Which prompts to track first when you already rank on Google

The blank page is the hardest part of AI visibility tracking. You've decided to monitor your brand in ChatGPT, Gemini and the rest, you open the setup screen, and it asks for prompts. Which ones? Teams usually respond in one of two ways: they invent twenty prompts in a meeting that all sound like the homepage, or they stall.

If your site already ranks on Google, you're holding a better starting point than a meeting. The keywords you rank for tell you what topics you're credible on, and those are the topics where you'd expect to appear in AI answers too. Here is how to convert them into a first prompt set.

Why start from ranking keywords

Three reasons, none of them magic.

First, they're real. They reflect queries people typed, not what the team wishes people asked.

Second, they mark where you have content. If you rank for "how to reconcile bank statements", you have a page on it. If that page doesn't appear in AI answers, you've found a gap worth investigating. If you track a prompt for which you have no page, a miss is expected and tells you little.

Third, they're organised by page. That makes it easy to connect a prompt to the URL that should answer it.

There's a limit. Google keywords are short and keyword-shaped, and people prompt AI in full sentences, often with context. A ranking keyword is a seed, not a finished prompt.

Step 1: pull your ranking keywords

Use Search Console for your own site: queries with impressions and clicks over the last three months. Group them by page. Ignore branded queries for now, since you'll track those separately.

Filter to the keywords that matter commercially, meaning the ones where a visitor could become a customer. A page that ranks for a hundred informational queries nobody buys from shouldn't dominate your first set.

If you want to see a competitor's ranking keywords, you need a third-party keyword tool, and numbers from those are estimates. Treat them as directional.

Step 2: turn each keyword into prompts

For each keyword you choose, write two or three prompt versions:

A direct question. "How do I reconcile bank statements for a small business?"
A recommendation request. "What's the best software for reconciling bank statements?"
A comparison. "X versus Y for bank reconciliation."

Mix prompt types deliberately. Information prompts show whether your content gets cited. Recommendation and comparison prompts show whether you get named. Our post on prompt volume in AI search explains why volume is estimated and shouldn't be your only filter.

Add the persona. "I'm a bookkeeper at a 10-person agency" changes the answer. If you have a priority buyer, phrase some prompts as that buyer.

Step 3: keep the first set small

Resist the temptation to track 300 prompts on day one. A first set of 30 to 50 prompts, split across your key topics, is easier to read and to act on. You'll learn which prompt types matter before you scale.

A balanced starter set might include:

a handful of branded prompts,
prompts for your top commercial topics,
competitor comparison prompts,
a few prompts for topics where you rank well on Google but suspect you're invisible in AI.

That last group is where the interesting findings come from. A keyword you rank for in the top three that produces no AI mentions is a clear case of a page that wins the old game and loses the new one.

Step 4: sanity-check and prune after a few weeks

After a few weeks of data, look at which prompts yield stable results and which swing wildly. Drop prompts that no real buyer would write. Add prompts for the sub-questions you see in the answers. Our guide to query fan-out explains why the engine's own sub-searches are worth mining for new prompts, and the prompt tracking tools comparison covers tools.

Where Promptwatch fits

The conversion from keywords to prompts is repetitive: export, filter, rephrase, paste, repeat. The tool we'd use for it is Promptwatch, and it's the one recommendation here.

Per the Promptwatch changelog, Prompt Explorer gained a Ranking keywords tab on September 19. You enter a page or domain, and Promptwatch looks up the Google keywords it ranks for, then turns them into prompt suggestions that you can add to a monitor. It sits next to the other sources: Search Console, People also asked, seed query and website page. So you can start from your own domain, a single page, or a competitor's domain.

Two details help with the blank page. The Prompt Explorer also supports Search Console as a source, which matches step 1 for your own site, and the platform's prompt tracking comes with search volumes, difficulty scores and citation rates, which helps with prioritizing. Since September 21, new monitors also get starting prompts, so a new project doesn't open empty.

On November 2, the changelog notes that Agent Chat now works inside Prompt Explorer. It sees the suggestions in the open tab and the ones you ticked. You can ask it to "track the ticked suggestions", which adds them the same way the Track button does, and it can track them into a new monitor. Before it changes anything, it asks for every value it needs instead of guessing a monitor or setting.

The mechanics suit the "start small" advice: you tick 30 to 50 suggestions, and they go into a monitor, rather than adding hundreds.

We'd still apply judgment. Suggestions generated from keywords inherit the keyword shape. Read them, rewrite the clumsy ones and make sure the set has prompts for the persona that matters. Automation produces candidates, not a strategy.

A quick way to check your first set

Before you commit, ask three questions of the list:

Could a real buyer plausibly type this?
Do we have a page that should answer it?
Would a good answer change a decision for us, such as content, linking or outreach?

If a prompt fails all three, remove it. If it passes only the second, keep it as a control.

Setting expectations

Tracking tells you where you stand. It won't lift your visibility by itself. The value of a good first set is that it makes the next steps obvious: which pages to refresh, which gaps to fill, which competitors keep showing up.

If you rank on Google today, you've already done the hard part of finding what you're credible for. The first prompt set is mostly a matter of translating it.