Teams keep asking whether they need a second research program for the AI era. No: keywords and prompts are the same intents wearing different lengths, and a mapped program, one row per intent, two phrasings per row, serves both surfaces from one effort.
Here is the mapping, the page anatomy that serves both, and the paired scorecard that tracks it.
The mapping table
| Keyword class | Its prompt twin | The page both want |
|---|---|---|
| best [category] | "what should I use for [situation]?" | Ranked list with criteria and picks per use case |
| [X] vs [Y] | "[X] or [Y] for someone like me?" | Comparison with a verdict per situation |
| [X] alternatives | "I want to leave [X], what instead?" | Alternatives list organized by leaving-reason |
| how to [task] | "walk me through [task]" | Numbered steps that survive excerpting |
| what is [term] | "explain [term] simply" | Definition up top, depth below |
| [X] worth it | "should I actually pay for [X]?" | Honest assessment with the deciding factors |
Read the third column twice: it is one page per row, not two. The keyword phrasing and the prompt phrasing are entrances to the same intent, and the page that satisfies the intent serves both doors.
The workflow: inflate, do not duplicate
Start from your keyword list, because that is where the demand data lives. Take the priority intents and re-inflate each into prompt form using the persona-constraint method: who asks this, with what limits, in what situation.
Prioritize the overlap: intents with real search volume and a recommendation-shaped prompt twin. Those rows are where classic traffic and AI presence pay from the same page, which is the entire economic argument.
The page anatomy that serves both
Title and H1 carry the keyword phrasing, because rankings still read them first. The opening block answers the prompt directly, the verdict or the shortlist in the first lines, because that is what gets extracted into answers.
The body covers the intent's fan-out: the sub-questions, the situations, the trade-offs, which serves deep readers, long-tail rankings and AI Mode's decomposed searches simultaneously.
And the FAQ carries the prompt phrasings almost verbatim as question headings. That section is where the two vocabularies formally meet: humans skim it, engines quote it, and the conversational wording finally gets to exist on the page without wrecking the title.
The paired scorecard
Tracking follows the same one-row logic: each intent gets two cells, its ranking and its answer presence.
| The pair reads | It means | Fix |
|---|---|---|
| Ranks well, never cited | Findable but not quotable | Add the extractable answer block |
| Cited, ranks poorly | Quotable but weak authority | Classic SEO: links, internals, refresh |
| Neither | The page misses the intent | Rebuild against the mapping row |
| Both | The system working | Copy its pattern to the next row |
Rankings come from your normal tracker, answer presence from the citation report and prompt tracking. The diagnostic power is in the mismatch rows: each one names which half of the page needs work, which is more than either metric says alone.
One program, quarterly
Run it on the research cadence you already have: quarterly, refresh the keyword data, re-inflate the new priorities into prompts, update the pairs. The output stays one list, one set of pages, two doors served, and no second program to maintain.
The one-line takeaway: every keyword class has a prompt twin, so map them: keyword in the title, prompt answered in the first lines, fan-out in the body, prompt phrasings in the FAQ. Track each intent as a rank-plus-presence pair and let the mismatches assign the work.