Why Your Insurance Website's Subsidy Calculator Is Wrong

The number your calculator gave out last week

Most agency websites built in the last few years have some version of a quote tool on the ACA page. Sometimes it is a real calculator with a few input fields. More often it is a static example, a number typed in once when the page was built: "clients like you save up to $400 a month," or a little slider that spits out an estimate against numbers nobody has opened since launch. Either way, somebody built it once, and then the agency moved on to selling.

That was a survivable shortcut for a few years. It is not one anymore. The temporary premium tax credit enhancements that Congress passed under the American Rescue Plan and extended under the Inflation Reduction Act expired at the end of 2025, and the subsidy cliff at 400% of the federal poverty level, the hard ceiling those enhancements had removed, is back for plan year 2026^1,2. A household that priced coverage on your site's calculator using last year's assumptions, or using a formula frozen from whenever the widget was built, is not getting an approximation. It is getting a number computed under rules that stopped applying on January 1.

This is not a marginal shift. KFF's tracking of the 2026 marketplace found the average monthly premium payment for subsidized enrollees rose 58%, from $113 in 2025 to $178 in 2026, alongside a 37% increase in average deductibles to a record $3,786¹. Those are national averages across every income band, not a worst case. For a specific household sitting near the cliff, the swing is far larger, and it is the swing this guide walks through in real numbers a few sections down.

Why a subsidy formula cannot be hardcoded once

The premium tax credit is not a fixed dollar amount you can bake into a page template and leave alone. It is the output of three separate inputs, and every one of the three changes on its own schedule, independent of the other two.

The federal poverty guideline is republished by HHS every January, and it moves every single year, not by a rounding error but by a real percentage tied to consumer prices². The benchmark premium, the second lowest cost silver plan in a household's county or rating area, is repriced by insurers every plan year and can move sharply county to county, which is exactly why KFF's own calculator draws directly from CMS and state exchange filings rather than a single national number⁵. And the applicable percentage table, the schedule that converts a household's income as a percent of poverty into what it is expected to contribute toward that benchmark premium, is republished annually by the IRS in a revenue procedure sized for the coming tax year⁹. This year that table did not just get new numbers. It reverted to the ACA's original statutory structure now that the temporary enhancement expired, which reinstates the flat cutoff at 400% of poverty where a household received no credit before the temporary law lifted that ceiling starting in 2021.

A calculator that hardcodes any one of those three numbers is accurate for exactly one plan year, and only if nothing about the underlying law changes in the meantime. A calculator that hardcodes all three, which is what most static widgets do, was already fragile. This year it broke on schedule, the moment the enhanced credits expired and the cliff came back.

How the premium tax credit is actually calculated

It helps to see the mechanism plainly rather than take "the formula changed" on faith. The premium tax credit equals the benchmark silver plan premium for the household's area minus the household's required contribution, and the required contribution is the household's income multiplied by a percentage set by the applicable percentage table, based on where that income falls as a share of the federal poverty level.

Notice what is missing from that table: a single number you could type into a form once. Every row is a moving part, and the last row is the one that changed most for 2026, since the "required contribution exceeds the premium" case did not produce a zero credit for the several years the enhanced subsidies were in effect. It produces one now.

The cliff, in real numbers

KFF worked an example that makes the mechanism concrete rather than abstract: a 60-year-old couple earning $85,000 a year, about 402% of the federal poverty level, just over the line, sees yearly premium payments rise by more than $22,600 once the enhanced credits expire, because they go from receiving a substantial credit to receiving nothing at all³. That is not a gradual taper. Two percentage points of income either side of that line is the entire difference between meaningful help and none.

Those two right hand figures are simple arithmetic on the published 2026 guideline, not a separate government estimate, and that is deliberate. Anyone can check that math against the Federal Register notice in about a minute⁴. A household applying for coverage with income just under one of those numbers is looking at a real credit. The same household a few thousand dollars over it is looking at none. A calculator that does not know where that line sits this year cannot warn a prospect they are near it, which is exactly the conversation an agent should be having before the application, not after.

What actually changed for 2026

It is worth separating what changed from what stayed the same, since not every part of ACA subsidy math moved this year.

The benchmark premium itself, the SLCSP, also moves every year independent of any of this, since it is set by what insurers actually filed for the coming plan year in each rating area, which is one more reason a number typed into a page template during a previous plan year cannot be trusted without checking it again.

One more variable: which exchange you are even quoting

There is a fourth moving part most agency calculators skip entirely: which exchange a household actually shops on. CMS's own 2026 national snapshot splits the 23.0 million total plan selections into 15.8 million through HealthCare.gov, the federal platform covering 30 states, and 7.2 million through the 20 states plus DC that run their own state-based exchange, each with its own eligibility and enrollment system⁶. A calculator built once around HealthCare.gov's federal poverty guideline and federal applicable percentage table is still using the right numbers for a household in a state-based exchange state, since the federal poverty guideline and the tax credit formula itself are national. What differs by state is the benchmark premium data feed and, in a handful of states, an additional state-funded subsidy layered on top of the federal credit.

California, New York, and several other state-based exchange states run their own supplemental subsidy programs that reduce premiums further for residents who qualify, on top of whatever the federal premium tax credit calculates. A calculator that only knows the federal formula will understate the actual help available to a household in one of those states, the mirror image of the more common problem this guide has been describing. If your agency writes business in more than one state, or in a state-based exchange state specifically, that is one more reason a single hardcoded formula cannot serve every visitor correctly, and one more reason the fix has to be a live data connection rather than a page someone edits once a year.

What a wrong number costs an agency

A wrong quote is not a rounding problem for the prospect who gets one. CMS's own final snapshot for the 2026 open enrollment period recorded 23.0 million plan selections nationwide, through January 15, 2026⁶, and KFF separately estimates plan sign-ups actually fell to about 23.1 million, down from 24.2 million in 2025, with projected effectuated enrollment dropping by 4.8 million as higher net premiums pushed some applicants away entirely¹. A share of that decline is people who priced coverage against a number that turned out wrong once they actually applied, got sticker shock at the real figure, and walked away rather than come back to the agent who quoted it.

For the agency, the immediate cost is time. A prospect who planned around your site's estimate and then hears a different number from the Marketplace application calls back confused, sometimes annoyed, and the producer spends the call re-explaining math the website should have gotten right the first time. The slower cost is trust. An insurance shopper comparing three or four agencies is already skeptical of round numbers and vague savings claims. A specific number that turns out wrong is worse than no number at all, because it reads as either careless or misleading, and a prospect rarely stays to find out which.

There is a compliance angle worth naming plainly rather than dodging. TPMO marketing rules and state insurance advertising regulations generally require that quoted figures be accurate and not misleading. A calculator running last year's formula was not built to deceive anyone, but "the widget was outdated" is a weak answer if a regulator or a client ever asks why a published estimate was materially wrong. The fix here is not a legal opinion, it is simply keeping the underlying data current, which removes the question before it gets asked.

How to check what your site is doing right now

Before anything else, find out what your own site is actually running. This takes about ten minutes and does not require touching any code.

The honest DIY fix

You can do a version of this yourself, and for a single producer with a light site and no development budget, it may be the right stopping point. The honest fix is not to rebuild a live calculator from scratch. It is to stop pretending a static widget is one.

If your site currently states a specific dollar figure or percentage, the lowest effort fix is to replace it with a link to a calculator that is actually kept current. KFF's Health Insurance Marketplace Calculator pulls current CMS and state exchange premium data, was last refreshed with 2026 premiums on March 16, 2026, and states its own limitations plainly, including that it relies on user-entered income rather than a verified figure and does not add a tobacco surcharge⁵. HealthCare.gov's own eligibility tool does the same for Federally-facilitated Exchange states. Neither is a fake widget. Both are honest about being an estimate, which is the standard your own site should meet too.

If you would rather keep a tool on your own domain, the minimum discipline is a yearly calendar entry: recheck the federal poverty guideline every January when HHS republishes it, recheck your state's benchmark premium data at the start of open enrollment, and confirm the applicable percentage logic against the current IRS revenue procedure before the new plan year opens. That is three checks a year, not a rebuild, and it is genuinely better than nothing. Where it falls short is a benchmark premium that reprices mid-year in some states, or a household near the cliff who needs a precise answer rather than a once-a-year approximation.

There is also a manual version of this that costs nothing but a few minutes on a call. Rather than quoting a number from memory or from an old page, pull up the current federal poverty guideline for the household's size, note how many percentage points of poverty their stated income represents, and walk them through whether that lands them comfortably under the 400% line, close to it, or over it. A prospect who hears "here is exactly why you qualify, and here is the number your income would need to cross before you would not" trusts that explanation more than a black box widget ever earns, wrong or right. The transparency is the product here as much as the number itself.

What a real fix requires

CMS holds its own approved partners to a higher standard than an annual manual check. Its Enhanced Direct Enrollment program requires primary entities to integrate with more than 20 real-time APIs covering eligibility determination, plan comparison, and enrollment, specifically so a consumer working with an approved partner sees the same live numbers HealthCare.gov itself would show⁷. That is the honest benchmark for what "real time" means in this space. A calculator is only as good as how current its underlying feed is, and a feed that updates once a year, or once at launch, is not that.

This is the exact problem the Brain, our data engine, exists to solve. It stays connected to CMS, the Census Bureau, HealthCare.gov, and dozens of other federal sources, so a calculator built on it is not running a number someone typed in during a previous plan year. Every build we ship includes calculators wired this way rather than a decorative form with an example number, verified live on the Digital Foundation pricing page, where the Website tier starts at $84 a month¹⁰. Our own custom builds page states the same standard plainly: quoting tools "backed by live government data, so the numbers are right for their county, not a fake widget," scoped on a call for an agency that needs something more specific than the base tier¹¹.

It is fair to ask whether this is worth fixing if your agency is small and your traffic is mostly referral based. If nobody is landing on your ACA page cold and pricing coverage against your calculator before ever talking to you, the honest DIY fix above, checked once a year, is a reasonable stopping point. Where this matters most is an agency actively running paid traffic or organic content to a marketplace or subsidy page, where a stranger is forming their first impression of your accuracy from a number your site generated with no human in the loop.

What changes once it is accurate

Strip away the mechanics and the outcome this guide is describing is narrow. A prospect who prices coverage on your site sees a number close to what they will actually see on the real application, which means the conversation you have with them starts from agreement instead of correction. The calculator updates itself every January when HHS republishes the poverty guideline, and every plan year when benchmark premiums reprice, without anyone on your team remembering to open the file.

There is a second, compounding benefit worth naming. A page built on a fast, statically rendered site, the kind Astro produces, is also the same foundation that helps a site get found organically and cited by name in an AI generated answer, since Google states plainly that Core Web Vitals performance aligns with what its core ranking systems reward⁸. Fixing a quoting tool and building a fast, AI-citable site are not two separate projects. A sourced, accurate, quotable number on a fast page is exactly the kind of passage an answer engine can lift and attribute correctly, which a stale, unsourced estimate never will be.

Questions agencies ask