Google Search Console Import: Topics and Prompt Explorer for AI Visibility
The week-one workflow we run for every client: seed the AI prompt list from their own Search Console queries, then promote by hand.
Every client kickoff includes the same argument. Someone proposes a prompt list from a brainstorm: what would our customers ask ChatGPT? The list comes out aspirational, ten variations of the flagship product's name and a few questions the CMO wishes people asked. We have learned to skip the brainstorm and start from evidence instead. The client's own Search Console queries are the best public record of how their buyers actually phrase things, and that is what we import.
The brainstorm fails for a reason worth naming. The people in the room are not the buyers, and the buyers are not in the room. A marketing team writes prompts that sound like the brand. Buyers write prompts that sound like their problem. Search Console is the closest thing to a transcript of the buyer's phrasing, because it records what people typed when they were already looking for help. That is a better seed than any whiteboard.
The mechanics, on Promptwatch, where we run client programs: the GSC connection is read-only OAuth at the organization level, set up once in Settings. One property pairs with one project, and multiple Google accounts can be connected, which matters for us because clients grant access from their own accounts, not ours. The sync pulls up to a year of performance data, clicks through position, and the queries land in Prompt Explorer as candidates.
Read-only matters to the client's security team. The grant does not change GSC settings, sitemaps, or URL Inspection. It reads. That is usually the line that unblocks the ticket. The year-long window matters to us, because a 28-day GSC view makes seasonal pages look dead. A year of history lets us see the query that spiked in March and went quiet in July, and decide whether it is worth a tracked slot. Without the longer window we would cut pages that only earn their keep in one quarter, and that cut would cost the client real demand the moment the season returned.
Candidates, not prompts
This distinction carries the whole workflow. An imported query costs nothing and commits nothing. A promoted prompt occupies one of a finite number of tracked slots and gets checked daily across the AI engines. So week one looks like this: import, then a working session where a human cuts the candidate list down hard.
The slot budget is the reason the cut has to be honest. Every prompt you promote eats capacity and gets checked every day on every engine the plan covers. A list of 200 prompts sounds thorough. In practice nobody reads 200 rows a week, and the ones that matter get buried. A list of 40 prompts that a human actually opens is the program. The import gives you the raw material. The cut is the work.
Our cut criteria are not secret. We keep queries with commercial or comparison intent, queries where the client has a page worth citing, and queries whose fan-outs reveal a cluster the client can plausibly own. We drop navigational queries, one-off news spikes, and near-duplicates that fan out to the same sub-questions. The scoring helps here: each candidate shows a volume band and a difficulty score, and difficulty reflects how authoritative the currently cited sources are. A high-volume prompt where the models already cite three government sites is a worse first target than a mid-volume prompt cited from two thin blog posts. The mid-volume prompt has a path to a citation. The high-volume one has a wall.
The cluster point is the one most teams miss on the first pass. A single query is one row. A fan-out that keeps circling the same five sub-questions is a topic. If the client can own that topic with three or four pages, those four pages are worth more than four unrelated prompts scattered across the list. We read the fan-outs during the cut, not after promotion, because that is where the cluster shows up. Promote first and you lose the chance to see the shape of the demand before you spend the slots on it.
A number we watch clients misread: their Google position. Ranking first in Google does not predict an AI citation, because models assemble answers from whatever sources fit the specific prompt. The import is valuable precisely because it surfaces this gap. The queries where you win Google clicks and get zero AI mentions are the clearest priority list a GEO program can start from. Google's generative AI performance reports stay in Search Console for Overviews. They do not replace the prompt list we promote by hand.
That gap, between Google clicks and AI mentions, is the whole reason the import beats the brainstorm. The brainstorm assumes the brand is visible. The import shows where the brand is visible in Google and invisible in ChatGPT, which is the exact set of queries a GEO program should attack first.
What the seeded list feeds
Once promoted, prompts run daily on the engines the plan covers, and everything downstream keys off them. Citation analytics show which pages get cited for those prompts, and content gaps get computed against the prompt set rather than against a generic topic universe. On Professional, Business, or an agency plan, crawler logs also show whether the bots fetch the pages that should be cited. A prompt list seeded from real queries makes each available report more honest.
Honest is the word. A content gap report run against a generic topic universe produces generic briefs. The same report run against the prompts your buyers actually type produces briefs you can defend in a QBR. The crawler log join is the same: a fetch on a page that maps to a promoted prompt is a signal. A fetch on a page with no prompt behind it is noise. When the prompt list is real, the reports stop being decoration and start being a worklist.
Plan-wise, prompt capacity is the first constraint to know: 50 on Essential ($95/mo), 150 on Professional ($245/mo), 350 on Business ($579/mo), unlimited on the agency plans we run. Essential has no listed crawler-log allowance. If the seeded-list workflow must include Agent Analytics, use Professional, Business, or an agency plan. We rarely promote more than 80 prompts for a mid-size client in quarter one. Fewer prompts, reviewed weekly, beat a large list nobody reads.
The plan choice follows from the workflow, not the other way around. If the program is prompts and citations only, Essential is enough. The moment crawler logs become part of the brief, you are on Professional or an agency plan. We do not pretend Essential has logs, and we do not sell a $29 Promptwatch plan. There is no such plan.
FAQ
Are imported queries the same as tracked prompts?
No. An imported query costs nothing and commits nothing. A promoted prompt occupies a tracked slot and gets checked daily. Week one is import, then a human cut.
Does ranking first in Google predict an AI citation?
No. Models assemble answers from whatever sources fit the specific prompt. The queries where you win Google clicks and get zero AI mentions are the priority list.
How many prompts do we promote in quarter one?
We rarely promote more than 80 prompts for a mid-size client in quarter one. Fewer prompts, reviewed weekly, beat a large list nobody reads. Agency plans have unlimited prompt capacity.
If you want the import and the first cut done with someone who has run this a few dozen times, write to hello@1001seomedia.com. Bring GSC access; that is the only prerequisite.