Is that AI spending actually expensive for your life? I built a standard household from official statistics and ran the numbers

Part 1 established that my AI spending is 7.7 times the average, and Part 2 built a ruler: judge it as a ratio to take-home pay, with 5% as the guideline ceiling and 10% as the AI-poverty line.

But something nagged at me the whole time I was writing Part 2. Does a ratio to take-home pay really describe someone's life?

The weakness of a take-home ratio: it ignores rent and family size

Part 2's ruler was AI spending divided by take-home pay. Clear, and instant to compute. But take two people both taking home 300,000 yen:

• Living alone in Tokyo, rent 75,000
• Living with family, rent 0

If both pay 30,000 a month for AI, both are at 10%. Part 2 gives them the same verdict: the AI-poverty line. The weight on the household is obviously different. The second person has that entire 75,000 sitting in their free money.

Family size does the same thing. A household of two adults and two children spends more than twice as much on education and food as someone living alone. Taking 30,000 out of that budget for AI cuts something different than it does for a single person.

So what we actually want isn't "what percentage of take-home pay" but "what percentage of what's left after living costs and savings". In Part 2's 50/30/20 terms: how much room AI takes inside the 30% you spend freely.

The problem is that asking every reader to enter their living costs is unrealistic. Food, utilities, communications, transport... told to fill in every field, I would close the tab.

Generate "your standard household" from statistics first

So I flipped it. Rather than asking for living costs, generate a standard household close to that person from statistics, show it, and let them correct only what's wrong.

Two official statistics from Japan's Statistics Bureau.

• Family Income and Expenditure Survey (2025, income and expenditure edition). What a household spends per month, by family structure: food, utilities, communications, culture and recreation, social expenses and so on
• Housing and Land Survey (2023). Monthly rent by prefecture and household size

The household survey alone can't give rent: it mixes in homeowners, so housing costs come out very low. So rent alone comes from the Housing and Land Survey, by prefecture and household size.

The reader enters five things: monthly take-home pay, prefecture, family structure, housing type (rented, mortgage, owned outright) and total monthly AI spending. From that, a standard household appears.

For example, 300,000 yen take-home, Tokyo, living alone, renting (yen per month, adjusted to take-home pay from the statistics):

Add the savings guideline (a tenth of take-home pay, matching J-FLEC's "a tenth of annual income"), 30,000 yen, and 41,428 remains. That is the money you can actually spend freely.

If your food bill is lower, or your phone plan is 3,000 yen, correct that field and it recalculates. You don't have to fill in everything.

Where I didn't use the statistics as-is

Honestly: used raw, the statistics broke the checker.

The household survey averages come from households with disposable income (close to take-home pay) of about 310,000 a month for one person, 490,000 for a couple, and 600,000 for a couple with two children. So anyone earning below average blows past their take-home pay on default living costs alone, and lands in the top verdict after paying one yen to an AI. The first time I tried it, a couple in Osaka on 400,000 take-home paying 15,000 for AI got "you are working for the AI", which is obviously wrong.

So every item except rent is scaled by "reader's take-home pay ÷ the statistics' disposable income" (clamped between 0.5x and 1.5x so it can't go extreme). Rent uses the prefecture's statistic as-is, because rent does not obligingly fall in proportion to your income. The scaling factor is shown on screen.

One more thing. The household survey's "culture and recreation" already includes that household's entertainment spending. AI spending might belong inside it, but this checker treats AI as paid on top of the standard household. If you're replacing entertainment with AI, reduce the culture and recreation field accordingly.

The detailed processing (which table, which column, adding vehicle costs into transport, how single-parent households are handled) is at the end of this post.

So how does it land for your life?

The checker itself is Japanese-only for now → AI subscription poverty checker v2. Nothing you enter is sent anywhere; it calculates inside the page. The five inputs are take-home pay, prefecture, family structure, housing type and monthly AI spending.

The verdict lines, like Part 2's, are this blog's own. Up to 15% of your free money is "ordinary", past 33% (a third) is "deep in it", past 50% is "AI subscription poverty", past 100% is "working for the AI". Calling it poverty when more than half your free money disappears into AI seems a reasonable place to draw the line.

The same 10% of take-home pay, judged differently

Taking the earlier example at 30,000 yen a month for AI:

The same 10% of take-home pay shifts the verdict by one level depending on whether the rent is Tokyo rent. And the 5% that Part 2 called "nothing to worry about" exceeds a third of free money for someone living alone in Tokyo, landing on "deep in it". The take-home ratio was too lenient for people living alone in expensive cities — that's what this exercise found.

I'm freelance, so I'll run it against averages

Comparing myself to an average salaried income isn't very meaningful, so I ran the numbers as an average Japanese household with a head of household around 35.

In the 2025 Family Income and Expenditure Survey, worker households with a head under 40 have an average head-of-household age of 34.3. Close enough to a 35 model. Average household size is 3.20 people and disposable income is 541,537 yen a month (Statistics Bureau, summary of 2025 average results).

In the checker I picked "couple with one child" as closest to 3.20 people, and used the national average rent for a three-person household, 74,935 yen, as-is.

• Head of household 35
• Couple with one child, three people
• Household take-home 541,537 a month
• Renting, rent at the national three-person average of 74,935
• AI spending the same as mine: 44,408

The National Tax Agency's age-banded data puts average pay for 35 to 39 at 4.82 million yen a year. That's one person's gross salary, not a household's take-home, so I used the household survey's disposable income, which goes straight into the checker (National Tax Agency, 2024 survey of private-sector salaries).

This isn't a reconstruction of my life. It's what my AI spending looks like dropped into an average household around 35.

The result:

On the take-home ratio alone it's 8.2%: Part 2's "deep in it". Subtract average living costs and savings first, and more than half of the 80,656 in free money goes to AI.

The standard household's 32,507 yen of culture and recreation may contain spending that AI has replaced. Zero that field out entirely and free money becomes 113,163, with AI at 39.2%. The verdict drops to "deep in it".

That said, when a freelancer uses AI for work, it belongs in business expenses or investment, not household entertainment. If you already deduct AI costs as a business expense before arriving at take-home pay, entering the same AI cost into this checker double-counts it.

Dropped into an average household, my AI spending is heavy. But that alone doesn't make me AI-poor. In the end, as in Part 1, it comes down to ROI: am I producing more work and time than the AI costs?

The limits of the statistics, stated

• "Couple with one child" and "couple with two children" are approximated by three- and four-person households, because the household-type tables don't break down by number of children
• "Single parent with children" uses the survey's "households of one parent and unmarried children"
• Rent is the average excluding zero-rent households. If you live in company housing or with family, choose "owned" for housing or set rent to 0
• The statistics are from 2025 (household survey) and 2023 (housing survey). Prices move, and these drift
• The one-tenth savings guideline and the verdict lines (15/33/50/100%) are, as in Part 2, this blog's own

So who is this for?

• If Part 2 told you "under 5%, you're fine" → if you live alone somewhere with high rent, check again here
• If you have a family → this should feel closer to reality than a take-home ratio
• If you're "recovering it at work" → as in Parts 1 and 2, this checker isn't for you. Use ROI

That closes out the AI subscription poverty series.

• Part 1: Can you actually go broke on AI subscriptions?
• Part 2: Is AI subscription poverty real? A spending line from household benchmarks
• Part 3: this post

The statistics used, and how they were processed

• Family Income and Expenditure Survey 2025, income and expenditure edition. Table 1 for single-person households, table 3-1 by household size, table 3-6 by household type. All using the "worker households" column (the all-household average includes unemployed and elderly households, which drifts from a working reader's budget). The denominator for the take-home scaling is that column's disposable income
• Housing and Land Survey 2023, table 123-1, monthly rent by building type and household size. The average for rented main households excluding zero-rent, by prefecture, for one to four people and five-plus
• "Transport" is the survey's transport plus vehicle-related costs. "Everything else" is other consumption expenditure minus social expenses. Everything else is the table value as-is
• The survey's "housing" figure is not used (homeowners in the mix push it down)
• The original Excel files downloaded from e-Stat, their SHA256 hashes and the normalisation script are in the blog's repository. The numbers are not padded