How to find what people will pay for, with AI: the buyer-question method
Last updated · By Orenston
Disclosure: some links on this site are affiliate links, marked "sponsored". If you buy through one, Orenston may receive a commission at no extra cost to you. It is general information, not a forecast of results. Many people never earn a commission.
Short answer: Start from questions real people already ask, not from a product. Use AI to cluster those questions by intent, check who answers them (in search and in AI answers), match a program you're approved for, and score each cluster before you build. AI does the sorting; the inputs must be real.
The method (7 steps)
- Pull real questions. Export the queries your site already gets from Google Search Console (last 28 days), or collect questions people ask in public forums. Note the date window.
- Cluster by intent. Have your AI assistant group the questions by what the person wants: compare, price, deal, how-to or find a provider.
- Check who already answers. Search each cluster's main question. Note who ranks and what format wins: a data table, a price index, a short video.
- Check AI answers. Ask the same question in two or three AI search tools and write down which sources they cite.
- Match a program. Attach a product or service you are approved to promote. If none fits, the cluster is content-only for now.
- Check the terms. Make sure the tactic you plan (coupon page, paid ads, comparison) isn't banned by that program.
- Score and decide. Score each cluster 0-3 on six factors, then mark the top three build, test or skip.
Score each cluster (0–3 per factor, 18 max)
| Factor | 0 | 1 | 2 | 3 |
|---|---|---|---|---|
| Measured demand | none | a little | some | a lot (your own data) |
| Ranking gap | not ranking | page 3+ | page 2 | page 1, low click rate |
| Competitor weakness | strong, data-rich pages | some gaps | thin or undated | no one answers it |
| AI-citation opening | authorities cited | mixed | thin sources | no good source cited |
| Program fit | no approved program | weak fit | good fit | good fit + clear attribution |
| Terms safety | tactic banned | unclear | allowed with limits | explicitly allowed |
Two hard rules: Program fit = 0 can't be "build", only "test" (content without links) or "skip". Terms safety = 0 is always "skip".
A prompt to start with (free)
Here are {N} real search queries from {date window}. Group them into clusters by what the searcher wants (compare, price, deal, how-to, find a provider). For each cluster give: a name, the queries in it, the intent, and one question a single page could answer. Don't add queries that aren't in my list, and don't estimate search volume.
[GATED · Starter] Prompt pack and scoring sheet
Members get: 8 prompts (clustering, competitor format check, AI-answer check, terms check), the scoring sheet as a spreadsheet with the hard rules built in, and a worked walk-through on a sample query list.
[GATED · Operator] Agent runbook
Members get: a runbook for running this weekly with an AI agent: pulling Search Console data through an MCP connector, writing one JSON record per cluster, and a human review step before anything is built or published.
FAQ
How do I find a profitable niche with AI?
Don't ask an AI to guess. Give it real questions people already ask, from your Search Console data or public forums, and have it cluster them by intent. Then check who answers each one and whether a program you're approved for fits.
Can AI estimate search volume?
Not reliably on its own. Use measured data, such as your own Search Console impressions or a keyword tool, and label anything you didn't measure as unmeasured.
What if no affiliate program fits the demand?
Treat it as a content-only test, or skip it. A cluster with no approved program shouldn't be built as a money page.
Related: How to choose an affiliate program · Get cited by AI search