Form statistics from 11.7 million submissions, 2022 to 2026: spam per form by year, a monthly 2026 breakdown, who fills in forms, and what forms ask for.

Most of what reaches a web form in 2026 is not written by a person. formcarry received 11.7 million form submissions between January 2022 and October 2026 and flagged 8.5 million of them as spam. In October 2026, 92 of every 100 submissions were spam. These form statistics cover five years, with a month-by-month breakdown of 2026.
The numbers come from our own submission records, which is why they look different from the usual lists of form statistics. Those lists measure what happens before the submit button: how many visitors start a form and how many finish it. We measure what arrives after it: who sent it, from where, on which device, with which fields, and whether it was a person at all.
Data source: formcarry's own submission records. formcarry is a form API that has run since 2017: a form on any website posts to a formcarry endpoint, and formcarry stores the submission, filters spam and delivers it.
Sample: every submission received between January 1, 2022 and October 31, 2026: 11.7 million submissions to more than 20,000 forms. 2026 covers January to October, so year-over-year comparisons use the same ten months of each year where it matters.
Method: spam is what formcarry flagged when the submission arrived. "Typical" means the median form, so each form counts once and a few very large forms cannot move the figure. A "targeted form" is a form that received at least one spam submission in the month, and spam per targeted form divides the month's spam by those forms, so the figure does not rise just because more forms exist. Figures marked "pooled" divide totals, and large forms weigh more in them. Fields are counted from the field names in each submission, so "carries an email field" means the field was sent, not that the address is real. Country is the location of the sender's IP address.
Limits: the sender's country is recorded since 2023 (completely since July 2023), and the device since August 2026. Detection changed over five years, so part of any rise in flagged spam is better detection. Form type is worked out from form and field names.
Raw spam totals rise whenever more forms exist, so they say little on their own. The useful question is how much spam one form receives. We count it per targeted form: a form that received at least one spam submission that month.
| Year, January to October | Spam per targeted form, per month | Share of submissions that was spam | Flooded forms per 1,000 targeted forms |
|---|---|---|---|
| 2022 | 86 | 73.4% | 5.7 |
| 2023 | 70 | 62.2% | 4.1 |
| 2024 | 82 | 72.4% | 4.1 |
| 2025 | 117 | 81.8% | 10.1 |
| 2026 | 98 | 85.8% | 17.5 |
Each year covers the same ten months, because 2026 ends in October. A flooded form received 1,000 or more spam submissions in one month.

Three things in this table matter.
First, the long-run rise is real but modest. A targeted form received 86 spam submissions a month in 2022 and 98 in 2026, 14% more.
Second, the averages hide the direction. 2025 was the heaviest year so far and 2026 began well below it. Then September and October 2026 became the two heaviest months of the whole five years, at 157 and 159 per targeted form. The earlier peak was 148, in September 2022.
Third, the share is pooled, and a pooled share describes the forms under attack more than it describes your form. Spam is concentrated: a typical web form gets 16.1% spam, and a minority of forms get most of it. Source: formcarry's analysis of 2,580 active forms, January 2024 to October 2026. Among forms with steady traffic, 27% received more spam than real submissions in 2026, up from 18% in 2022.
What to do with this: do not size your spam problem from an average. Look at your own form's flagged share for a month. If it is a few percent, a quiet form is normal. If spam outnumbers real submissions, you are in the minority that bots have found, and the rest of this report is about you.
Our numbers sit inside a wider pattern. Imperva's 2026 Bad Bot Report measured automated traffic at more than 53% of all web traffic in 2025, with human activity down to 47%.
The year has two halves. From January to June, a targeted form received between 63 and 84 spam submissions a month. From July the figure climbed every month, and by October it was two and a half times February's.
| Month, 2026 | Spam per targeted form | Share of submissions that was spam | Flooded forms per 1,000 targeted forms |
|---|---|---|---|
| January | 78 | 82.2% | 4.0 |
| February | 63 | 80.8% | 2.2 |
| March | 65 | 81.9% | 2.5 |
| April | 69 | 82.1% | 4.1 |
| May | 84 | 84.0% | 5.5 |
| June | 84 | 83.0% | 5.2 |
| July | 104 | 86.2% | 10.7 |
| August | 113 | 85.1% | 15.7 |
| September | 157 | 90.5% | 61.4 |
| October | 159 | 92.1% | 63.4 |
One number is missing from the table because it barely moves. Of the forms that received anything in a given month, between 60% and 66% received spam, every month of the year. What changed in 2026 is not how many forms bots reach. It is how hard they hit the ones they reach.
A flood is easy to define: one form, 1,000 or more spam submissions, one month. Through the first half of 2026 that happened to between 2 and 6 of every 1,000 targeted forms. In September and October it happened to more than 60 of every 1,000, about one targeted form in sixteen.

The number of targeted forms stayed between 1,800 and 2,050 all year, so this is not growth in the number of forms. A jump like that across unrelated forms in the same weeks is not many separate spammers. It is the same operation working through a list of form endpoints.
The source of the traffic says the same thing. In 2025, a form under attack received spam from 26 countries in a typical month. In 2026 that fell to 14 countries, because the traffic moved into fewer networks. Source: formcarry's analysis of its submission records, January 2023 to October 2026.
| Where spam was sent from (IP address location) | 2024 | 2025 | 2026, January to October |
|---|---|---|---|
| Russia | 35.1% | 38.2% | 36.6% |
| Netherlands | 6.2% | 7.7% | 34.0% |
| United States | 7.7% | 7.2% | 6.8% |
| Countries and territories seen | 189 | 195 | 161 |
The Netherlands went from under 8% of flagged spam to 34% in one year. Month by month it rose from 12% in January 2026 to more than 40% from July onward. A country does not start sending five times more spam on its own. The Netherlands hosts a great deal of rented server capacity, and my reading is that this is spam leaving home connections and moving into data centers, where one operator can run thousands of sessions. An IP address shows where a server sits, not who rents it.
What to do with this: blocking by country is the first thing people try, and this table is why it disappoints. The traffic moves, and a block on a hosting country also blocks real customers who use a VPN. Filtering has to judge the submission, not the map.
The old picture of form spam is a script that posts gibberish. That picture is out of date.
Between August and October 2026, formcarry recorded the user agent of 829,006 spam submissions. Only 1,444 of them named an automation tool. Every other one presented itself as an ordinary browser.
The content gives nothing away either. A crude bot posts whatever it likes to the endpoint. These post exactly the fields the real form has: in 2026, 98.4% of spam submissions included the form's email field and 89.2% included its message field. Source: formcarry's analysis of 1.9 million spam submissions, January to October 2026. A spam submission has the same fields as a real one, in the same shape, so a filter cannot tell them apart by structure.
Imperva describes the same shift across the web: bot requests "often appear legitimate, with well-formed requests" and differ only in intent and scale.
What to do with this: checks that look for an empty field, a missing header or an obvious script no longer catch much, because that is not what spam looks like. A hidden trap field still costs nothing and is worth keeping. But the work has moved to judging content, sender and behaviour together, and to updating those judgments as the traffic changes.
Not everything automated is advertising. In the summer of 2026 we found scanners walking through form fields with database attack strings, the kind used to test whether a site passes form input straight into a query.
Over 30 days we counted 4,894 of these payloads. The scanners were not visiting pages and filling in forms. They already had the endpoints and posted to them directly.
These are not spam in the usual sense. A content filter asks whether a message reads like a sales pitch, and an injection string does not. formcarry blocks them.
What to do with this: if your form posts to code you wrote, assume every field will receive hostile input, and never build a query or a shell command from it. If it posts to a form API, ask what happens to a payload like this before it reaches your inbox, your spreadsheet or your CRM.
Real submissions tell a quieter story. The senders are spread across the world, about half of them are on a phone, and a large share of them write outside working hours.
Country. Real submissions came from 214 countries and territories in the first ten months of 2026. For 35% of forms, the United States is the main country. The United Kingdom, Germany and India follow with 4% to 5% each, and 102 different countries are the main source of at least one form. Most forms are more international than their owners expect: 82% receive submissions from more than one country, and 45% from five or more.
Device. About half of a typical form's submissions come from a phone: 48.5%. The kind of form changes that a lot. A typical newsletter signup gets 78.6% of its submissions from phones. A typical contact form gets 43.1%. Source: formcarry's analysis of forms with device data, August to October 2026.
Time. In 2026, 42% of real submissions arrived outside 9am to 6pm in the sender's local time. Roughly a quarter came in the evening and the rest overnight and early in the morning. This figure is pooled, and local time is estimated from the sender's country.
Phones matter for more than layout. Zuko's form benchmarking data, drawn from 93 million tracked sessions, shows that 55.5% of desktop visitors who start a form complete it, against 47.5% on mobile.
What to do with this: test the form on a phone first, since that is where half your senders are. And decide what happens to a message that arrives at 9pm: an auto-response and a notification that reaches someone are the difference between a lead and a missed one.
Forms got longer. The typical form asked for 4 fields in 2022 and asks for 5 in 2026, and the extra field is usually a phone number.
| Year | Fields on a typical form | Forms that ask for a phone number | Forms that received a file |
|---|---|---|---|
| 2022 | 4.0 | 33.1% | 2.0% |
| 2023 | 3.3 | 30.0% | 1.9% |
| 2024 | 4.0 | 48.8% | 3.0% |
| 2025 | 4.6 | 49.0% | 3.8% |
| 2026 | 5.0 | 55.4% | 5.8% |
Each year counts the forms that received at least 10 real submissions that year, between 2,000 and 2,600 forms. Over the same five years, the share of forms that ask for an email address rose from 69% to 84%, and the share with a message box from 39% to 57%.
The same thing shows inside 2026 when you count per submission instead of per form. This view is pooled, so long forms that receive many submissions (orders, bookings, applications) weigh more, which is why its field count is higher than the typical form's.
| Month, 2026 | Fields per real submission | With a phone field | With a file attached |
|---|---|---|---|
| January | 8.6 | 65.4% | 2.3% |
| February | 9.0 | 63.9% | 2.9% |
| March | 9.1 | 63.1% | 3.1% |
| April | 9.5 | 69.3% | 3.9% |
| May | 10.5 | 72.9% | 4.0% |
| June | 10.5 | 70.8% | 3.5% |
| July | 11.0 | 73.2% | 3.2% |
| August | 10.4 | 80.2% | 3.4% |
| September | 11.1 | 70.3% | 4.0% |
| October | 11.8 | 73.0% | 4.7% |
The mix of form types explains the direction. Among forms with steady traffic in 2026, 56% are contact forms, 15% are order or booking forms, 9% are lead or quote forms and 4% are job applications. Forms are doing more of the work that used to need an app: taking orders, bookings and applications, with files attached.
What to do with this: every field you add is data you now hold. A form that collects phone numbers and uploaded documents deserves the same care as a customer database: know where each submission is stored, who can read it and where it is forwarded.
Put the five years together and the job of receiving a form submission has changed.
In 2022 a form handler could be forty lines of code and a mail function, and a hidden field stopped most of the junk. In 2026 the same endpoint is found by scanners, flooded in coordinated waves from rented servers, and written to by bots that send every field the way a customer's browser would. The handler now has to tell those apart, keep working through a flood, refuse attack payloads, and still deliver the one real message in ten.
That is the case for a managed form service, and it is a practical one, not a slogan:
You do not need any of this for a form that receives a few messages from people who already know you. You need it once bots have found the endpoint. In any month of 2026, about two in three forms that received anything received spam.
These are projections from the data above, not predictions.
Spam will stay the majority of what forms receive. The share has risen for three years running, and the last two months of the series are the heaviest per form. Plan for nine in ten.
Telling a bot from a person by the shape of a submission will stop working. It mostly has. Messages will read better, too. The filter that works in 2027 judges many signals at once and is retrained often.
Forms will be treated as an attack surface. The injection scans of 2026 were crude. Expect more of them, aimed at whatever sits behind the form: a database, a spreadsheet, an email template, an AI agent that reads the inbox.
More of the reading will be done by software. If nine in ten submissions are noise and the tenth arrives at 9pm, people will want help sorting what matters. This is where formcarry is heading. The MCP server already lets Claude, ChatGPT or Cursor read submissions and check where each one was delivered. AI reports are coming in early 2027, and so are Workflows, a visual integration builder.
It depends on whether bots have found the form. A typical web form gets about one spam submission in six. A form that bots target can receive more spam than real messages, and in 2026 that was true of roughly one in four forms with steady traffic. Check your own flagged share over a month before comparing yourself to an average.
It stops part of it. A captcha challenge raises the cost of each submission, which removes the cheapest bots. It does not stop operations that solve or bypass challenges, and it adds a step for real senders. Use it on forms that are under attack, alongside filtering, and leave it off forms that are not.
Usually not. In 2026 a third of flagged spam came from IP addresses in the Netherlands, up from under 8% a year earlier, so a list of blocked countries goes out of date within months. Country blocks also stop real senders who travel or use a VPN.
A contact form is a free way to put a message in front of a person, with a link in it, from an address that looks like the site's own. Some operations also use forms to test stolen data, to plant links, or to probe the site behind the form for weaknesses.
We plan to refresh this report once a year, with the same method, so the yearly figures stay comparable. The data here ends on October 31, 2026.
Three findings carry the report. The spam a targeted form receives grew modestly over five years and then doubled inside 2026, ending the period at its highest level. The attacks are coordinated, arriving in floods from rented servers and shaped like submissions from a real browser. And real forms got longer and more valuable in the same years, with phone numbers and files in them.
If bots have found your form, filtering is no longer something you set up once. Get started free and let formcarry handle it.
About the numbers: they come from formcarry's analysis of its own submission records, January 2022 to October 2026. "Spam" is what formcarry flagged when a submission arrived. "Typical" means the median form. The 16.1% spam figure and the device figures use 2,580 active forms, each with at least 20 real submissions between January 2024 and October 2026; device data exists for 592 of them. "Forms with steady traffic" means forms with at least 20 real submissions in the year, and the form type mix is counted on those forms for 2026. Country and time-of-day figures for real submissions use forms with at least 10 real submissions in 2026. Pooled figures divide totals. Country is the location of the sender's IP address and is recorded since 2023. Device is recorded since August 2026. The count of attack payloads covers 30 days in August and September 2026.
Nusu Alabuga founded formcarry in 2017 and still runs it: the ingest pipeline, the spam filtering, the email delivery and the incident tracking that watches customers' forms. A product engineer by trade, with years of work in payments, security, invoicing and banking infrastructure.
WebsiteXLinkedInGitHub