather than reflecting random phrasing. The point is not to pick prompts that flatter the client. The point is to mirror how a real buyer would ask, then watch whether the answer engine surfaces the business, a competitor, or nobody useful at all. We document the wording so anyone on the team, or the client themselves, can reproduce the exact check months later. A monitoring program that changes its questions every week measures nothing; it just generates a feeling of activity. Fixing the set first is the unglamorous step that makes everything after it trustworthy.
The four prompt types we run
We sort every monitoring prompt into four buckets. Brand prompts name the business directly: "What do you know about [business name] in Fort Lee?" Category prompts describe the service without a name: "Best commercial real estate brokers near Hackensack for leasing office space." Buyer-intent prompts add a decision context: "I run a 20-person dental practice and need a med-spa build-out consultant in Bergen County, who should I talk to?" Competitor prompts test the neighborhood: "Compare the top three catering companies for corporate events in Jersey City." Each bucket answers a different question. Brand prompts test recall. Category prompts test whether you show up before someone knows your name. Buyer-intent prompts test fit for a described situation. Competitor prompts test relative standing in a crowded field. We deliberately keep the wording natural and a little messy, the way people actually type into an AI tool, because over-polished, keyword-perfect prompts produce results no real buyer would ever trigger. A baseline of twenty to thirty prompts, split across the four buckets, is usually enough to see a pattern without drowning in noise.
Why this matters for AI visibility
Most businesses only ever check the brand prompt, type their own name, see a paragraph, and assume they are visible. That is the least useful test. By the time a buyer types your exact name, they have already heard of you elsewhere. The category and buyer-intent prompts are where new demand actually lives, and they are far harder to win because the engine is choosing among options. Tracking all four together shows whether visibility is real or just brand recall. One education client was named confidently in every brand prompt but appeared in zero category prompts for "after-school STEM programs near Edgewater." That gap told us exactly where the content work needed to go, which a single vanity check would have hidden completely. The four-bucket view also protects against false comfort: it is entirely possible to look strong on brand recall while losing every prompt that actually drives new business, and only the full set makes that visible.
What we measured
For each prompt we record three things: was the business mentioned, was it mentioned first or buried, and did the engine cite a source URL we could trace. The source URL matters most. When an engine cites a page, we can see which content it trusted and reinforce it. When it mentions a business with no citation, the mention is fragile and can vanish on the next model update. Across one recent client baseline we found brand prompts returned a mention roughly nine times out of ten, while category prompts returned one well under half the time. We also logged which engine behaved differently, because Perplexity, ChatGPT, and Gemini do not weight sources the same way and a win on one is not a win on all three.
What still needs work
This method has real limits and we name them. AI engines change their models without notice, so a clean result one week can shift the next for reasons outside anyone's control. Prompt wording is a judgment call; we cannot test every phrasing a buyer might use, so we sample the common ones and accept that the long tail is unmeasured. We also cannot prove a mention caused a lead, only that visibility exists. And no monitoring set can promise placement, because the engines decide what to surface. We treat the prompt log as a thermometer, not a guarantee. It tells us where attention is and is not landing, which is enough to direct the work honestly. Finally, the log measures presence, not the quality of the answer surrounding it; an engine can mention a business in a lukewarm or hedged way, and counting that as a clean win would overstate progress. So alongside the yes-or-no of being named, we keep a short note on how the business was framed, because a confident, well-sourced mention and a passing aside are not the same result even when both technically count.
Next action and how to start
If you want to know where you actually stand, write down five prompts a real buyer in your category would type, run them in ChatGPT and Perplexity today, and note whether you appear and whether a competitor does. That ten-minute exercise usually surprises owners more than any report. From there, the structured version is an AI Visibility Snapshot, where we build the full four-bucket prompt set for your business and market and give you the baseline in plain language. If you would rather see the method before committing, the Living Lab page documents how we track our own visibility the same way. Either path starts with one honest question set instead of a vanity search.