The AI-resistance filter — six questions I run every new income idea through

Why a filter exists at all

By the end of 2024 there were roughly five times more “passive income ideas” available online than there were categories that worked. AI commodified content production, social platforms compressed organic reach, and the gap between a category’s marketing yield and its realized yield widened across the board. The honest editorial problem became: how do you decide which of the next “make $5K/mo from your laptop” pitches to even take seriously?

The filter below is what we run every candidate idea through before it gets a page on this site. It’s six questions. The first three identify the structural shape of the income. The last three stress-test it against the post-2024 reality. An idea has to pass at least five out of six to make it into the catalog. The ones that pass cluster in surprising places — boring capital plays, narrow audience-driven distribution plays, regulated niches — and the ones that fail almost universally share the same defect.

This is the methodology behind every recommendation on the site. It’s also a tool you can use on any idea you encounter elsewhere.

The six questions

1. What is the actual bottleneck?

Every income stream has one scarce input that gates how much income it can produce. The question is which one. Roughly, the options are:

• Production. “I can write / design / code / film / record / produce faster than the average person.” Until 2023 this was a real, scalable bottleneck. After AI flooded every production category, it’s not. Any income whose bottleneck is “make more stuff than the next person” is now competing against an essentially infinite supply of free-or-cheap competitors.
• Distribution. Audience, email list, search rankings, app-store position, network of buyers. Building distribution still takes years and cannot be AI-shortcut at scale. Anyone with a real distribution channel in 2026 has a defensible position.
• Capital. Money deployed, properties owned, shares accumulated. Capital plays don’t care about AI’s effect on production — a dividend stock pays the same dividend whether or not its annual report could have been written by Claude. Capital is the most AI-resistant bottleneck of all.
• Skill or judgment. Specific decisioning ability that doesn’t yet automate — picking domains that will flip, picking websites to acquire, picking neighborhoods for rental property, picking which of 50 SaaS to buy on Acquire. Survives until the specific skill commodifies, which for most judgment-heavy categories is much further out than people assume.
• Regulatory access or license. Specific permission to operate that AI can’t grant itself — broker license, medical practice, accounting certification, legal practice, banking. Effectively permanent moats.
• Network or relationships. Specific people who buy from you because of trust accumulated over time. Not transferable, not AI-replicable, and not built quickly.

The first filter question: what’s actually scarce here? If the only honest answer is “I’m willing to do production work the average person isn’t,” the category mostly already failed.

2. What is the moat, and is it AI-resistant?

The bottleneck identifies the scarce input. The moat identifies whether the operator can keep that scarcity. AI-resistant moats cluster in five places:

• Capital deployed against a real asset.
• Regulatory or licensing access.
• An audience that trusts you for reasons AI can’t replicate (specific publicly-stated wrong predictions you’ve learned from, specific niches you’ve built credibility in over years, etc.).
• A network of specific people who buy from you.
• A judgment quality demonstrated through a long enough track record to be hard to fake.

AI-vulnerable moats — almost universally — sound like “I produce X faster, better, or cheaper than the average person.” That moat is now actively dissolving across most categories. If your moat description has the words “faster,” “more efficient,” “more scalable,” or “better quality at the same price” without specifying why those advantages will still exist in three years, it’s probably already gone.

3. What is the realistic forward margin (not the current margin)?

The fatal error in evaluating an income idea in 2026 is using current-state economics to forecast three-year economics. The current margins on most production-bottlenecked income categories overstate the future. Etsy POD margins compressed from 25-35% to 8-15% between 2020 and 2024. Generic affiliate site margins compressed from 40-60% to 5-20% over the same period. AdSense CPMs on saturated content compressed 50-70%. The pattern is identical: AI flooding the supply side compressed every category whose moat was production speed.

The forward question: what does this category’s median operator’s margin look like after another 2-3 years of AI-driven supply expansion? If the answer is “materially lower than today,” the idea is buying the top of its category. If the answer is “the same or higher,” there’s something AI-resistant in the structure.

The categories where forward margins stay flat or expand are almost always either capital plays (dividends, P2P, royalty investing), regulated plays (vertical SaaS, licensed services), or distribution plays with real audiences (niche affiliate with genuine expertise, paid newsletters with engaged subscribers).

4. Does the customer-side market still exist?

A subtle failure mode is the category where the producer’s edge collapsed and the customer’s willingness to buy collapsed simultaneously. Dropshipping is the canonical example — not only did the operator’s moat (ad arbitrage) dissolve, but customer trust in single-product Shopify stores collapsed at the same time. The category died on both ends.

Course launches in 2025-2026 have a related problem — operators can still produce a course, but the customer-side appetite for “course as the answer to my career problem” softened materially as AI tools commodified the underlying skill the courses claimed to teach. A course on “learn Python” sells worse in 2026 than 2022 because the buyer correctly estimates AI will close the skill gap faster than the course will.

The fourth filter question: does the customer in this category still exist in 2026-2028, and are they willing to pay the historical prices? For many categories the answer is “yes but for less” or “yes but only for premium positioning.” For some, the answer is “no, this category dispersed into adjacent categories that look different.”

5. What is the median operator’s hourly return after the filter?

Every published “passive income” comparison stacks the headlines. Median realized hourly returns tell the honest story, and they’re consistently 2-5x worse than the marketing copy. The fifth filter question: when you assume the operator is genuinely the median (not the top 5% featured in the marketing), what does the realistic hourly return look like over a 24-36 month window?

This number is what determines whether a category is worth the effort for the person reading the article. A category that produces $50/hour for the top 5% but $5/hour for the median should be filed under “top-decile activity, not passive income.” Most categories that get sold as passive income fail this test.

6. Is this a category, or a feature?

The last and most overlooked filter question. A “category” is a durable business shape that survives multiple platform shifts, technology changes, and competitive cycles — domain investing, dividend investing, owned-and-operated content sites, real estate rental, niche affiliate, paid newsletters. A “feature” is an income that depends on a specific platform’s current rules and dies (or pivots painfully) when the platform changes — Etsy POD circa 2018-2021, AdSense YouTube channels in saturated niches, Amazon FBA brand-arbitrage, TikTok organic reach in 2021-2022.

Features can produce real income for a few years. Categories produce income across decades. The filter question: does this still exist as an income stream if the underlying platform changes its rules? If the answer is “no, this dies the moment Etsy changes its algorithm / Amazon changes its policies / Meta changes its ad rules,” the category is a feature, and should be sized accordingly.

The filter applied — worked examples

Dropshipping (Facebook-Ads-to-Shopify, 2018-2021 era). Bottleneck: production + ad arbitrage. Moat: “I find winning products faster” — AI-vulnerable. Forward margin: actively collapsing. Customer-side: trust collapsed. Median hourly return: negative for >90% of operators. Category-or-feature: feature, depended entirely on iOS 14.4 era Facebook ad costs. Score: 0/6. Pages on this site treat this as a failed category and redirect toward private-label or buying existing stores — see the is-dropshipping-dead-2026 piece.

Print-on-demand on Etsy. Bottleneck: design + Etsy SEO. Moat: niche-brand language and review accumulation. Forward margin: compressed but stabilizing in branded niches. Customer-side: Etsy buyers still exist for niche-branded products. Median hourly return: $5-15/hour for the median, $25-40/hour for the top decile. Category-or-feature: feature (depends on Etsy’s algorithm) but a long-lived one. Score: 3-4/6. Page exists with explicit framing that this is a small business, not passive — see the Etsy POD breakdown.

Dividend ETF compounding. Bottleneck: capital. Moat: literally just having money in productive assets — fully AI-resistant. Forward margin: same as 2015. Customer-side: themselves. Median hourly return: infinite (zero time investment beyond initial setup). Category-or-feature: category, survives every platform shift. Score: 6/6. Page exists, and it’s one of the recommendations we apply our highest endorsement to — see dividend stock portfolios.

Niche affiliate sites with genuine expertise. Bottleneck: distribution (built audience and topical authority). Moat: real first-person review depth, original research, named author profiles — AI-resistant if the human angle is genuine. Forward margin: better than 2022 once Google completed its Helpful Content updates and removed the generic competitors. Customer-side: still robust in non-saturated niches. Median hourly return: $20-50/hour for the median operator with a real site. Category-or-feature: category, but the rules of execution changed sharply. Score: 5/6. Featured on the site — see niche affiliate sites.

Retail forex trading. Bottleneck: skill (a narrow edge). Moat: dissolving against AI-driven institutional execution. Forward margin: roughly flat, but the realistic hourly return was always worse than the marketing implied. Customer-side: 95% of retail traders consistently lose. Median hourly return: negative for the median operator over multi-year windows. Category-or-feature: category, but barely. Score: 1-2/6. Page exists as a blog opinion framing why this isn’t passive income, not as an idea page in the catalog.

Buying a small SaaS or content site. Bottleneck: capital + operational judgment. Moat: capital deployed + cash flow generated by the acquired asset. Forward margin: stable to improving as more inventory comes onto marketplaces post-2024. Customer-side: each acquired asset has its own customers — diligence covers that risk before purchase. Median hourly return: 30-40% cash-on-cash returns at this capital tier. Category-or-feature: category. Score: 6/6. Featured — see buy a website.

Generic AdSense YouTube channels. Bottleneck: production volume. Moat: faster production than the average person — AI-vulnerable. Forward margin: actively compressing (CPMs down ~40% in saturated niches over 2023-2025). Customer-side: ad-driven, supply-side flooded. Median hourly return: $2-8/hour after editing, thumbnail, scripting time. Category-or-feature: feature, depends entirely on YouTube’s algorithm and CPM rates. Score: 0-1/6. Not on the site. The trading-bot, niche-tutorial, and named-personality YouTube categories are different stories and may merit their own evaluation.

Domain investing. Bottleneck: capital + judgment. Moat: pattern recognition that hasn’t commodified (despite AI valuation tools). Forward margin: stable. Customer-side: enterprise buyers still pay 3-5x multiples for the right name on 12-36 month holds. Median hourly return: hard to measure (lumpy income) but real for disciplined operators. Category-or-feature: category. Score: 5-6/6. Featured — see domain investing.

The pattern

Notice what passes the filter and what doesn’t. The categories that score 5-6/6 cluster in:

• Capital plays (dividends, P2P, royalties, REITs, buying assets).
• Distribution plays with real audience moats (niche affiliate with E-E-A-T, paid newsletters, owned-media properties).
• Regulatory-protected services (vertical SaaS, licensed practice, specialized services).
• Judgment-driven asset acquisition (domains, sites, businesses, real estate).

The categories that score 0-2/6 share a common defect: their moat depended on the operator being faster, more efficient, or higher-volume than the average person at producing some output AI now produces for free. Dropshipping, generic AdSense, generic affiliate, KDP self-publishing at scale, course launches into saturated niches, POD without brand-niche specificity, signal services without verified edge, MLMs, copy-trading at scale.

The pattern isn’t moralistic. It’s mechanical: when the production-speed moat dissolves, every category that depended on it follows. The categories that survive have moats AI can’t substitute for. Pick those.

How to use this

When you encounter a new “passive income idea” — in this site, on YouTube, in a course pitch, in a friend’s enthusiastic message — run it through the six questions before forming an opinion. Most ideas fail at question 1 (what’s actually scarce here?) or question 2 (is the moat AI-resistant?). The ones that survive both are worth deeper diligence; the ones that don’t are usually selling you something else (a course, a tool, a referral).

The filter doesn’t kill creativity or new ideas — many new categories will emerge in 2026-2028 that pass it cleanly. RWA tokenization, vertical SaaS in regulated niches, sync licensing on short-form platforms, and several others mapped in our state of passive income piece all score 5+/6. The filter just keeps you out of the 60-70% of currently-marketed categories that already structurally failed and haven’t priced it in yet.

The honest passive-income map in 2026 is narrower than it used to be, and harder to fake. The filter is how we keep this site narrower than the easy version of it would be.