Every IRS audit rate ever published is a partial count, and the agency prints the warning in a footnote
The IRS says 0.1 percent of 2023 individual returns have been examined. That figure is not an audit rate. It is a reading of how far the examination cycle had run on 30 September 2025, and comparing it with a settled year is the single most common error made about tax enforcement.
0.1 percent. That is the share of individual income tax returns filed for tax year 2023 that the Internal Revenue Service had examined as of 30 September 2025, and it is close to the least informative true number available about American tax enforcement.
It is not the audit rate for 2023. It is a measurement of how far the examination cycle had run on one particular date, taken while most of the examinations that will eventually be opened on 2023 returns had not been opened yet. The IRS says so itself, in a footnote nobody quotes. Next to the tax year 2022 and 2023 columns of Table 3-1 of the 2025 Data Book sits a dagger, and the dagger note reads: "The shaded tax years show returns still within the normal 3-year statute of limitations. Therefore, the percentage covered and recommended additional tax will increase in future years as additional examinations are completed."
That warning is fifteen years old and almost universally ignored. So we measured what it is worth.
Money & World took three consecutive editions of the IRS Data Book and set the same tax years against each other, edition by edition, to establish how much a published coverage figure moves after it is first printed. The answer is that it moves by about 45 percent in the first year and barely at all after the third. Which means the honest comparison is not between a recent year and an old one. It is between two years measured at the same age.
Made that way, the comparison says something the rounded percentages hide.
What is the IRS audit rate?
The most recent settled figure is 0.32 percent. That is the share of individual income tax returns filed for tax year 2021 that the IRS has examined, or roughly one return in 315. Tax year 2021 is the most recent year the agency itself treats as complete, because 2022 and 2023 remain inside the three year assessment window.
Coverage is not uniform. It falls through the middle of the income distribution and rises sharply at the top, and the shape has changed as much as the level. The table below is computed from the return counts in Table 3-1 rather than copied from the rounded percentage column, and it reproduces the IRS figures wherever the agency prints one.
| Total positive income | TY2010 | TY2015 | TY2019 | TY2021 |
|---|---|---|---|---|
| All individual returns | 1.01% | 0.58% | 0.32% | 0.32% |
| No total positive income | 20.58% | 7.00% | 5.63% | 1.85% |
| $1 under $25,000 | 1.03% | 0.66% | 0.44% | 0.51% |
| $25,000 under $50,000 | 0.61% | 0.40% | 0.21% | 0.22% |
| $50,000 under $75,000 | 0.69% | 0.54% | 0.19% | 0.15% |
| $75,000 under $100,000 | 0.65% | 0.50% | 0.19% | 0.16% |
| $100,000 under $200,000 | 0.85% | 0.48% | 0.21% | 0.16% |
| $200,000 under $500,000 | 2.28% | 0.56% | 0.26% | 0.22% |
| $500,000 under $1,000,000 | 3.62% | 1.22% | 0.77% | 0.60% |
| $1,000,000 under $5,000,000 | 8.19% | 2.61% | 1.76% | 0.94% |
| $5,000,000 under $10,000,000 | 13.54% | 5.17% | 3.40% | 3.86% |
| $10,000,000 or more | 21.48% | 9.50% | 11.55% | 6.59% |
| Returns claiming the earned income tax credit | 1.81% | 1.17% | 0.79% | 0.70% |
Two features of that table are worth sitting with.
The first is the collapse at the top. A return reporting total positive income of $10 million or more was examined at 21.5 percent for tax year 2010 and at 6.6 percent for tax year 2021. The rate for the band below it, $1 million to $5 million, fell from 8.2 percent to 0.9 percent. Nothing in the middle of the distribution fell by anything like that proportion, for the simple reason that there was not as far to fall.
The second is that the bottom of the distribution is not the bottom of the enforcement. For tax year 2021 a return reporting between $1 and $25,000 of total positive income was examined at 0.51 percent, and a return reporting between $100,000 and $200,000 at 0.16 percent. Returns claiming the earned income tax credit, 15.2 percent of all individual returns filed for tax year 2023, were examined at 0.70 percent. That is more than four times the rate for the $100,000 to $200,000 band and rather more than the overall average.
Why does the published audit rate keep going up?
Because examinations are opened over several years and the table is printed once a year. A tax year first appears in the Data Book about eighteen months after its returns were due, when only a fraction of the eventual examinations exist. Each subsequent edition adds the examinations opened since.
The size of that effect can be measured, and it has not been published anywhere we can find. Take the same tax year out of three consecutive editions and read the individual return counts.
| Tax year | First published | One edition later | Two editions later |
|---|---|---|---|
| 2021 | 341,276 (0.21%) | 492,012 (0.31%) | 510,375 (0.32%) |
| 2022 | 326,411 (0.20%) | 472,613 (0.29%) | not yet published |
| 2023 | 159,266 (0.10%) | not yet published | not yet published |
Tax year 2021 gained 44.2 percent more examined returns in its second edition and 3.7 percent in its third, a total increase of 49.5 percent over its first published figure. Tax year 2022 gained 44.8 percent in its second edition. The two independent observations agree closely, and they agree with the mechanism: the assessment window closes three years after filing, so almost all of the additions arrive at once and then stop.
The money moves further than the count. Recommended additional tax for tax year 2022 individual returns went from $1.81bn at first publication to $5.85bn a year later, an increase of 223.5 percent. Field examinations, which produce most of the dollars, take far longer to close than correspondence examinations, so the earliest reading of any year is weighted towards the cheap cases.
None of this is hidden. Footnote 4 to Table 3-1 states that the percentage "may increase in future years as additional returns are selected for examination", and the notes at the foot of the table describe the whole table as "a snapshot in time of the examination process". The information is there. It is simply not carried by anybody quoting the number.
Is the IRS opening fewer examinations than it used to?
Yes, and by about half. Compared at the same age, the first published count of examined individual returns fell from 341,276 for tax year 2021 and 326,411 for tax year 2022 to 159,266 for tax year 2023. That is 51.2 percent below the 2022 figure and 53.3 percent below 2021.
This is the comparison the rounded percentages make impossible, and it is the one that answers the question people are actually asking. Each of the three counts was taken at the same point in the cycle, in the edition where its tax year first appeared, so the difference between them is not an artefact of measurement timing.
The shortfall is not spread evenly. Broken out by income band, at equal maturity:
| Total positive income | TY2021 first published | TY2022 first published | TY2023 first published |
|---|---|---|---|
| $1 under $25,000 | 0.40% | 0.38% | 0.16% |
| $200,000 under $500,000 | 0.09% | 0.12% | 0.04% |
| $1,000,000 under $5,000,000 | 0.50% | 1.07% | 0.22% |
| $5,000,000 under $10,000,000 | 1.41% | 3.11% | 2.01% |
| $10,000,000 or more | 2.91% | 3.95% | 5.16% |
| Returns claiming the earned income tax credit | 0.68% | 0.74% | 0.32% |
One band rose in all three years. Examinations opened on returns reporting $10 million or more of total positive income went from 2.91 percent of that band for tax year 2021 to 3.95 percent for 2022 and 5.16 percent for 2023, an increase of 77 percent across two tax years. Every other line in the table is lower for 2023 than for 2022, and the two largest populations fell by more than half: the $1 to $25,000 band by 57.2 percent and earned income tax credit claimants by 56.6 percent.
So the aggregate halving is a composition change rather than a general retreat. The examinations that disappeared are the cheap high volume ones. The examinations at the top of the income distribution were opened faster than in either of the two preceding years.
Whether the 2023 figure ends up where 2021 did depends on additions that have not happened yet. Apply the multiple tax year 2021 eventually reached, 1.4955, to the 2023 count and it implies about 238,000 examined returns and a final coverage rate near 0.15 percent, against 0.32 percent for 2021. That is arithmetic on two observations, not a forecast, and the IRS makes no such projection. The composition shift is a reason to treat it cautiously: a year running ahead at the top and far behind in the correspondence bands will not necessarily follow the aggregate path of a year that was not.
What does an IRS audit actually consist of?
A letter, in four cases out of five. The IRS closed 497,621 examinations in fiscal year 2025. Of those, 94,562 were field examinations and 403,059 were correspondence examinations, so 81.0 percent of all audits and 89.2 percent of individual audits were conducted through the mail.
The two are different events with different consequences. Table 3-2 gives recommended additional tax of $19.10bn from field examinations and $7.73bn from correspondence, which works out at about $201,980 per field examination against $19,176 per correspondence examination, a ratio of about ten and a half to one.
The split tracks income closely. Of the 4,060 individual examinations closed in fiscal year 2025 on returns reporting $10 million or more, 3,680 were field examinations. Of the 131,153 closed on returns reporting $1 to $25,000, 3,537 were. A taxpayer in the top band who is examined meets an examiner. A taxpayer in the bottom band who is examined receives a notice.
How many taxpayers does the IRS contact without auditing them?
About six times as many as it audits. Alongside 497,621 closed examinations, fiscal year 2025 produced 987,460 closed cases under the Automated Underreporter Program, 592,773 under the Automated Substitute for Return Program, and 1,440,328 math error notices. That is 3,020,561 contacts against 497,621 audits, a ratio of 6.1 to one.
None of those three programs is an examination and the Data Book counts them separately, which is correct: the underreporter program matches third party information returns against filed returns, the substitute for return program constructs returns for people who did not file, and a math error notice adjusts an arithmetic or clerical error without the procedural protections an examination carries. But they are all the IRS writing to a taxpayer to say that the return is wrong, and they account for six of every seven such letters.
Table 3-8 puts additional assessments from the two automated programs at $5.90bn and $2.93bn respectively, from a combined 1,534 full time equivalent positions. The audit rate does not describe any of this. Anyone using it as a measure of the odds of hearing from the IRS is understating them by roughly a factor of seven.
How many revenue agents does the IRS have?
Fewer than at any point since before 2019, and more than at any point since before 2010, depending on which column of the same table is read.
Table 6-3 of the 2025 Data Book reports revenue agents two ways. Full time equivalent positions realised rose from 8,983 in fiscal 2024 to 11,025 in fiscal 2025. The number of employees in pay status as of 30 September fell from 10,736 to 8,609 over the same period, a decline of 19.8 percent. Total IRS employment in pay status fell from 99,628 to 80,967, down 18.7 percent, which the table attributes in a footnote to employees electing the deferred resignation programmes, voluntary early retirement authority and voluntary separation incentive payments that took effect on or before 30 September 2025.
Both numbers are accurate. Full time equivalents average staffing across the year and the separations landed at the end of it, so the fiscal 2025 book records a year of high average staffing that closed with a substantially smaller workforce than it opened. Which of the two describes examination capacity for tax year 2024 and 2025 returns is the question, and the answer is the second one.
Set against the long series, the position is starker. Revenue agent full time equivalents ran at 13,879 in fiscal 2010 and 8,526 in fiscal 2019, with end of year headcounts of 14,588 and 8,681. The 8,609 in pay status at the end of fiscal 2025 is 41.0 percent below the fiscal 2010 figure. Returns filed went the other way, from 230.4 million in fiscal 2010 to 271,442,355 in fiscal 2025, an increase of 17.8 percent.
What the numbers do not settle
Three limits matter enough to state plainly rather than bury.
The bands are total positive income, not adjusted gross income, and Table 3-1 defines it as the sum of all positive income items with losses excluded. A taxpayer with large offsetting losses can appear in a higher band than their tax return otherwise suggests, which matters most at the top and in the "no total positive income" line, where the population itself swings violently from year to year: 964,115 returns for tax year 2023 against 3,631,912 for 2020.
Tax years 2020 and 2021 were filed and processed through disrupted seasons, and their examination cycles began under the same conditions. Comparisons that lean on those two years carry that with them.
And the coverage rate counts returns, not people. A single examination can reach several related returns across years and forms, and Table 3-1 counts those closures against the tax year and form type to which they belong. It is a measure of returns touched, which is what it says it is, and not a probability that any given filer will hear from the government.
Method
Coverage percentages in this piece are computed as closed examinations plus examinations in process, divided by returns filed for the tax year, using columns 1 to 3 of Table 3-1. That is the definition footnote 4 gives for the published column 4, and the computed figures reproduce the IRS rounded percentages throughout.
The multi-edition comparison uses the same table from three consecutive Data Books: the fiscal 2023 edition, where tax year 2021 first appears; the fiscal 2024 edition, where tax year 2022 first appears; and the fiscal 2025 edition, where tax year 2023 first appears. Tax year 2022 does not appear in the fiscal 2023 edition, which is what establishes each year's first publication. Each edition measures as of 30 September of its fiscal year.
Personnel figures are from Table 6-3 of the 2025 Data Book and the corresponding tables in the 2019 and 2010 editions, where the series is labelled Table 32 and Table 30 respectively. Full time equivalents and end of year headcount are reported separately in all three and are not interchangeable.
Frequently asked questions
What are the chances of being audited by the IRS? For the most recent settled year, tax year 2021, the IRS examined 0.32 percent of individual income tax returns, or about one in 315. The rate published for 2022 and 2023 is lower, but those years are unfinished and will rise. Rates by income band are in the table above.
Why do different sources give different IRS audit rates? Usually because they are quoting different tax years measured at different stages of completion, and sometimes because they are quoting a fiscal year count of closed examinations rather than a tax year coverage rate. A figure without a tax year attached to it cannot be checked.
Does claiming the earned income tax credit increase the chance of an audit? Returns claiming the credit were examined at 0.70 percent for tax year 2021, against 0.32 percent for individual returns as a whole. The IRS reports the figure separately in Table 3-1 and notes that the counts are also included in the income bands above it.
Is an IRS letter the same as an audit? No. Correspondence examinations are audits and made up 81.0 percent of those closed in fiscal year 2025. Automated underreporter notices, substitute for return cases and math error notices are counted separately by the IRS, and together they outnumbered examinations by about six to one.
