AiformleadsAiformleads

Almost half of your form submissions are bots: how to detect fake leads in 2026

Team at AiformleadsWritten with AI assistance17 min read

In summary

  • The Ad Fraud Report by Opticks Security (published on October 14, 2026, covering over 2 billion clicks from 500 advertisers across 243 territories between January 2025 and March 2026) places the proportion of lead generation conversions that are malicious bots at 46.7%, compared to 18.3% in e-commerce.
  • The difference is not the sector: it is the friction. A payment gateway requires a functioning card; a form only requires typing. Opticks measures 2.5 times more bots when there is no payment involved.
  • The spam of 2026 is no longer gibberish. They are agents based on language models that navigate at human pace and fill out with plausible names, syntactically valid emails, and well-formed phone numbers. DataDome recorded 7.9 billion requests from AI agents just in January and February 2026.
  • CAPTCHA is no longer the frontier: the academic article Broken Gates (Ousat et al., reviewed on November 5, 2026) documents a 100% success rate of resolution services against challenge-based defenses.
  • What does work is measuring submission behavior: completion time, consistency between responses, origin verified with hidden fields and UTM, scoring rules that penalize incoherence, and a step-by-step funnel where a human drops off and a bot does not.

If you manage lead generation, you've been noticing it for months: the volume goes up, the cost per lead goes down on the panel, and yet the sales team calls and there's no one on the other side. It's not an impression. It's the central data of 2026, and it changes the question that needs to be asked: it's no longer "how many leads have I brought in," but "how many of these leads exist." This article reviews the published figures, the costs that do not appear in any report, and the specific signals that betray an automatic submission — the same ones that Aiformleads, the lead generation tool by AIsToHire, converts into tags and scoring at the moment of submission.

Line of human figures in front of a giant web form where half of them are translucent silhouettes of circuits
Line of human figures in front of a giant web form where half of them are translucent silhouettes of circuits

How many of the leads I receive are bots? Almost half

46.7% of lead generation campaign conversions correspond to malicious bots, according to the Ad Fraud Report by Opticks Security published on October 14, 2026. The sample is not small: over 2 billion clicks from more than 500 advertisers and media agencies across 243 territories, analyzed between January 2025 and March 2026. During that period, Opticks recorded 416.3 million bot sessions, of which 370.9 million (89.10%) were malicious bots and only 45.4 million legitimate crawlers.

Put in context with other conversion goals, the contrast is the relevant finding:

Conversion GoalConversions that are malicious bots
Lead generation (form)46.7%
E-commerce (purchase)18.3%

Source: Ad Fraud Rates by Channel, Opticks Security, data from January 2025 to March 2026.

In other words: if you received 400 quote requests this month, the reasonable statistical scenario is that about 180 do not correspond to any person. And they are not randomly distributed across your channels: they concentrate where the form is shorter and the campaign is more aggressive.

Why does a form attract more bots than an online store?

Because the conversion goal determines the fraud profile. A purchase forces the bot to overcome a payment gateway with a valid card, which is an expensive and verifiable filter; a form only requires it to type text, which is free. Opticks quantifies it: there are 2.5 times more bots when there is no payment gateway in the path to conversion.

This asymmetry explains why lead generation has become the preferred target. It is the only mass funnel in digital marketing where the conversion costs nothing to the attacker and, on the other hand, does cost money to the advertiser: each false submission consumes bidding budget, takes up a spot in the CRM, and triggers a notification to a salesperson.

Two funnels compared: the e-commerce one hindered by a payment gateway and the one for forms with the door open for all bots to pass through
Two funnels compared: the e-commerce one hindered by a payment gateway and the one for forms with the door open for all bots to pass through

And it is not a problem that resolves itself. The report State of Bot & Agent Security 2026 by DataDome (published on September 22, 2026, covering over a trillion requests across more than 75,000 sites between July 2025 and June 2026) measured a 124% increase in malicious bot traffic, more than nine times the growth of human traffic in the same period.

What changed in 2026: form spam no longer looks like spam

Until recently, a fake submission could be recognized at a glance: names with random consonants, a link in the comments field, the same text repeated a hundred times. In 2026, the dominant profile is different: agents built on language models that load the page, scroll, pause for a few seconds before clicking, and write coherent responses to the question. Plausible names, emails with valid syntax, well-formed phone numbers, and free text that, read loosely, passes for human.

The scale is documented. DataDome recorded 7.9 billion requests from AI agents in January and February 2026 (4.11 billion in January and 3.45 billion in February), and 29.65 billion in the first half of 2026 compared to 23.06 billion in the second half of 2025. More relevant for those managing forms: in the first half of 2026, AI agents generated 605.6 million requests to sensitive endpoints — login pages, forms, carts, and payment flows.

Metric (DataDome, Jul. 2025 – Jun. 2026)Value
Requests from registered AI agents52.7 billion
Growth of AI traffic+82.3 %
Growth of malicious bot traffic+124 %
AI requests to sensitive endpoints (H1 2026)605.6 million
Sites that do not block any type of bot (out of 21,491 analyzed)65.3 %
Sites with full protection2.4 % (down from 8.4 % in 2024)

The qualitative leap is well summarized by a figure from the same report: monthly requests from AI bots to login pages rose from 11.9 million in January 2026 to 99.7 million in June. And the precedent for where this is heading was set by email: according to data from SlashNext collected by Enilon, malicious phishing emails grew by 1,265 % since the launch of ChatGPT in the fourth quarter of 2022. The form is lagging behind the inbox, not ahead.

Is putting a CAPTCHA useful? Less and less

As a standalone barrier, no. The research Broken Gates: Re-evaluating Web Bot Defenses in the Age of LLM Agents (Ousat, Turkmen, Rampersaud, Bailey, and Kharraz; presented on July 21, 2026, and revised on November 5, 2026) documents that commercial resolution services achieve a 100 % success rate against challenge-based defenses, while their average rate against reCAPTCHA v3 —which scores behavior instead of presenting a puzzle— is only 23 %.

The practical conclusion of that work is what matters here: what determines whether a defense accepts the agent is not the puzzle, but the browser environment and session history. An agent with a persistent Chrome profile and an established consumer account gets through where a newly created one does not.

This does not mean removing the CAPTCHA. It means stopping treating it as the filter and starting to treat it as another layer. Common implementation guides, like the one from ManagedNerds on fake form submissions (published on April 21, 2026, and updated on November 12, 2026), recommend combining it with honeypots, server-side validation, and service-specific questions, and they remind of an operational limit that is easy to forget: the free tier of reCAPTCHA covers 10,000 checks per month per organization, with a recommended score threshold of 0.5 in version 3.

The real cost of a fake lead: budget, time, and a scoring system that stops working

The visible cost —the bidding budget burned on clicks that were never going to buy— is the least of the three. The second is the time of the sales team, which calls, leaves messages, retries, and marks "no answer" on records that do not exist. The third is what truly hurts in the medium term: the lead scoring model gets contaminated and stops being useful for prioritization.

The mechanism is described by 4Thought Marketing (September 29, 2026, updated October 22): automatic submissions "accumulate the same points" as a legitimate prospect, so a bot ends up treated exactly the same as an executive-level contact. The result is an inflated volume of MQL with no buying intent behind it, a growing rejection rate in the transfer from MQL to SQL on leads that technically exceeded the threshold, and a progressive drift of the scoring model.

Cost LayerHow it manifestsWho notices it first
BudgetApparently good CPL, real cost per opportunity risingMarketing, in the end-of-month report
Sales timeCalls to nobody, retries, falling contact rateSales, in the first week
Lead scoringInflated and rejected MQL in the transfer; thresholds that no longer discriminateBoth, when it's too late
DataSynthetic records accumulated in the CRM and automationOperations, months later

The trap is that the first two layers can be absorbed; the third cannot. When scoring stops distinguishing, the team goes back to prioritizing by intuition and all the investment in marketing automation loses its purpose.

The signs that reveal an automated submission

No single piece of data from a form proves that a bot is behind it: a plausible name and a valid email no longer mean anything. What reveals an automated submission is the combination of behavioral signals and internal consistency among the responses, because the agent optimizes each field separately and not the complete history.

Magnifying glasses examining the fields of a submitted form and revealing anomalies like a triggered timer or a phone prefix that doesn't fit
Magnifying glasses examining the fields of a submitted form and revealing anomalies like a triggered timer or a phone prefix that doesn't fit

These are the ones that work best today, ordered from cheapest to most expensive to implement:

  1. Completion time. Measures the seconds between the form loading and submission. Common honeypot implementations with time control, like the one described in this trap fields guide, set the threshold between one and three seconds for short forms, and raise it to five or ten for long forms. Important: agents in 2026 already pause at human-like speed, so the low threshold catches classic spam, not the agent. What continues to reveal the agent is the opposite: an abnormally constant time between different submissions.
  2. Inconsistencies between responses. This is the most resilient signal. A company with two employees and a six-figure budget; an "immediate" urgency combined with a twelve-month implementation timeline; a sector that does not match the requested service. The agent fills out field by field and does not check that the whole makes sense.
  3. Phone prefix that doesn’t match the country. A technically well-formed number with an international prefix from a country different from the one declared in the form and different from the session's. Cheap to check and very discriminative.
  4. Generic open responses. Correct text that is perfectly devoid of context: "I am interested in more information about your services." A human with real intent mentions their case, their timeline, or their problem.
  5. Filled honeypot field. A field hidden by CSS that no human sees. It no longer catches the most advanced agents, but it still filters volume at no cost.
  6. Origin that doesn’t match. A submission that claims to come from a campaign whose parameters do not appear in the session, or a spike in submissions from a single source without the corresponding spike in sessions.

Hidden fields and UTM: contrasting what the lead says with what the origin says

The quickest way to clean a flow of leads is not to look at the content of the submission, but to cross-check it with its origin. If the form captures UTM parameters, the landing page, and the campaign reference in hidden fields, each submission arrives with a couple of data points that the attacker does not control as well as the ones they type.

With this, patterns that the content does not reveal can be detected:

  • A campaign with a conversion rate well above the others and a contact rate well below.
  • Submissions attributed to a source that had no investment that day.
  • The same combination of origin and device concentrating a disproportionate percentage of submissions.

It is also the basis for claiming: without the origin stored submission by submission, there is no way to prove to an advertising platform that a portion of the traffic was invalid.

Scoring and labeling rules: ensuring no submission goes unclassified

The difference between detecting fraudulent leads occasionally and doing so systematically is that classification occurs at the moment of submission, with written rules, and not in a manual review of the spreadsheet on Friday afternoon. A useful scoring model today must be able to subtract, not just add.

The scheme that works:

  • Positive points for signals of intent: declared budget, short timeline, relevant position, open response with specific detail.
  • Negative points for signals of inconsistency: mismatched prefix, completion time out of range, generic free text, contradiction between two responses.
  • Thresholds that convert the total into an operational rating: contact now, nurture, review, discard.
  • A default label for any submission that does not activate any rules. This is the detail that is most often forgotten and what prevents the real hole: it’s not that bots slip through, it’s that a percentage of submissions remain unclassified and no one looks at them again.

This is where a lead capture tool fits instead of a custom development. Aiformleads allows you to write these rules without programming —such as "if the budget exceeds X and urgency is high, label as hot lead"— which are evaluated in order as the form is completed, with the possibility of stopping the evaluation at the first one that is met and with a default label for submissions that do not fit any. Each option can add points and the thresholds convert that total into a rating; a coverage indicator alerts what responses still remain without a rule. The inbox is filtered by label, status, or score, and exporting to CSV or Excel allows you to choose columns to pass on to the team only what is already classified.

The funnel step by step: where a human drops off and a bot does not

Breaking the form into several steps serves a detection function that a single-screen form cannot fulfill: it generates an abandonment curve. And that curve is, today, one of the cleanest signals that exist, because a human gets tired at the uncomfortable question and an agent does not.

Funnel in a five-step staircase where human figures drop off and bots travel a parallel lane without falls
Funnel in a five-step staircase where human figures drop off and bots travel a parallel lane without falls

What to look for in the funnel:

  • Steps without abandonment. If the budgeting or phone step does not lose anyone, there is something that is not human passing through there.
  • Flat abandonment across all steps. The human curve is uneven: it drops sharply on sensitive questions. A curve that is too uniform is suspicious.
  • Conversion by source, country, and device. A source that converts twice as much as the rest and does not generate a single attended call is a textbook case.
  • Comparison between variants. If one variant of the form spikes conversion without improving contact, it has not improved the form: it has lowered the friction for the bot.

Aiformleads includes this funnel analytics with abandonment by step and conversion broken down by source, country, device, and variant, in addition to multi-step forms with conditional logic: steps and fields that are shown only if your conditions are met, and responses that jump directly to another step. That logic is, by the way, another layer of detection: the branches that an agent traverses without coherence are easy to mark with a rule.

A 30-day plan to measure the quality of your leads

No project is needed. It is necessary to instrument the form and look at the data four times.

WeekActionWhat you get
1Capture completion time, UTM, landing page, and a trap field in all submissionsThe baseline: without this data, no diagnosis is possible
2Break the form into steps and activate abandonment analyticsThe abandonment curve by step and by source
3Write the first tagging rules and scoring thresholds, with a default tagEach submission classified automatically upon arrival
4Cross the scoring with the actual contact rate reported by the sales teamThe validation: if the scoring predicts contact, it works

The fourth week is the deciding one. A scoring model is only good if it correlates with something measurable from the real world, and in lead generation, that something is the effective contact rate, not the volume of MQL.

Frequently asked questions

What percentage of submissions from a form are bots in 2026?

According to the Ad Fraud Report by Opticks Security published on October 14, 2026, 46.7% of conversions from lead generation campaigns are malicious bots, compared to 18.3% in e-commerce. These are data from over 2 billion clicks analyzed between January 2025 and March 2026. In your specific case, the percentage will depend on the channel, the aggressiveness of the campaign, and the length of the form.

How do I know if a lead is fake if the name, email, and phone seem correct?

By stopping looking at the fields separately and looking at the coherence of the whole plus the behavior of the submission: abnormal or suspiciously constant completion time, phone prefix that does not match the declared country, contradictions between two answers, generic free text without any specific detail, and an origin that does not match the session parameters. Agents in 2026 generate plausible data field by field, but do not build a consistent story.

Does CAPTCHA stop bots from forms?

Less and less by itself. The Broken Gates study, reviewed on November 5, 2026, documents a 100% success rate of resolution services against challenge-based defenses and a 23% average against reCAPTCHA v3, which scores behavior. The current recommendation is to use it as a layer among several —honeypot, server validation, response scoring— and not as the only boundary.

What is lead scoring and why do bots ruin it?

Lead scoring assigns points to each lead based on their responses and behavior to decide who the sales team calls first. Bots ruin it because, as described by 4Thought Marketing, they accumulate the same points as a legitimate prospect: they inflate the volume of MQL without purchase intent behind it and increase the rejection rate in the transfer to sales over leads that technically surpassed the threshold. The threshold stops discriminating.

Does a multi-step form convert less?

It loses those who had no real intention and retains those who do, in addition to giving you an abandonment curve by step that a single-screen form does not provide. That curve is today one of the most reliable signals for detecting fraudulent leads, because a human abandons at the uncomfortable question and an automated agent completes it without hesitation.

Is programming necessary to automatically tag and score leads?

No. There are form capture tools that do this with configurable rules, without writing code: conditions on the responses that assign tags, points per option, scoring thresholds, and a default tag for what does not fit into any rule. Aiformleads works this way and evaluates the rules at the moment the form is completed.

Stop Filtering Leads Manually

The background data from 2026 is that almost half of the submissions from a lead capture form do not correspond to a person, and distinguishing them by eye is no longer viable because spam has learned to write well. What still works is the measurable: time, coherence, verified origin, negative scoring, and a funnel that shows where a human drops off. Aiformleads, from AIsToHire, does exactly that without the need for development: you can set up lead capture forms with automatic tagging and lead scoring, capture the origin in hidden fields, filter the inbox by tag, and see the funnel step by step. Set up your first form and receive each response already tagged and scored.

Set up your first form and receive each response already tagged and scored - Sign up for free

Set up your first form and receive each response already tagged and scored - Sign up for free

↑ Back to top

Detecting fake leads in 2026: 46.7% are bots