How to build an AI prompt set from Search Console queries

A marketing manager who owns a Search Console account already owns a list of the questions that buyers type. This guide rewrites that list as a prompt set for an AI visibility tool. The example is the export of our own site, fullreach.ai. Our guide on which kinds of prompt name brands measures what each kind of question does to the number. This guide covers the step before it: where the questions come from.

Why do Search Console queries tell you which prompts to track?

Search Console lists the exact words that buyers typed before Google showed your page. Some of those words reach the AI platforms too. In the 28 days to 30 October 2026, ChatGPT and Grok ran 430 distinct web searches before they answered our prompts. In those 28 days, 10 of the 430 searches matched one of 8 export queries word for word, if you ignore case and the year. A prompt set is the list of questions that an AI visibility tool asks each day. An impression is one appearance of your page in Google's results for one query.

A new site gets few clicks, so the impressions carry the signal. Our export holds 400 queries with 2,736 impressions and 5 clicks in the 90 days to 29 October 2026. All 5 clicks came from the four queries that contain our brand name. The other 396 queries sent no visitor, but they held 2,700 impressions.

Position does not change which queries become prompts. In the 90 days to 29 October 2026, the 204 queries that ask for a tool had a median position of 88. Only 8 of them placed in the first 20 results in those 90 days. Our page held an average position of 93.5 for "best ai visibility tools" in the same 90 days. In the 28 days to 30 October 2026, Grok searched "best AI visibility tools 2026" 16 times before it answered.

How do you export the queries?

Export the Queries table of the Performance report for the last 90 days. A shorter window misses the questions that buyers ask a few times a month.

1. Open Search Console, then the Performance report for your property.
2. Set the date range to the last 3 months, and open the Queries tab.
3. Export the table as a spreadsheet. Keep the columns query, clicks, impressions and position.
4. Add three columns: group, prompt and topic. The next sections fill them.

A site with more than one country needs one export per country. Filter the report by country before each export, because a prompt runs in one market.

How do you group the queries by intent?

Sort each query into one of seven groups by the words it contains. The table shows the groups, the rule that sorts a query into each one, and what the group becomes. The counts are our own export, 400 queries in the 90 days to 29 October 2026.

Two of three impressions went to requests for a tool. That is the group that an AI visibility tool measures best, because an answer to such a question lists tools. The how-or-why group is small, 33 queries, but it holds the questions that your pages must answer.

The "Other" group needs a person. Ours held feature names such as "share of answer", segments such as "ai visibility for small business", and four misspelled names of other companies. A feature name becomes a how-or-why prompt. A segment becomes a request for a tool with the buyer in it. Drop a wrong company name.

How do you turn a query into a prompt?

Rewrite the query as the full question that a buyer speaks, and keep the group's purpose. The table pairs seven queries of our export with the prompt in our own set that covers each one. Our project tracks the category we sell in, AI visibility tools, and the topics are the ones it uses.

Follow five rules when you rewrite a query:

1. Write one full question. "ai visibility agency" becomes a question about the best tool for agencies, not two words.
2. Keep the buyer or the need when the query has one. "for small business" and "for agencies" stay.
3. Keep brand names out, except in the Comparisons topic.
4. Remove the year, the quotes and the plural. "best generative ai seo tools 2026" and "generative ai seo software" become one prompt.
5. Write the question in the language of the export. A German export becomes German prompts.

Which queries do you drop, and which get their own topic?

Drop the queries that contain your brand name, and drop every search operator. Give the queries that name a rival a topic of their own.

In the 90 days to 29 October 2026, our four brand queries held 36 impressions and all 5 clicks. Each one contains the words "full" and "reach". They measure people who already know us, not buyers who choose. In the same 90 days, our six operator queries held 197 impressions, 7% of the export. Five start with "site:reddit.com", and one excludes sites with "-site:". A query with a "site:" operator is not a question that anyone asks an AI platform.

The rival group was the second largest in the 90 days: 103 queries and 490 impressions. 42 of them named one rival, Peec AI, for 237 impressions. 33 compared two rivals with "vs". 25 asked for a rival's alternatives or competitors. 11 asked for a rival's price. Each kind becomes a prompt, and all of them go in one Comparisons topic.

How many prompts does an export give?

Count the distinct wordings after you strip the filler words, and keep the wordings with ten or more impressions. In our 90-day export, the 204 requests for a tool reduce to 101 wordings once "best", "top", "tool", "software", "platform" and the year are gone. 15 of the 101 wordings reached ten or more impressions. Those 15 held 1,600 of the group's 1,790 impressions, 89%. 47 wordings got one impression each.

So the tool requests of our export give 15 core prompts. The price, how-or-why and rival groups add their own prompts to those 15. Our guide on how many prompts give a stable number gives the size that a full set needs.

For comparison, our own project asked 47 prompts in English on 7 AI platforms in the 28 days to 30 October 2026. Three of them started later in that window, so 44 ran on all 28 days. The 47 sit in nine topics:

• AI visibility tools, 8 prompts
• Alternatives to GEO tools, 7
• AI ranking tools, 6
• Comparisons, 6
• SEO for generative search, 5
• Brand visibility in AI, 5
• Measuring AI visibility, 5
• Pricing, 3
• Agencies, 2

On 30 October 2026, the project also held 24 paused prompts in four neighboring topics. The Agencies topic matches the segment queries of the export, and the Measuring topic matches the how-or-why queries.

What to do next

1. Export the last 90 days of queries. Fill the group column with the seven rules above.
2. Write one prompt for each wording with ten or more impressions. Put the rival names in a Comparisons topic.
3. Start the set, and read it after a week. Our guide on generative engine visibility gaps lists the prompts whose answers name a rival and not you. Our guide for in-house marketing teams holds the weekly read.

For how FullReach AI collects the answers, read the methodology.