A search intent AI prompt earns its place when it produces one label per keyword and forces you to justify each one. "Write an article about X" hides the decision that matters most: whether the reader wants to learn something, compare options, or buy something today, because each one demands a different page.
Label the keyword list first, group the labels, then brief the page. The prompts below do it in that order, and they include the reasoning behind each label so you can overrule the model instead of accepting it.
The keyword classifier prompt
Classify each keyword below by the intent a person most likely has when they type it. Use exactly four labels: informational (they want to understand or learn), commercial investigation (they are comparing options before a decision), transactional (they intend to buy, book, or sign up now), and navigational (they want a specific site or page). For each keyword give: the label, the signal in the wording that produced it, and a one-sentence description of what the searcher needs to see on the page. Where the wording could support two labels, give the primary and secondary label and explain what would move it to the other. Do not invent search volumes or ranking data. If a keyword is ambiguous enough that you cannot choose, mark it uncertain rather than guessing. Keywords: [paste list].
Fill in these details
- Your keyword list: Ten to fifty phrases. Long lists are fine, but the output stays readable only if the model returns one line per keyword.
- The four labels: Keeping them fixed is what makes the output sortable in a spreadsheet later.
- Your business: Add one line telling the model what you sell, because the same keyword means different things to a publisher and to a software company.
The "signal in the wording" column is the part worth reading. A single word such as "versus", "near me", or "template" often carries most of the intent, and noticing that in a list of a hundred keywords is exactly the work you do not want to do by hand.
What each intent class demands from the page
The label matters because it changes the structure, not just the tone. Knowing the four patterns before you read the labels keeps the output usable.
- Informational: Answer the question in the first two paragraphs, then add depth the searcher did not ask for.
- Commercial investigation: Put comparison, options, and limits on the page, since the reader is deciding and wants to know what does not work.
- Transactional: One obvious action, one price or next step, and the minimum number of reasons to hesitate.
- Navigational: Do not compete for it. Publish the useful version and let the brand query go to the brand.
Most underperforming pages are not poorly written. They are informational pages trying to close a sale, or product pages buried in history.
A prompt that labels a whole list and groups it
Run this after the classifier to turn a hundred lines into a content plan you can hand to a writer or build into a calendar.
Group the labelled keywords below into clusters. Give each cluster a name, the keywords it contains, the intent label that dominates it, and the one page type it should produce: guide, comparison page, product page, tool page, or FAQ section. Flag any cluster where the keywords have mixed intent, and say whether those keywords should share one page or be split. Rank the clusters by how close they are to a decision the buyer is already making, using only the intent labels as the basis for that ranking. Do not estimate demand or difficulty. Clusters: [paste classified list].
Mixed-intent clusters are the useful finding here. A cluster containing both "bookkeeping software" and "what is bookkeeping" is really two pages, and splitting it is cheaper than writing one page that satisfies neither.
Rewriting the outline once the label is known
Here is a draft outline for [page topic]: [paste outline]. The dominant intent is [label], from these keywords: [list]. Rewrite the outline so the structure serves that intent. Put the answer or the decision in the first section. Order the remaining sections by what a reader with this intent needs next, not by what is interesting to write. Name the format for each section: prose, table, steps, or image caption. Add one line saying which section would fail if the intent label is wrong. Do not add keywords the list does not contain and do not add a call to action unless the intent is transactional.
The last instruction gives you a test. If the page breaks when the label changes, you can see it in the outline, rather than discovering it after publication.
Worked example: fifty keywords, four page types
A small-business owner has fifty phrases from a support inbox and a spreadsheet of search terms. None carry labels yet.
| Group after labelling | Intent | Page produced |
|---|---|---|
| "bookkeeping software for sole traders" | Transactional | Product page |
| "bookkeeping software compared" | Commercial investigation | Comparison page |
| "what is a bookkeeping journal" | Informational | Guide, linked to the product page |
| "bookkeeping software login" | Navigational | No new page |
The owner expected one guide. The labels produce three pages and one decision not to compete for a navigational term. The transactional page is the one that was missing, and the informational guide answers questions the support inbox shows people already ask.
Worth noting the limits of this example. A labelled list tells you what each page should be, not whether that page can rank for the phrase. Ranking also depends on the competition, the site, and factors search engines do not publish, so treat the labels as a planning tool rather than a forecast.
How to check the labels before you brief a writer
1. Read the signal column, not just the label
If the reason given for a label is "the wording suggests", the label is a guess. Check the actual queries your customers send and correct the rows that disagree.
2. Open ten of the phrases and read the current results
Look at what already ranks for a phrase. If most results are transactional for an informational phrase, either your label or the searcher's behaviour is different from your assumption, and that is worth knowing.
3. Check whether one page already owns the intent
If an existing page covers the intent, refreshing it is usually cheaper than writing a new one that competes with it.
4. Split clusters where the label disagrees
Two labels in one cluster mean two jobs. Two pages can do both jobs well, one page rarely does.
Common search intent prompt mistakes
- Classifying after writing: Labeling afterwards justifies whatever you already wrote. Label first.
- Using two labels at once: "Mostly informational" is a decision postponed. Pick one and record your doubt separately.
- Ignoring navigational terms: Building a page to appear for a brand's own name usually wastes the page.
- Treating the label as a ranking promise: Intent describes the reader. It does not describe whether you will win the page.
- Leaving mixed clusters untouched: Splitting them early costs one outline, and saves a page that satisfies nobody.
Frequently asked questions
How many keywords should I classify in one prompt?
Ten to fifty per run. Above that the model starts compressing and you lose the per-keyword reasoning that makes the labels reviewable.
Can ChatGPT tell me the search volume for each keyword?
Not reliably, and you should not ask it to. Volume and difficulty change constantly and live in specialist tools; use a model to classify wording, and take figures from a source that actually measures them.
Does intent decide how long the article should be?
No. Intent decides what the page must contain and in what order. Length follows the job, and a short transactional page can beat a long guide for the same audience.
Label first, then write
Run the classifier on your next keyword list, group the results, and brief one page from the group you were about to write anyway. For more search and publishing prompts, browse the SEO optimization prompt collection.