From Search Console query to AI tracking prompt: a mapping method

To turn a Search Console query into an AI tracking prompt, treat the query as evidence of what people searched for on Google, then rewrite it as the full question a buyer would put to an AI assistant. A query is a fragment ("document review software"); a prompt is a complete question with a buyer, a context and a decision behind it. Sort your queries by impressions and intent, drop the ones that make poor questions, rewrite the rest, and log every decision. The result is a shortlist you can defend, though it stays an editorial choice: Search Console reports Google results, not AI questions.

In short

• A Search Console query is a search string typed into Google. An AI tracking prompt is a full question you choose to monitor. They are related, not interchangeable.
• Use queries as evidence to prioritise and phrase prompts, never as a measure of how often AI engines are asked something.
• Not every query deserves a prompt. Branded, navigational and single-word head terms usually make poor ones.
• Rewrite each keeper as a buyer question with a role, a situation and a decision, then log the source query, the rewrite and the reason.
• Each country you track a prompt in uses a plan slot, so a written reason for every inclusion and exclusion keeps the list small and reviewable.

What is the difference between a keyword and an AI tracking prompt?

A keyword or query is a short string a person typed into a search box. A prompt is a full question that a monitoring tool runs against AI engines on a schedule. DiscoveredBy's key terms page says it plainly: a prompt is not a keyword, and it is not something a user typed. It is a question you chose to monitor because it resembles what your buyers ask.

Two consequences follow. First, queries carry little context. "Contract review software" does not say who is buying, what they are comparing, or what happens next. A prompt has to, because an AI engine answers the question as written, and a vague question tends to get a generic answer. Second, a prompt is something you decide to run. Search Console tells you what surfaced your pages; it cannot tell you which questions to monitor. That is judgement, and this post gives you a structure for it.

The glossary entry for prompt covers the term; the prompts docs page covers how a prompt runs in DiscoveredBy.

What can Search Console actually tell you?

Search Console tells you which Google searches surfaced your pages, how often, and at what average position. It does not tell you what anyone asked an AI assistant.

The DiscoveredBy Search performance screen reads your Search Console property and shows queries, pages and query-and-page pairs, each list up to 100 rows ranked by clicks. The default window is the last 28 days, and you can switch to 7 or 90. That is a useful pool of candidates, with three limits worth knowing before you start:

• It only contains queries that already surfaced you. A question your buyers ask that you do not rank for is invisible here. Silence is not zero demand.
• Impressions are a Google measure. An impression means one of your pages appeared in a Google result for that query. The prompts docs say it directly: nothing in the product measures how often a question is put to ChatGPT or Gemini, and Google demand is not a stand-in for it.
• Average position is a blend. It is weighted by impressions across the window, so a query that ranked well on some days and poorly on others shows a middling number.

Used honestly, Search Console gives you a defensible way to rank candidate questions against each other. It does not give you a forecast.

How do you turn a query into a prompt?

Work through five steps. Each produces a column in the mapping sheet below.

1. Collect a candidate list. Copy or export the top queries from your window in Search Console, or read them off the Search performance screen, with impressions, clicks and average position. Include the page that ranked (the query-and-page tab shows the pairs), because the page often reveals the intent the bare query hides.
2. Classify the query. Note whether it is branded (your name), competitor-branded, navigational ("login"), informational, commercial or comparison. DiscoveredBy's prompt classifications use intent (informational, navigational, commercial, comparison, troubleshooting) and buyer stage (awareness, consideration, decision), so using the same words now saves rework later.
3. Decide keep, merge or drop. Keep queries that imply a real buying question. Merge near-duplicates. Drop the rest and write down why.
4. Rewrite each keeper as a question. Add the buyer, the situation and the decision. Keep it in natural language, as you would ask a colleague.
5. Record the evidence. Log impressions, position and the ranking page, plus a note on what you expect to learn. That note keeps the prompt from drifting into a vanity metric later.

For the wider question of which prompts belong in a starter set, see How to choose AI tracking prompts that reflect buying decisions. This post is narrower: it starts from data you already have.

Enlarge image

Enlarge image

Which queries make poor prompts?

Four kinds of query usually deserve a "drop" or a different treatment.

• Your own brand name. "Quillstone pricing" tells you people already know you. It may be worth a fact-oriented prompt later, but it is not a discovery question. Whether to track branded questions at all is a separate decision, so make it deliberately.
• Navigational queries. "Quillstone login" is someone trying to reach a site, not choosing a product.
• Single-word head terms. A bare "software" has no buyer in it. DiscoveredBy's own demand matching ignores queries with fewer than two meaningful words for the same reason.
• Queries whose page is unrelated to the buying decision, such as a career or press page that happens to rank.

Competitor-name queries ("Clausewise alternative") are the interesting exception. They are usually comparison-stage questions with real commercial weight, so they often belong on the list, rewritten so the question stays neutral.

The query-to-prompt mapping sheet

This is the deliverable. Copy it into a spreadsheet; one row per query you considered, including the ones you dropped.

QUERY-TO-PROMPT MAPPING SHEET

Project / brand:                  Window (7, 28 or 90 days):
Date pulled:                      Search Console property:

COLUMNS (one row per candidate query)
 1  Query (exact text from Search Console)
 2  Impressions in window
 3  Clicks in window
 4  Average position (blended; lower is better)
 5  Ranking page (URL)
 6  Query type: branded | competitor | navigational | informational | commercial | comparison
 7  Decision: KEEP | MERGE into row # | DROP
 8  Reason for the decision (one line, mandatory, also for DROP)
 9  Prompt wording (a full buyer question, for KEEP rows)
10  Buyer stage: awareness | consideration | decision
11  Intent: informational | commercial | comparison | troubleshooting
12  Countries / cities to track
13  Tags to attach (for later filtering)
14  What I expect to learn (one line)
15  Duplicate check: is a tracked prompt already this question? Y/N
16  Status: proposed | added | rejected in review
17  Review date

RULES
 - Every dropped query gets a reason; unexplained drops get revisited.
 - The rewrite adds a buyer and a situation. It does not add claims about my product.
 - Do not sum impressions across rows that share a query; count each query once.
 - Impressions are appearances in Google results. Never present them as AI question volume.

Enlarge image

If you plan to add prompts in bulk, DiscoveredBy's import takes a file with a column of prompt text, and optional country and tags columns. A header named prompt, query or question is recognised for the text. Imported prompts start unclassified, so classify them afterwards. Each new country costs one prompt slot, the same as adding one by hand. See Prompts for file limits.

How do you rewrite a query as a buyer question?

Add three things the query leaves out: who is asking, what situation they are in, and what they need to decide. Then check that the question still reads like a person, not like a keyword stuffed into a sentence.

A useful test: could a colleague in the buyer's role plausibly type it into an AI assistant? If it sounds like a search box, it has not been rewritten yet.

Two cautions. Do not smuggle your own brand into an unbranded question, because a question that names you measures something different from one that does not. And keep wording stable once tracking begins: changing the wording changes the question being measured, so results before and after the edit are no longer like for like.

How does DiscoveredBy use your Search Console data?

DiscoveredBy uses Search Console data in three separate ways, and none of them turns a query into a tracked prompt automatically. The documentation describes each.

Google demand on prompts. With Search Console connected, each tracked prompt carries a figure: the impressions Google recorded in the last 28 days for the queries the prompt covers. A query counts toward a prompt when every meaningful word of the query appears in the prompt, the query has at least two meaningful words, and singular and plural are treated as the same. It is best for ranking your own prompts against each other. The 1 to 5 score is relative to your own project (and only appears once at least three of your prompts match a query), and a dash means no matching query, which is not the same as zero demand. Two prompts can share a query, so there is no project total.

Suggestions. Suggested prompts come from a research run that reads your site, builds a profile of the business, and asks a model for buyer-intent questions. Your best-performing Search Console queries are handed to that call as extra context. A Google query never becomes a suggestion by itself; it only reaches the queue folded into the profile a model reasoned over. See Suggestions and opportunities.

Search Console opportunities. A separate list finds (query, page) pairs where the page averaged position 5 or better over 28 days, drew at least 100 impressions, the query matched an active tracked prompt, and that prompt's citation rate sat below 20 percent. That is a reason to check a page you already rank for, and it only exists once you track the matching prompt. It is also a different list from the Quick wins on Search performance, which covers positions 4 to 20 with at least 10 impressions.

So the mapping sheet is the human step the product deliberately leaves to you. Once prompts are tracked, the same connection can show which of them match queries Google has already shown your pages for.

Worked example: Quillstone maps seven queries

Quillstone sells document-review software to mid-sized legal and compliance teams. Its SEO lead pulls a 28-day window and lists seven queries. The impressions sum to 12,000.

Four queries are kept: 4,200 + 1,900 + 640 + 1,250 = 8,000 impressions, two thirds of the 12,000. Their rewrites:

Row 3 is worth a second look. Quillstone should decide deliberately whether the competitor name in the prompt is a feature or a distortion, since it will change which brands the answers name.

Two Search Console lists sit beside the sheet. Rows 1, 2, 5, 6 and 7 fall in the 4 to 20 position band with at least 10 impressions, so they would appear as Quick wins, whatever the sheet decides. The opportunity list is stricter: it needs position 5 or better, at least 100 impressions, a match to an active tracked prompt, and a citation rate below 20 percent for that prompt. Row 3 (position 3.2, 640 impressions) could qualify once its prompt is tracked and measured. Row 2, at position 6.1, could not, however well it performed. That difference is why the two lists should not be read as one.

What this method cannot tell you

• It cannot show AI demand. Impressions measure how often your pages appeared in Google results. A question with few impressions might be asked constantly of an assistant, and a heavily searched query might rarely be.
• It cannot show what you are missing. Queries you do not rank for never appear. Add prompts from sales calls, support tickets and buyer interviews too.
• It does not prove a prompt is worth the slot. Each country you track a prompt in uses one of your plan's prompt slots (see plans and limits), so treat every KEEP as a hypothesis and review it after a few collection cycles. Audit your prompt list before adding more prompts covers that review.
• Ranking well on Google does not mean AI engines cite you. The two are measured separately; see Google rankings versus AI citations.
• Rewrites can drift. A buyer question you invent may not match how real buyers phrase it. Keep the source query in the sheet so you can check.

Common mistakes

• Pasting every query in as a prompt, then discovering the list is full of branded and navigational terms.
• Treating the impressions column as a ranking of AI importance.
• Adding a competitor's name to a prompt without recording that it shifts what the answer measures.
• Rewriting a prompt in the middle of a comparison period.
• Dropping queries without a reason, so the next person re-adds them.
• Summing impressions across prompts that share a query.

Frequently asked questions

Should I track every query that appears in Search Console?

No. Most queries make poor questions: branded, navigational or single-word terms. Track the ones that imply a real buying decision, and keep the rest in the sheet with a reason.

Does high Google impression volume mean AI assistants are asked that question often?

Not necessarily. Impressions count appearances in Google results. DiscoveredBy does not measure how often a question is put to ChatGPT or Gemini. Use impressions to rank your own candidates against each other.

Does DiscoveredBy turn Search Console queries into prompts for me?

Not directly. Your best-performing queries are given as context to the research run that proposes suggestions, but a Google query never appears on the queue as-is. You still decide what to track.

What if a good buying question has no Search Console query behind it?

Track it anyway if your buyers ask it. Search Console only reports queries that surfaced your pages. Sales calls, support tickets and customer interviews fill the gaps.

How many Search Console queries should I convert?

There is no correct number. Start with the few that carry the most decision weight, and expand after you have looked at real results. See How to choose AI tracking prompts that reflect buying decisions for sizing a starter set.

Next step

Connect Search Console, work through the mapping sheet for your top 25 queries, then add the KEEP rows as prompts and check whether a Google demand figure or a dash appears beside each one. Once results arrive, AI visibility, brand mentions, and citations explains how to read them. You can see the numbers and screens in the analytics features overview, and connect a property from integrations. Start tracking your prompts.