TL;DR. Google Search Console is a free Google tool that records the exact queries people typed before your site appeared in their results. For AI SEO, that query history is a goldmine: you rephrase real buyer queries as the questions people ask ChatGPT, and build the 50-prompt tracking list the rest of the playbook runs on. By the end of this project you can do it in one evening.
- Watch us mine our own Search Console (the receipts)
- What is Google Search Console?
- What does prompt mining look like when it works?
- Why is this Step 2 of The Playbook?
- How do you mine Search Console for AI prompts, step by step?
- How do you make the list your own?
- How do you train someone else on this?
- Where do you share your result?
- Frequently asked questions
Watch us mine our own Search Console (the receipts)
This is Project 2 of 7 in the AI SEO playbook. Project 1 gave you ten generic prompts; this one replaces guessing with your own data.
What is Google Search Console?
Google Search Console (GSC) is Google's free reporting tool for site owners. It shows every query your site appeared for, how many people saw you, how many clicked, and where you ranked. No tracking code to write, no subscription, and setup takes about ten minutes: search "Google Search Console," add your site, and verify ownership.
Most owners either never open it or stare at the wrong numbers in it. The value for AI SEO is one specific report: Performance, filtered to 12 months of queries. That is a year of your buyers telling you, in their own words, what they wanted right before they found you.
What does prompt mining look like when it works?
This is the step everyone skips, and it is the most valuable one in the whole playbook. Every client engagement we run starts here: before we write a word of content, we export the business's real queries and turn the high-intent ones into tracked AI prompts. The 50-prompt lists behind our client dashboards, including the scan that found a national sauna brand named in only 17% of its buyer questions, are built with exactly the method below.
The reason it works is simple: AI engines answer the same underlying questions people type into Google, just phrased as full sentences. "cryotherapy sore muscles" in your GSC becomes "does cryotherapy help sore muscles?" in ChatGPT. Same buyer, same intent, new battlefield.
Why is this Step 2 of The Playbook?
Because fewer than 50 tracked prompts means you are reacting to noise. One AI answer changing means nothing; a pattern across fifty means everything. Project 1 established your baseline on ten generic prompts. This project builds the full list that Project 7 will measure weekly, and that Project 6 will turn into your content plan. Think of GSC as your customers' question diary: you are not inventing prompts, you are transcribing them.
How do you mine Search Console for AI prompts, step by step?
One evening. Here is the drill:
- Verify GSC if you have not (ten minutes; the "Domain" property type covers your whole site).
- Open Performance, set the range to 12 months, and export your queries (the Export button gives you a spreadsheet).
- Highlight every query that sounds like a person with a problem. "does cryotherapy help sore muscles," "acupuncture for migraines cost," "gym with childcare near me." Skip your own brand name and one-word terms.
- Rephrase each one the way someone would ask an AI, as a full question. "acupuncture migraines cost" becomes "how much does acupuncture for migraines cost in [city]?"
- Add the questions real customers actually say. Ask your last ten: "What did you type or ask to find us?" Their wording beats your guesses.
- Add one line to your intake form: "How did you hear about us? Google / ChatGPT or AI / referral." That single field becomes your proof later, because "heard from AI" leads are the number that pays the bills.
- Build the list to 50 tracked prompts covering your services, your locations, your prices, and the "best of" questions from Project 1.
We exported our own 12 months of Search Console: 48 total queries, and the top four are misspellings of our own name. 507 impressions on "bread chaser," zero clicks on every non-brand query. Before this playbook, nobody found us for what we do, only for who we are.
Buried under the brand noise: real buyer problems we almost rank for ("seo for cryotherapy" at position 26, "med spa marketing agency," "health and wellness aeo"). Rephrased as AI questions, they seeded our 50-prompt tracking list, which now feeds the weekly share-of-voice number in Project 7. We also ran the pro upgrade below: validating the list against a commercial keyword database is how we learned one category term carries nearly triple the cost per click of the label we had been using, and that single number repositioned the whole company. Our version of the intake field: every score submit already carries a source tag into our CRM, so "heard from AI" is a countable number.
The receipts, including the unedited export, live in the Project 2 post in our build-in-public thread on X.
How do you make the list your own?
The upgrade: write your intake question in your own voice and put it everywhere, not just the form. Train whoever answers your phone to ask it and log the answer. Businesses that ask "what did you search to find us?" out loud collect prompt ideas no export will ever show, because spoken questions are exactly how people talk to AI.
The pro upgrade (optional): validate your list against a commercial keyword database. Search Console has one blind spot, and it is a big one: it only shows queries where you already earned at least one impression. The searches you never surfaced for are invisible. Agencies close that gap with paid keyword databases that report real Google search volume, cost per click, and competition for any term you type in, plus a snapshot of who actually ranks. Feed your 50-prompt list through one and three things happen: dead prompts get culled (a question nobody asks is not worth tracking), hidden neighbors appear (the tool suggests related terms with real volume), and CPC becomes a buyer-intent meter, because the price advertisers pay per click is the market's honest vote on which searches turn into customers.
That last one is not theory. When we ran our own list through this step, we found that one category term costs advertisers nearly three times more per click than the label we had been using for ourselves. Three times the buyer intent, in a category almost nobody had claimed. We renamed the company's positioning around it the same week. If you do not want to buy and learn a keyword database, this is precisely the layer a done-for-you partner runs for you, and it is a fair question to ask any agency you interview: "which terms did you validate, and what did the click prices tell you?"
Selling software instead of services? The B2B SaaS upgrade mines your sales-call transcripts instead of the front desk, because a category creator's call recordings are the keyword database no tool sells. Product brand? The eCommerce upgrade mines the engines' own fanout vocabulary: best, reviews, 2026, vs.
How do you train someone else on this?
Swap exports with another owner. You mine their queries, they mine yours. A stranger spots the buyer questions you have gone blind to, and explaining why a query is or is not a prompt makes the judgment automatic. Peer, mentee, mentor: three swaps and this is instinct.
Where do you share your result?
Post one surprising query. Every export contains at least one question you never guessed your buyers were asking. "People find my studio by searching THIS?" is a genuinely fun post, and it links back to this project for the next owner. Tag us and we will feature the best finds on the playbook.
Score yourself: your 50-prompt list existing and getting re-run on a schedule is +1 on the playbook's 0 to 5 AI-awareness score. Next: Project 3 fixes the fundamentals AI reads before it trusts anything you publish.
Frequently asked questions
Is Google Search Console free?
Yes, completely. It is Google's own reporting tool for site owners. Verification takes about ten minutes and there is nothing to install on most modern site platforms.
How many prompts do I need to track?
Fifty. Fewer than that and you are reacting to noise: one answer changing means nothing, a pattern across fifty means everything. Cover your services, locations, prices, and the "best of" questions.
What if my site is new and GSC is empty?
Start from the customer side: ask your last ten customers what they typed or asked to find you (or businesses like yours), use Project 1's ten prompts as the spine, and borrow question patterns from review sites in your category. Your GSC fills in as your visibility grows.
Does Search Console show what people ask ChatGPT?
Not directly; no tool does. But buyers ask AI the same underlying questions they type into Google, phrased as full sentences. Rephrasing your real GSC queries as questions is the closest honest proxy, which is why this step matters.
Two ways to run this playbook. Get your free AI Visibility Score in about 30 seconds: 50 prompts, 3 AI engines, scored live, plus a free 20-minute walkthrough with Matthew. Or get The Playbook and keep doing it yourself.
