Ai Search

The 20 Prompt Panel for AI Visibility

An AI visibility prompt panel is a fixed list of buyer questions you run against AI assistants on a schedule, so you can see whether, and how, your business shows up in the answers they give. The twenty prompt version is deliberately small: short enough to run by hand or on a cadence, and stable enough that this month's run can be compared directly to the one before it.

Most people check their AI visibility once. They type their company name into ChatGPT, read the reply, and either panic or relax. A prompt panel replaces that single snapshot with a repeatable instrument. The prompts do not change. The engines do not change. The only thing that changes between runs is the answer, and that is exactly what makes a result a measurement rather than an anecdote.

What the panel measures, and what it cannot

An AI assistant does not hand back a rank. It hands back a sentence. So a prompt panel measures what a sentence can actually reveal, not a position number on a results page.

Mention: does the answer name your business, in plain words, inside the response.

Citation: does the answer link to or source your own website. This is not the same as a mention, and it usually matters more, because a citation is a signal the model trusts you and can sometimes send a real click.

Shortlist: when the prompt asks for options, are you named among them, and where in the order.

Framing: how the model describes you, and whether the facts it states about you are correct.

These come apart in useful ways. A mention with no citation means the model learned about you from a source it will not credit. A citation with no shortlist spot means you were a source the model read but not an option it picked. The panel keeps these separate, because each one points at a different fix. What the panel cannot do is promise a position. AI answers are not a leaderboard you can buy your way onto, and any tool that claims a guaranteed spot is selling something that does not exist.

Why twenty prompts, and not two or two hundred

Two prompts is a hunch. Two hundred is a project that gets run once and never again. Twenty is the number that survives contact with a real schedule.

The whole value of a panel is that you re-run it unchanged. A trend only exists if the inputs hold still. A list so long that nobody repeats it is worse than a short list you actually run every month, because the long list gives you one data point and then dies. Twenty well chosen prompts cover the handful of ways a genuine buyer phrases the thing they want, without turning the panel into a chore.

What the twenty prompts should cover

Category questions: best option in a city, top choice for a specific need, who to hire for a job.

Comparison questions: one provider versus another, or alternatives to a competitor a buyer already knows.

Problem questions: the symptom someone types before they even know the name of the service.

Direct questions: your own business name, to check what the model already states about you unprompted.

Qualifier questions: price, timeline, and who the service is for, the practical screens a buyer applies before they call.

How a run works across engines

Each prompt goes to each engine you care about: ChatGPT, Google AI Overviews, Perplexity, and Gemini. The same words produce different answers on each one, because each engine retrieves and grounds its reply in its own way. Perplexity shows its sources openly on every answer. ChatGPT may reply from its training, from stored memory, or from a live retrieval step, and its willingness to cite shifts depending on which path it took.

Answers also drift between two runs on the same engine. The models are non-deterministic, and the web they read keeps moving underneath them. One run, on its own, is noise. A panel run on a steady cadence turns that noise into a line you can read: are you holding, climbing, or slipping. Location and personalization can change what you see too, so a run is only comparable to another taken under the same conditions.

Record the full answer text, not just a yes or no, so you can watch the framing shift over time.

Log the engine, the date, and whether each hit was a mention, a citation, or both.

Keep every prompt identical between runs. The moment you reword one, its history resets to zero.

Reading the results and what they tell you to fix

A prompt panel is a diagnostic. It does not improve anything by itself. It points at the work.

When competitors get named and you do not, the gap is almost always sources: the model is reading pages, directories, and discussions that talk about them and stay silent about you. When you are cited but the details are wrong, the gap is your own structured content and the entity data the model grounds on. When you are mentioned but never make the shortlist, the gap is positioning and proof on the pages the model already reads. Each pattern has a different answer, and the panel is what tells you which one you are looking at.

Where the panel fits a program

A prompt panel earns its keep as a before and after. Run it at the start to set a baseline, then re-run it on a cadence while the real work happens underneath: content structuring, entity cleanup, crawler access, and earning citations. The panel is the scoreboard. It is not the game, and it does not score itself.

Ninetyfifth Marketing runs a panel like this as part of how it monitors AI visibility, and the discipline is the same every time: same prompts, same engines, same cadence, so the trend stays honest. To see how that monitoring sits inside a full SEO and AI search program, read /services/seo/. For how engagements are structured and priced, see /pricing/seo/.

Questions

How many prompts should an AI visibility panel have?

Enough to cover the real ways a buyer asks, and few enough that you will actually re-run them. Twenty is a practical sweet spot for most local and service businesses. The exact number matters less than keeping it fixed between runs, because a changing list has no trend to read.

What is the difference between a mention and a citation in a prompt panel?

A mention is when the answer names your business in plain text. A citation is when the answer links to or sources your website. A citation usually carries more weight, because it signals the model trusts your site enough to credit it and can send a real click. The panel tracks both separately because the fix for each is different.

How often should I run the prompt panel?

On a steady cadence, often monthly, so you build a trend instead of a snapshot. AI answers drift between runs because the models are non-deterministic and the web keeps changing, so a single run tells you very little. Running the same prompts on the same engines at the same interval is what turns that drift into a readable line.

Can a prompt panel guarantee I will show up in ChatGPT?

No. A panel measures where you stand and whether your work is moving the line. It does not promise a position, because AI answers are not a ranking you can buy into. Anyone guaranteeing a spot in an AI answer is selling something that does not exist.

Do I need to run the panel on more than one AI engine?

Yes, if your buyers use more than one. ChatGPT, Google AI Overviews, Perplexity, and Gemini each retrieve and ground answers differently, so your visibility can look strong on one and weak on another. Running the same prompts across each engine shows you where the gaps actually are.