Google Search Console Import: Topics and Prompt Explorer for AI Visibility
Your first prompt list already exists in Search Console. How the GSC import into Prompt Explorer works, and where buyers overdo it.
The hardest part of starting AI visibility tracking is not the tool. It is the blank prompt list. Most teams stare at an empty field labeled "add a prompt" and type whatever the meeting suggested, which is how you end up tracking questions nobody asks. There is a better seed source, and you already own it: your Google Search Console queries. A prompt list built from meeting guesses tracks the questions the room found interesting, which is rarely the questions the buyers actually type.
Promptwatch lets you pull existing GSC queries in as a starting point instead of building a prompt list from scratch. That single sentence is the whole pitch, but the mechanics decide whether it works for you, so here they are. The pitch is short because the idea is simple. The execution is where most programs go wrong, and the wrong move is almost always promoting too many queries too fast.
How the connection actually works
The connection lives in Settings and uses read-only Google OAuth at the organization level. You can connect multiple Google accounts, and each Promptwatch project takes one GSC property. Once linked, the sync pulls up to a year of performance data: clicks, impressions, CTR, and position. URL import is optional, and URL Inspection data comes along, which means indexing status shows up next to AI crawler reads if you also have logs connected. The read-only grant matters because it answers the security question before the security team asks it: Promptwatch can read your GSC data, it cannot change it, and the property stays under your control.
The queries land in Prompt Explorer. From there you promote the ones worth tracking into monitored prompts, and this is the step where judgment beats volume. A query that earns Google clicks is evidence of demand, not proof it is worth an AI-visibility slot. A query can bring a thousand Google clicks a month and still be a poor AI prompt, because the AI answer space for that question may be dominated by a single source, or the question may not be one a model would decompose. Prompt allowances are finite: Essential at $95/mo carries 50 prompts, Professional at $245/mo carries 150, Business at $579/mo carries 350. Agency plans from Kick-off at $199/mo have unlimited prompts, which changes the calculus but not the principle. A tracked prompt nobody reviews is noise with a renewal date. Unlimited prompts do not make unreviewed prompts useful, they just make more of them.
What the promoted prompts gain
Here is why importing into a platform beats keeping the queries in a spreadsheet. Each promoted prompt picks up search volumes, a difficulty score, and query fan-outs, the sub-questions a model generates when it decomposes your prompt. Volume is shown as a band from 1 to 10k+ and is computed from keywords attached to the prompt, weighted toward AI search demand rather than raw Google numbers. Difficulty estimates how contested the answer space is, based on the past 30 days. A spreadsheet of GSC queries gives you the query and the click count. The platform gives you the query, the click count, and a read on whether the AI answer space for that query is worth fighting for.
So the loop looks like this: GSC tells you what people asked Google, the import turns that into candidate prompts, and the scores tell you which candidates are worth your finite slots. Then daily checks across ChatGPT, Gemini, Claude, Perplexity, and the rest tell you whether the models mention you when those questions get asked. Each step feeds the next, and the step that requires a human is the promotion step. The import is wide, the promotion is narrow, and the daily check is what turns the slot into a measurement.
What to avoid
Do not bulk-promote a year of queries. GSC will happily hand you thousands of long-tail variants, most of which fan out to the same underlying question. Import wide, promote narrow. A year of queries promoted at once fills the prompt allowance with duplicates of the same intent, and the daily check then spends its budget measuring the same answer five ways. And do not treat the GSC connection as a replacement for GSC itself; Google's own reporting stays the source of truth for Google clicks. The generative AI performance reports stay there too. The import is a seeding mechanism, not a migration. GSC remains the official record of what happened on Google, and Promptwatch is the record of what happened on the AI side of the same questions.
No other tool in our coverage documents this exact workflow, though Semrush AI Toolkit ($99/mo per domain) approaches from the other side by generating prompt approximations from keyword data. Generated approximations are not your buyers' language. Queries your site already earned are closer. A generated prompt is a guess at what a buyer might type. A query your site earned is a question a buyer actually typed, and that is a stronger starting point for an AI-visibility slot.
Setup walkthrough: connecting Search Console. Method: how we rank.
FAQ
Does a GSC query automatically become a tracked prompt?
No. Queries land in Prompt Explorer. You promote the ones worth tracking. Import wide, promote narrow. The two-step split is deliberate, because a query that earns Google clicks is not automatically a prompt worth an AI slot, and the human judgment is what keeps the prompt allowance useful.
How many prompts do brand plans allow?
Essential at $95/mo carries 50 prompts, Professional at $245/mo carries 150, Business at $579/mo carries 350. Agency plans from Kick-off at $199/mo have unlimited prompts. The allowances scale with the plan, and the promotion discipline matters more on the smaller allowances, where a wasted slot is a larger share of the budget.
Is the GSC grant read-only?
Yes. The connection uses read-only Google OAuth. Promptwatch does not edit Search Console. The grant is read at the organization level, it supports multiple Google accounts, and each Promptwatch project takes one GSC property, so the scope of the read is bounded by the property you point it at.