Classify a keyword list by search intent and the page type it needs
A prompt that labels each query by intent, names the page format that satisfies it, and tells you which of your existing pages is the wrong shape for its keyword.
- Works in
- ChatGPT, Claude
- You need
- A list of queries · The URL currently targeting each one, if there is one
- Written for
- what is search intent
Scored by our own engine
This page, run through the audit we sell. Measured 6 August 2026.
A page can be about exactly the right subject and still be the wrong answer, because the searcher wanted a table and you wrote an essay. The prompt below labels every query by intent, names the page format that satisfies it, and, where you supply a URL, tells you whether the page you already have is the wrong shape for the query it targets.
The four labels are too coarse to build from
Informational, navigational, commercial investigation, transactional. Every guide on search intent stops there, and none of those four tells you what to build. “Commercial investigation” describes a state of mind. “Wants a comparison table with prices” describes a page.
That is why the prompt asks for both. The four labels keep the classification consistent across a long list, and the sub intent is the part you hand to whoever is writing. If you only take one column from the output, take the sub intent.
The wrong shape verdict is the reason to run this
Classifying keywords you have not built for yet is planning. Classifying keywords you already rank for, against the pages currently targeting them, is diagnosis, and it is much faster to act on.
The common finding is a guide sitting on a comparison query. It has links, it has age, it has some position in the results, and it is losing to thinner pages that happen to be shaped correctly. The fix is not more words. It is a table.
What the model cannot do here
It cannot see the results page. Everything in the output is inferred from the language of the query and from what the model absorbed about the topic during training, which means it will be confidently wrong on queries whose results changed recently. The prompt makes it say so instead of pretending, and flags the queries where different searchers genuinely want different things. Those flagged rows are the small set worth checking against a live search yourself.
You are an SEO strategist. I will give you a list of queries. Classify each
one by search intent and tell me what shape of page satisfies it. Where I
have given you a URL, tell me whether that page is the right shape.
For every query, give me:
1. The intent, as exactly one of: informational, commercial investigation,
transactional, navigational. Pick one. A query that genuinely sits
between two goes in the more commercial of them, and you flag it.
2. The sub intent in three or four words, because the four labels are too
coarse to build from. "Wants a definition", "wants a comparison table",
"wants a price", "wants a step by step", "wants somebody to do it".
3. The page format that satisfies it: definition page, guide, comparison,
listicle, product page, category page, calculator or tool, case study,
pricing page, contact or quote page.
4. The proof the reader expects on that page before they will act. For a
comparison it is a table with sources. For a service query it is
evidence you have done it before. For a how to it is a worked example
with real values.
5. If I gave you a URL, a verdict: right shape, wrong shape, or cannot tell
from the URL alone. When it is the wrong shape, say what the page is
currently and what it would have to become.
Then, separately:
- Any two queries in my list with the same intent and the same format that
should therefore be one page, not two.
- Any query where you think the intent is genuinely contested, meaning
different searchers typing it want different things. Say which reading
you would build for and why.
- Any query you could not classify, and what you would need to know.
Output the classification as a markdown table:
Query | Intent | Sub intent | Page format | Proof expected | Current URL verdict
Constraints:
- Do not guess what the results page currently shows. You cannot see it.
Classify from the language of the query and say so.
- Do not label everything informational. It is the safe answer and it is
the one that makes the output useless.
- Do not invent a page format outside the list above. If none of them fit,
say so and describe what would.
My business: [ONE SENTENCE ON WHAT YOU SELL AND TO WHOM]
My queries: [PASTE THE LIST, WITH THE TARGET URL AFTER EACH ONE IF YOU HAVE ONE]What to change
Everything in square brackets is yours to replace. Nothing else needs editing.
[ONE SENTENCE ON WHAT YOU SELL AND TO WHOM]- Intent is partly a property of the query and partly a property of who is reading it. "Best CRM" is commercial investigation for a software vendor and close to navigational for a consultancy whose clients arrive already knowing what they want. One sentence resolves it.
[PASTE THE LIST, WITH THE TARGET URL AFTER EACH ONE IF YOU HAVE ONE]- One query per line. Add the URL you currently point at it, separated by a space or a comma, and the output includes the shape verdict, which is the most useful column on the table. Queries with no URL come back as a plan for a page that does not exist yet. Fifty to a hundred and fifty lines works well.
How to run it
- 01Pull the queries you actually get impressions for
Export from Search Console rather than from a keyword tool. Those are queries that already connect you to real searchers, and reclassifying them tells you where an existing page is answering the wrong question, which is faster to fix than building anything new.
- 02Attach the URL Search Console reports for each one
The page column is what makes the shape verdict possible. If you skip it you get a clean classification and no diagnosis, and the diagnosis is the reason to run this rather than reading a definition of search intent.
- 03Run it and read the wrong shape rows first
Wrong shape is the finding with the shortest path to a result. A guide sitting on a comparison query does not need more words or more links, it needs a table. That is an afternoon of work on a page that already has some standing.
- 04Check the contested queries against a live search
For every query the model flagged as contested, search it and look at the mix of page types that rank. The results page is the only place the answer actually lives, and this is the small number of queries where it is worth your time to go and look.
- 05Merge the pairs it told you to merge
Two queries with the same intent and the same format are one page. Redirect the weaker URL into the stronger one rather than leaving both, and update the internal links pointing at the one you retired.
- 06Rewrite one page and check the title still matches the intent
Change the shape of the highest value wrong shape page, then check the title tag reads like the format it now is. A comparison page with a guide style title gets skipped on the results page, and the title checker will show you how it truncates before anyone sees it.
Questions people ask
What is search intent?
Search intent is what the person typing a query is trying to accomplish, as distinct from the words they typed. The four standard categories are informational, navigational, commercial investigation and transactional. It matters because the search engines rank the page format that satisfies the intent, so a page can be about exactly the right topic and still be the wrong answer.
What are the four types of search intent?
Informational, where somebody wants to know something. Navigational, where they want a specific site. Commercial investigation, where they are comparing options before spending. Transactional, where they are ready to act. The four are coarse, which is why this prompt also asks for a sub intent in a few words, since "wants a comparison table" is buildable and "commercial investigation" is not.
How do I find the search intent behind a keyword?
Read the query language for the signals, then check the results page, because the ranking pages are the best available evidence of what searchers accepted. A model can do the first half quickly and cannot do the second at all, since it has no live view of the results. Use it for classification at volume and verify the contested ones yourself.
Can one page satisfy two search intents?
Sometimes, and it is usually a mistake to try. A page that answers a definitional question and sells a service tends to do neither well, because the proof each reader needs is different. The exception is a genuinely contested query where the same page can lead with the answer and offer the commercial next step below it, in that order.
Why does my page rank for a keyword but get no clicks?
Often because the page is the wrong shape for the intent. If the searcher wants a comparison and your listing promises a guide, they skip you for the result that promises what they came for. Ranking measures relevance to the topic. Clicks measure whether your listing looks like the format they were expecting.
Unsubscribe in one click. We never pass your address on.