When I validated the niche for this site, the first answer I got about its economics was wrong at the core. An AI, answering from memory, gave me a confident estimate of how many affiliate programs I could join. Checked against each vendor’s own program page, several of the best-known products in the niche had no program open to affiliates at all.
If I’d trusted that answer, I would have bought a domain and planned months of content around commissions that didn’t exist.
That’s the risk with validating a niche by asking an AI. A model asked “does this niche have potential?” will always give you a fluent, sensible-sounding yes-with-caveats. It has no stake in your decision and no way to know which of its numbers went stale last month. Picking a niche is the most expensive decision a solo builder makes, because a wrong pick costs months of content for a market that’s too small or one you can’t reach. You pay for that mistake with your own time.
This workflow keeps AI where it’s strong and out of where it’s dangerous. You set the criteria and collect the evidence. The AI analyzes only what you hand it. And the final call waits for a signal from real people.
I didn’t buy the leversolo.com domain until the research below was done.
The six steps
- Write pass/fail criteria, with thresholds, before you research.
- Collect the raw evidence yourself: SERPs and program terms.
- Have AI analyze that evidence, and mark what it’s guessing.
- Check every number that matters against the official source.
- Test reach and economics separately.
- Run a small real-world test, then commit.
Step 1: Write your criteria first, with thresholds
The most common mistake is setting the bar after you’ve seen the results. A number looks good, so it becomes the reason the niche “passes”. Write the criteria down first, each with a threshold, so that when results come in you only have to compare.
For a solo content and affiliate niche, five criteria is the minimum:
- Is there real buying intent? Are the core keywords the ones people type when they’re about to spend money (“best…”, “X vs Y”, “X alternatives”, “X pricing”), and do commercial pages show up in those results?
- Can you get in? On page one for those keywords, does any spot belong to a small independent site or a forum? If so, the page isn’t locked up by big publishers and the vendors themselves.
- Do the economics work? Affiliate income comes from a mix: average order value, commission rate, how many orders you can drive, and whether the commission recurs or pays once. Recurring commission steadies cash flow, but a one-time payout at a high rate on an expensive product can easily beat a small recurring one. The real test is how many conversions you need to reach your minimum income with this niche’s actual mix, and whether that number is within reach.
- Where is the audience, and can you reach it? List every traffic source the niche could use: search, communities and forums, social, video, email, referrals from other sites. For each, ask whether this niche’s audience is really there and what reaching them costs you in time or money. What counts as acceptable depends on your budget and stage. The criterion asks for one thing: at least some of these sources must be ones you can actually run, in your situation.
- Can you keep going? Do you have enough material to publish distinct, useful content in this niche for a long time, or will you run dry after a handful of articles?
Put them on one sheet before you start, with the threshold filled in and the result column empty:
| Criterion | Pass threshold (yours to set) | Evidence | Result |
|---|---|---|---|
| Buying intent | e.g. commercial pages in the top 5 for most core keywords | SERP log | |
| Can you get in | e.g. a forum or small site in the top 5 for at least 3 of 10 keywords | SERP log | |
| Economics | e.g. monthly target reachable with no more than N conversions | Program pages, pricing pages | |
| Reach | e.g. at least two traffic sources you can run today | SERP log, communities | |
| Depth | e.g. 30+ distinct article ideas from real questions | SERP log, “People also ask” |
The example thresholds are only there to show the shape. The exact numbers matter less than writing them down before you look.
Step 2: Collect the raw evidence yourself, before you ask AI
An AI answering from memory blends years of data: old prices, expired terms, programs that closed but that it still describes as open. The fix is the order of operations, evidence first and AI second. For niche validation, the raw evidence comes in two kinds:
- SERPs for the niche’s buying-intent keywords. You’re sizing up the competition: who holds the top spots (vendors, big publishers, small independent sites, forums, video), what kind of content wins, and whether there are ads. Note the date, time, and search settings, because results change with location and over time.
- The affiliate program pages and pricing pages of the products in the niche. That’s where commission rates, cookie windows, payout terms, and whether a program still accepts sign-ups actually live.
A note on scope: this evidence answers one question, whether the niche is viable. The prices and features of the tools you use to research belong in the tool section at the end.
A simple log keeps the SERP work honest. One row per keyword:
| Keyword | Date, time, location | Top 3 results are… | Forum or small site in top 5? | AI Overview? | Winning format |
|---|---|---|---|---|---|
| best X for Y | Nov 9, 9:00, US | vendor, publisher, forum | yes | yes | comparison |
From a real run: when I applied this step to the AI tools niche for solo builders, the SERP pass covered 10 buying-intent keywords, captured in one morning. The two most useful details weren’t ones I’d predicted: Google’s AI Overview appeared on all 10 results pages, and a forum held one of the top two spots for 9 of the 10 keywords. Both changed how I judged the later steps.
Those two details pushed the later steps in specific directions. Forums near the top suggested Google was rewarding first-hand experience on these queries, and that the audience gathers in communities, which makes forums a reach channel as well as a competitor. AI Overviews on every page meant plain informational content would get fewer clicks, so the opening was in content an AI summary can’t replace: step-by-step workflows and fresh, checked facts.
Step 3: Have AI analyze the evidence, and know what it’s guessing
Ask an AI one question and you get two kinds of output with very different reliability.
When it answers from memory, it’s fast, fluent, and unreliable on every specific number. How many affiliate programs exist, what they pay, what a plan costs: any of it can be wrong or out of date. Use this mode only to suggest a structure for your analysis and a list of things to verify.
When it works on the evidence you just gave it, you feed in your SERP log, the program terms, and the pricing pages, and ask it to cluster, compare, and find gaps using only those documents. Every result traces back to a source, so you can check it.
Treat anything from the first mode as unverified until it matches a source document. This prompt enforces that:
Here is my SERP log, the affiliate program pages, and the pricing pages for this niche.
1. Group the keywords by intent and by who currently ranks.
2. For each of my five criteria, say what this evidence shows.
3. List the sub-topics these results cover. Is there enough for a long-running site?
4. Answer only from the documents. Mark anything else as UNVERIFIED.Point 3 checks criterion 5, the one that’s easy to forget. The sub-topics and “People also ask” questions in your SERP evidence show how deep the niche goes.
The AI does the analysis. The conclusions come in the next step, from you.
Step 4: Check it against official sources, and keep only what matches
Compare both kinds of output from Step 3 against the source documents, and drop anything with nothing behind it. Three kinds of claims go wrong most often:
| Claim | How AI usually gets it wrong | Where to check |
|---|---|---|
| Prices of products in the niche | Mixes years, merges plans | The official pricing page, with the date you checked |
| Affiliate terms | Invents cookie windows, calls closed programs open | The vendor’s public affiliate program page |
| What the products can do | Repeats marketing copy as fact | Official docs and help pages: real features and real plan limits |
Product capability isn’t one of the five criteria, and it’s easy to skip for that reason. It checks the promise the whole niche rests on: buyers pay for a product to do a job, so if the marketing says one thing and the docs say another (especially about the limits of the plan most buyers will pick), the demand you measured in Step 2 is built on an inflated promise. Your economics in Step 5 will be off too.
One practical point about affiliate terms. Before you’ve chosen a niche, you almost certainly don’t have accounts in the affiliate networks, so you can’t look inside them the way many guides suggest. At this stage the right evidence is the vendor’s public program page, where rates, cookie windows, and payout terms are published to recruit affiliates.
Later, once you’re approved: established networks show a second layer that public pages don’t, such as real performance data (earnings per click, average conversion rate), tiered and unpublished bonuses, detailed rules on allowed channels (search ads, coupon sites), payout schedules, and creative assets. It won’t change your niche verdict. Use it to re-check terms before you add links and to rank programs against each other. Small in-house programs often have no such layer, so the public page is all you’ll get; note that limit.
From a real run: in validating the AI tools niche, the first estimate of how many affiliate programs were available came from the AI answering from memory, and it was wrong at the core. Checked against each vendor’s public program page, several very well-known products turned out to have no program open to affiliates at all. Had I trusted the first answer, the entire economics calculation would have rested on a list that didn’t exist.
Step 5: Test reach and economics separately
A niche with real demand can still be one you can’t work, for two separate reasons. Check them one at a time.
Reach. From the SERPs and communities you looked at in Step 2, where does this niche’s audience gather, and what do they read? Compare that with the traffic sources you listed under criterion 4. Which can you actually run, given your setup, your location, and each platform’s access rules? Cross off the ones you can’t use now, before you’ve produced anything for them.
Economics. Take the mix from criterion 3 (order value, commission rate, whether it recurs) and multiply by the number of conversions you can realistically expect early on. If the math only works when every assumption goes your way, the niche fails, even if the demand is real.
Step 6: Run a small real-world test, then commit
Steps 1 to 5 are research. They rule out the obviously wrong niches, and none of them can confirm a niche is right, because the market hasn’t said anything yet.
So before you commit for the long haul, test small on the distribution channel you actually plan to use. Publish a few real pieces, take them where the niche’s audience gathers, and measure against the criteria you set in Step 1: do people show up, do they click through to products, how do communities respond? Set a deadline and thresholds for the test in advance, and let that result make the decision.
“Small” can mean a whole website, if the website is cheap enough. When hosting costs nothing and the only fixed cost is a domain, a lean site is the test. What you’re really committing is months of production, and that’s the part to hold back until the signal comes in.
If the test fails, you’ve finished an experiment at a low price, months earlier than you’d have learned the same thing by grinding out content.
A recorded run
Here’s what happened when the full workflow was run on the niche of AI tools and automation for solo builders. Treat it as a worked example with real data, not a conclusion to copy.
The desk research took about two days, and the domain was bought only after it was done: comparing niche options, checking affiliate programs on each vendor’s public page, and reading buying-intent SERPs.
Result of Steps 1 to 5: a conditional pass. Buying intent is real, and in the automation topics page one isn’t locked up; small independent sites still hold spots. In SEO tools and video tools, the vendors themselves take most of page one, so a new site can’t compete head-on with “best tools” lists, and the opening there is workflow and how-to content. On economics, the strongest recurring programs are in automation, but part of the original product list had to go because those products had no open program.
The remaining condition is Step 6: the signal from the site’s first 30 days live. That test starts the day the site goes live.
Choosing tools for this workflow
The six steps don’t depend on any particular tool. They need three capabilities: search and synthesis with citations, a way to read SERPs, and a way to look up terms. Pick tools by budget and workload.
Free tier: enough to validate one niche. A free AI research tool (Perplexity’s free plan, for example) handles the analysis of the evidence you feed it. Google Search, checked by hand, covers SERPs. Google Trends gives a rough read on demand. Once your site is live, Search Console shows the real queries people use to find you.
Paid tier: when volume grows or you need hard numbers. Perplexity Pro is one option for the research side. According to Perplexity’s official help page (updated October 28, 2026), Pro’s current top models include Claude Sonnet 5.5, GPT-6.1 Sol, and Gemini 3.8 Flash, among others. The list changes over time, and the model picker in the app is the final word. Perplexity’s official pricing page on November 9, 2026 lists $17/month billed annually. A $20/month monthly price doesn’t appear on that page; it’s reported by independent sources, so if you quote it, label it that way. Perplexity has no open affiliate program (its Comet Referral program ended on November 6, 2025, per official documentation), so it’s here because it’s useful for research, and we earn nothing if you buy it.
When you need search volume and competitor data at a larger scale, a dedicated data tool like Semrush starts to earn its price. Semrush’s lowest plan is $139.95/month billed monthly (the “SEO” plan, checked on its pricing page on November 9, 2026). For a single niche at the validation stage, hand-checked SERPs are usually enough, and that money is better spent once you have cash flow.
When to stay free, when to upgrade
Stay on free tiers while you’re validating your first niche, running a few research sessions a week, and every decision in front of you can be checked by hand. That’s where most solo builders are at the start.
Upgrade your research tool when synthesizing many sources becomes a daily job. The clearest sign is spending more hours copying data around than analyzing it.
Upgrade to a data tool when your next decision needs numbers that hand-checked SERPs can’t give you (choosing between several candidate niches at once, for example), and when the tool costs less than the wrong decision it helps you avoid.
This workflow won’t hand you a “winning niche”. It gives you a decision backed by evidence, thresholds you set in advance, and a clear point to stop. Start with the criteria sheet in Step 1: copy the table, fill in your thresholds, and leave the result column empty until the evidence is in.
Next: if automation passes your test, Make vs n8n for solo builders covers the first tool decision. See how we check facts on the Methodology page, or browse the AI Research Workflow pillar.