How to Calculate the 2027 Social Security COLA Yourself, and Why Your Check Rises by Less

Will the 2027 Social Security COLA be 3.6%? Not on the numbers published so far. One of the three months that decide it is already in, and July on its own puts the raise at 3.1%.

The 3.6% figure circulating since spring comes from The Senior Citizens League, and it carries a condition that rarely travels with it: August and September have to run hot enough to pull the whole quarter up. June ran negative.

The formula is public, both inputs are public, and last year's answer is public. That combination is unusual, and it is what makes this a calculation you can hand to a chatbot and still verify. A model that computes the COLA and a model that recalls one produce paragraphs that look identical, so everything below is built around a single question: can it reproduce a number you already have? Every figure here was read from SSA, BLS and CMS pages on August 25, 2026.

If you only want the dollar figure, skip to the section on how much your check actually goes up. Everything before it is how to check that figure yourself.

What is the COLA, and why does it run on CPI-W instead of the CPI in the news?

The COLA comes out of a statutory formula on one price index, and that index is not the one reported every month.

The cost-of-living adjustment raises Social Security benefits once a year under section 215(i) of the Social Security Act. The Social Security Administration averages the CPI-W index for July, August and September, divides that average by the same quarter's average from the most recent year in which a COLA took effect, and rounds the result to the nearest 0.1 percentage point. The 2026 COLA came out of 317.265 against 308.729, which is 2.8%. For the 2027 COLA the base stays 317.265, because 2025 is the year that produced the last adjustment, and the comparison quarter is July, August and September 2026.

CPI-W covers urban wage earners and clerical workers. CPI-U, the index behind every inflation headline, covers all urban consumers, and the two sit at different levels: July 2026 read 327.104 on CPI-W and 333.918 on CPI-U. Same month, same country, 6.8 index points apart, because the sample households are different. Only CPI-W enters this formula. If you want the monthly release itself decoded, the CPI report walkthrough covers which tables matter, and the personal inflation rate post deals with the gap between any index and your own spending.

How do you know the AI ran the formula instead of remembering one?

You hand it a year whose answer is already published, and you refuse to move on until it lands on that answer.

SSA prints both 2026 inputs on the same page as the result. Feed 317.265 and 308.729 into the formula and the answer is 2.8%. A model that returns something else on a settled year is not going to be right about a year that has not finished happening.

Two wrong answers show up at this gate, and each one names its own bug. A 2.7% means the model truncated where the statute rounds. A 2.764884% means it divided correctly and then skipped the rounding step entirely. Neither is a cosmetic difference on a benefit: 0.1 percentage point on a $2,000 primary insurance amount is $2 a month, and COLAs multiply into the base that every later COLA works on.

Proving the model against a published exemplar is the same move that works on the federal aid formula, where the Education Department publishes a fully worked answer to check against. Anchor first, your own numbers second.

The prompt, and the five lines you fill in before pasting

One message does three jobs: it proves the model on 2026, runs three scenarios for 2027, and applies each result to a single benefit.

It also forbids the model from recalling a COLA figure, which matters more than it sounds. Ask a chatbot for the 2027 COLA with no constraints and it will hand back a number from its training data or from a news article it half remembers, formatted exactly as if it had been computed.

Fill in the five bracketed lines in STEP 3. The version below carries the figures used in the run further down: a $2,000.00 primary insurance amount, claimed at 62 against a full retirement age of 67, with the Part B premium at $202.90 now and $209.50 assumed for next year.

You are checking my Social Security COLA math. Use only the numbers I give you.
Do not look up or recall a COLA number from memory.

STEP 1 - VERIFY YOURSELF FIRST.
Compute: (317.265 - 308.729) / 308.729 * 100, then round to the nearest 0.1
percentage point. State the unrounded value and the rounded value.
If your rounded value is not 2.8, stop and tell me you got it wrong.

STEP 2 - CURRENT YEAR.
Base (2025 Q3 CPI-W average): 317.265
Known 2026 CPI-W: July = 327.104. August and September are not published yet.
Give me three scenarios and show every intermediate number:
  A) August and September both equal July
  B) August = 317.306 * 1.034 and September = 318.139 * 1.034
  C) the Q3 average required to produce exactly 3.6%, and the implied
     month-over-month rate from July needed to get there
For each: Q3 total, Q3 average (3 decimals), the percentage before rounding,
and the COLA after rounding to 0.1.

STEP 3 - MY CHECK.
My PIA: $2,000.00
Age I claimed at: 62     My full retirement age: 67
Medicare Part B premium deducted from my check now: $202.90
Part B premium to assume for next year: $209.50
Apply the official order: multiply PIA by (1 + COLA) and truncate to the next
lower dime; apply my early/delayed factor and truncate to the next lower dime;
subtract the Part B premium; truncate to the next lower dollar.
Do this for each of the three COLA scenarios.

OUTPUT:
1. A table: scenario | COLA | new PIA | gross monthly benefit | Part B | net check
2. One line per scenario: gross increase vs net increase, in dollars
3. A list of every number you used that came from me versus every number you
   assumed, labeled separately
4. Anything in my inputs that would change the answer and that you cannot verify

What the run returned, and the two places it slipped

The run cleared the 2.8% gate on its first try and printed every division, then got two of its own summary lines wrong.

Every index value in the body of that run matches SSA's published series, and every division is printed rather than asserted. The failures sit in the summary, which is the part a reader is most likely to quote.

The larger slip is the closing sentence. Scenario C's gross raise is $50.40, which is the new gross benefit of $1,450.40 ($2,072.00 × 0.70) minus today's $1,400.00, and $6.60 against that is 13.1%. The run copied the same share onto scenario A. Against A's gross raise of $43.40, $6.60 is 15.2%. Scenario A is the base case of this whole post, so the number most people would carry away is the one the run got wrong, and it got it wrong while every input feeding it sat correct in the table directly above.

The second slip is reasoning rather than arithmetic. Scenario B came out at 3.399787%, and the run called it fragile because it sits 0.0002 points from 3.4%. Rounding to the nearest 0.1 puts the boundary at 3.35, so B is a full 0.05 points clear of the line and rounds to 3.4% without difficulty. What is actually fragile about B is its premise, that the year-over-year rate holds at exactly 3.4% for two more months. Put plainly, B assumes the year-over-year rate holds and then reports 3.4%, so it is a premise dressed as a finding, not a third independent scenario.

I kept both rather than rerunning, because they show the shape of the risk. The model was never going to invent 327.104. It invented a summary.

The run was also right about something nobody asked it. Its list of unverifiable inputs flagged that "claimed at 62" is ambiguous: at 62 and one month the reduction is 29.583% rather than 30%, the factor becomes 0.704167, and the scenario A net check comes to $1,242 instead of $1,233. One month of claiming age moves the answer by $9, more than the whole distance between the 3.1% and 3.6% scenarios. The break-even post covers why SSA's own examples use 62 and one month.

What is the 2027 COLA so far, with only July in hand?

Three answers, depending on what August and September do, and only the first of them assumes nothing at all.

Scenario A is not a prediction that prices stop rising. The base it measures against is last year's third-quarter average, 317.265, and July 2026 already sits 3.1% above that. Most of that increase happened between last fall and this July, and it is baked into the July index level. Hold that level through August and September and the quarter averages the same 327.104, so the gap stays 3.1%. Zero further inflation still produces a 3.1% COLA. That is the floor unless prices fall outright before the end of September.

Scenario C is where the headline lives, and its requirement is the part the headline omits. To average 328.687 for the quarter, the index has to climb from 327.104 in July to 328.684 in August and 330.272 in September, a steady 0.48% a month. June fell 0.53% and July rose 0.01%. The last time CPI-W moved that fast was spring.

Read the shaded pair as a requirement, not a projection. Whether rent, gasoline and food deliver it before the end of September is a question I have no edge on, and neither does the model.

If the COLA is 3.1%, how much does your check actually go up?

The percentage applies to your full benefit, not to the check that arrives, and three separate steps sit between the two.

SSA fixes the order. Multiply the primary insurance amount by the COLA and truncate to the dime, apply the early or delayed claiming factor and truncate again, subtract the Medicare Part B premium, then truncate to the whole dollar. Claiming at 62 with a full retirement age of 67 fixes that factor at 0.700000, a 30% reduction that never expires and that carries forward through every future COLA.

Two things take the raise down from $43.40 to $36. $43.40 minus the $6.60 premium increase leaves $36.80, and then the final truncation to the whole dollar takes another 80 cents: $1,233.90 becomes $1,233 while the old $1,197.10 became $1,197.

That $6.60 is small by recent standards, and the familiar line about Medicare swallowing the COLA is a 2026 sentence being carried into 2027. In 2026 the standard Part B premium jumped $17.90, from $185.00 to $202.90, against an average retired-worker raise of $56.42 as the average benefit moved from $2,015 to $2,071. That $56.42 is SSA's own published figure, and Medicare took 32% of it. The 2026 Trustees Report projects $209.50 for 2027, a $6.60 increase, which is 15.2% of the $43.40 raise in the example above. For the average recipient the share is smaller: 3.1% of $2,071 is a $64.20 raise, our arithmetic rather than SSA's, and $6.60 is 10.3% of that.

Two cautions sit under that. The $209.50 is the intermediate projection in the Trustees Report rather than an announced figure, and CMS publishes the confirmed premium in the fall, usually November. And hold harmless does almost nothing at these numbers. The provision caps the dollar increase in your Part B premium at the dollar increase in your benefit, so with a $6.60 premium increase and a 3.1% COLA it binds only on benefits below about $213 a month. It also does not reach people paying IRMAA surcharges, the income-related monthly adjustment amount that raises Part B for higher incomes, nor people who enrolled mid-year, nor people who do not have the premium deducted from a check at all.

Three ways a chatbot gets the COLA wrong

Each of the three produces a plausible-looking percentage, and the 2026 gate in STEP 1 catches every one of them before you reach your own numbers.

The costliest is the wrong index. Ask about "the CPI" and a model may reach for CPI-U, which read 333.918 in July 2026. Measured against the 317.265 base that gives 5.25%, a figure so far outside the historical range that it would be obvious to anyone with a feel for what a normal COLA looks like, which is almost nobody.

Next is the year-over-year swap. CPI-W in July 2026 sat 3.4% above July 2025, and that number is printed in the BLS release. It is not the COLA. The COLA compares quarterly averages against a base year, and the two figures agree only by coincidence.

The mildest is truncation, which turns 2.764884% into 2.7% instead of 2.8%. On its own that costs $2 a month on a $2,000 benefit. As a signal it costs more, because a model truncating here is following a rule it made up, and the next rule it makes up may not be visible.

The index confusion is the one worth naming out loud in your prompt, since "the CPI" is genuinely ambiguous and the model resolves the ambiguity silently. It belongs to the same family as reading a nominal figure as a real one: the arithmetic is clean and the input was the wrong series.

What to watch between now and October

Two data releases and one announcement close this out, and the earlier release is the one that moves the answer most.

Two consequences follow the announcement that the COLA headline never mentions. The thresholds deciding whether your benefits get taxed, $25,000 of combined income filing single and $32,000 filing jointly, are not indexed to inflation, so every COLA moves more households above them. SSA publishes no count of how many cross in a given year, so any number you see for that is somebody's estimate. And IRMAA runs on a two-year lookback, which means income you receive in 2027 sets your 2029 premium rather than your 2028 one, so anyone telling you this raise will lift next year's Medicare surcharge is off by a year.

If I were living on a benefit check today I would budget on 3.1% and treat anything above it as a surprise, because scenario A is the only one of the three that assumes nothing about a month that has not happened.

FAQ

How do I know ChatGPT ran the COLA formula right instead of guessing?

Make it reproduce a year that is already settled. SSA publishes the 2026 inputs and the answer on the same page: (317.265 minus 308.729) divided by 308.729, times 100, rounded to the nearest 0.1 point, is 2.8%. Put that step first in your prompt and tell the model to stop if it does not land on 2.8. Two wrong answers each name a specific bug. A 2.7% means it truncated where the statute rounds. A 2.764884% means it did the division and skipped the rounding. Both signal a model applying its own rules, and a model doing that on 2026 will do it on 2027, where you have no published answer to check against. Ask for every intermediate value as well, because a printed division is checkable and an asserted percentage is not.

What is the 2027 COLA so far, with only July in hand?

It is 3.1%. July 2026 CPI-W came in at 327.104, published August 12. If August and September hold at that level, the third-quarter average is 327.104, which sits 3.1012% above the 317.265 base and rounds to 3.1%. If instead the year-over-year pace of 3.4% holds through September, the quarter averages 328.051 and the COLA is 3.4%. The 3.6% estimate from The Senior Citizens League needs a third-quarter average of 328.687, which means the index has to rise about 0.48% in August and again in September. June fell 0.53% and July rose 0.01%, so 3.6% requires a return to the pace of March through May. The two unknown months are published on September 11 and October 14.

If the COLA is 3.1%, how much does my check actually go up?

On a $2,000 primary insurance amount claimed at 62 with a full retirement age of 67, the gross benefit goes from $1,400.00 to $1,443.40, a raise of $43.40, and the net check goes from $1,197 to $1,233, a raise of $36. The gap is the Medicare Part B premium, projected at $209.50 for 2027 against $202.90 today. SSA applies the steps in a fixed order: multiply the primary insurance amount by the COLA and truncate to the dime, apply the claiming factor and truncate again, subtract Part B, then truncate to the dollar. Your own figures will differ on all three inputs, and the claiming factor moves the answer more than the scenario choice does. Claiming at 62 and one month rather than 62 and zero months changes the factor from 0.700000 to 0.704167 and the net check to $1,242.

Does Medicare eat the Social Security COLA?

Less in 2027 than in 2026, on current projections. In 2026 the standard Part B premium rose $17.90, from $185.00 to $202.90, while the average retired-worker benefit rose $56.42, from $2,015 to $2,071. The premium took 32% of that raise. The 2026 Medicare Trustees Report projects a $6.60 increase for 2027, to $209.50. Against a 3.1% COLA that is 10.3% of the average recipient raise of $64.20, and 15.2% of the $43.40 raise in the worked example on this page. The projection is not the confirmed number, and CMS publishes that in the fall. Hold harmless, the provision capping your premium increase at your benefit increase, binds only on benefits below about $213 a month at these figures, and it excludes people paying IRMAA surcharges, people who enrolled mid-year, and people who do not have the premium deducted from a check.

Sources

• Social Security Administration, Latest Cost-of-Living Adjustment, read August 25, 2026 (the section 215(i) formula and the 0.1 point rounding rule; the 2026 COLA of 2.8%; the 2025 third-quarter average of 317.265 from 316.349, 317.306 and 318.139; the 2024 third-quarter base of 308.729): https://www.ssa.gov/oact/cola/latestCOLA.html
• Social Security Administration, Consumer Price Index for Urban Wage Earners and Clerical Workers (CPI-W), read August 25, 2026 (the monthly CPI-W series; 2026 values from January 317.942 through July 327.104; footnote a on the missing October 2025 reading): https://www.ssa.gov/oact/STATS/cpiw.html
• Social Security Administration, Application of COLA to a Retirement Benefit, read August 25, 2026 (the order of operations and the truncation rule at each step): https://www.ssa.gov/OACT/COLA/colaapplic.html
• Social Security Administration, Early or Late Retirement, read August 25, 2026 (the 30% reduction at 62 for anyone born in 1960 or later; 5/9 of 1% per month for the first 36 months and 5/12 of 1% thereafter): https://www.ssa.gov/OACT/quickcalc/earlyretire.html
• Social Security Administration, 2026 Social Security Changes fact sheet, read August 25, 2026 (the average retired-worker benefit rising from $2,015 to $2,071 in January 2026): https://www.ssa.gov/news/en/cola/factsheets/2026.html
• Bureau of Labor Statistics, Consumer Price Index, July 2026, released August 12, 2026 (CPI-W up 3.4% over the year, down from 3.5% in June; CPI-U index level 333.918): https://www.bls.gov/news.release/cpi.nr0.htm
• Bureau of Labor Statistics, CPI news release schedule, read August 25, 2026 (August CPI on September 11, 2026 at 8:30 a.m. ET; September CPI on October 14, 2026): https://www.bls.gov/schedule/news_release/cpi.htm
• Centers for Medicare & Medicaid Services, 2026 Medicare Parts A & B Premiums and Deductibles, November 14, 2025 (the standard Part B premium of $202.90 against $185.00 in 2025; the $283 annual deductible; the 2026 IRMAA thresholds of $109,000 single and $218,000 joint): https://www.cms.gov/newsroom/fact-sheets/2026-medicare-parts-b-premiums-deductibles
• 2026 Annual Report of the Boards of Trustees of the Federal Hospital Insurance and Federal Supplementary Medical Insurance Trust Funds, Table V.E2 (the projected 2027 standard Part B premium of $209.50 and deductible of $292 on the intermediate assumptions; the hold-harmless provision and the groups it excludes): https://www.cms.gov/oact/tr/2026
• Social Security Administration, Income Taxes and Your Social Security Benefits, read August 25, 2026 (the combined-income thresholds of $25,000 and $32,000, which are not indexed to inflation): https://www.ssa.gov/benefits/retirement/planner/taxes.html
• The Senior Citizens League, COLA Watch, read August 25, 2026 (the 3.6% estimate for 2027, revised down from 3.8% after the July release): https://seniorsleague.org/cola-watch/
• Cold run transcript, Claude (Opus 4.8), August 25, 2026, single turn, no follow-up corrections and no rerun after the two summary errors surfaced.

Disclaimer

This is an educational explainer about a published federal formula, not financial, tax or benefits advice, and none of it is specific to your situation. The $2,000.00 primary insurance amount, the age-62 claiming date and the full retirement age of 67 are illustrative inputs you replace with your own from your SSA statement. The 3.1%, 3.4% and 3.6% figures are scenarios built on unpublished months rather than predictions: August 2026 CPI-W is released on September 11 and September 2026 CPI-W on October 14, and the COLA is not final until SSA announces it. The 2027 Part B premium of $209.50 is a Trustees Report projection on intermediate assumptions, and CMS has confirmed no 2027 figure as of August 25, 2026. State taxation of benefits, IRMAA surcharges, Part D premiums, the windfall elimination and government pension offset provisions, and spousal and survivor benefit rules all sit outside this calculation and can change the result. Check your own figures against your SSA account before acting on any of it.