Web60 Features
The CAPTCHA on Your Contact Form Is Quietly Turning Away Customers

The enquiries did not stop all at once. They thinned out over a few weeks in early autumn, which is exactly why nobody noticed. What follows is a composite, assembled from a pattern that turns up across small business websites rather than one firm's history, but every part of it is ordinary: a home-adaptations installer in Kerry, fitting stair lifts and level-access showers, whose customers are mostly in their seventies or are adult children filling in a form on a parent's behalf from a tablet at the kitchen table.
The website had been refreshed in the spring. Somewhere in that work, a Google reCAPTCHA box went onto the contact form to deal with spam. The phone kept ringing. Online, the form went quiet. And the spam? It kept arriving, a few junk messages a week offering SEO services and crypto partnerships, passing straight through the box that was supposed to stop them.
Nobody connected the two until a woman rang the office to say she had tried to send a request for a home assessment three times. The form asked her to pick out all the squares with traffic lights, and every time she got it wrong. She very nearly rang someone else instead. That story is the whole article in miniature. A CAPTCHA is a security control, and like every security control it has a failure mode. This one fails on your customers, silently, while the thing it was installed to block walks around it.
A test that bots pass and people fail
CAPTCHA is a simple bargain. Before the form will send, the visitor proves they are human by solving a puzzle a machine supposedly cannot. That bargain made sense when bots were crude. It makes far less sense now.
The most thorough recent look at this came from a research team led from the University of California, Irvine, published at USENIX Security 2023 [4]. Across their user studies, 1,400 participants solved 14,000 CAPTCHAs of the kinds in use on popular websites. Their findings suggest that bots outperform humans on both solving time and accuracy across a wide variety of CAPTCHA types. Read that again. The test designed to keep machines out is, on the evidence, easier for the machines.
Google is candid about the other half of the problem. Its own reCAPTCHA FAQ [1] acknowledges that being flagged "can unfortunately happen to good users", and gives the reasons: a shared network that someone else has abused, or an internet provider handing out an address with a poor reputation. Think about who sits behind those connections. A customer on hospital Wi-Fi visiting a parent. Someone on a mobile network that shares addresses across huge numbers of phones. A retired couple whose broadband provider recycles addresses that were misused by someone else last month.
The customers it excludes first
Then there is accessibility. The W3C, the body that sets web standards, has a working note devoted entirely to this problem [5]. Its wording is blunt: the interactive puzzle "inherently excludes many people with disabilities, resulting in a denial of service to these users." Low vision. Hand tremor. Cognitive fatigue.
For the Kerry installer, whose customers are almost by definition people for whom everyday physical tasks have become harder, the CAPTCHA was screening out the exact audience the business exists to serve.
The cost of getting this wrong is easy to describe. A customer who fails the puzzle twice does not ring you to explain. She closes the tab, searches again, and the next installer on the list gets the assessment, the quote and the job.
The invisible version fails more quietly
The usual response to complaints about puzzles is to switch to the invisible kind. reCAPTCHA v3 never shows a challenge. Google's documentation [2] says it "will never interrupt your users", and recommends running it on forms and "in the background of pages for analytics", because it "works best when it has the most context about interactions with your site." Instead of a puzzle, every visitor gets a score between 0.0 and 1.0. The website decides what to do with it, and Google suggests a default cut-off of 0.5. Anyone below the line is treated as a bot.
Operationally, this swaps a visible failure for an invisible one. That is worse, not better. With the puzzle, at least the customer knew something had gone wrong. A score threshold gives no such warning: a genuine enquiry that scores 0.4 might be rejected with a vague error or, depending on how the form plugin is configured, discarded while the visitor sees a cheerful thank-you message. The customer believes the request went through. You never knew it existed. Nobody complains about a message they think you received.
In the composite, this is the second chapter. Asked to get rid of the traffic-light puzzles, whoever maintained the site switched to the invisible version. The complaints stopped. So did a share of the enquiries, and this time there was no signal at all, just a quieter inbox that looked a lot like a slow season.
It is rarely just the contact page
Because v3 is designed to watch behaviour across a site, many setups load Google's script on every page, not only the contact form. That means third-party code running on your homepage, your service pages and your price list, for visitors who have no intention of filling in anything. Each external script is another request the browser makes before the page settles, and another dependency in your production environment that you do not control. On a fast, well-cached site that is a small cost. Still, it is a cost paid on every page view to protect a single form.
The consent question nobody mentioned
There is a privacy dimension too, and it is less settled than either Google or its critics tend to suggest. Google's position, in the same FAQ [1], is that reCAPTCHA "sets a necessary cookie (_GRECAPTCHA) when executed for the purpose of providing its risk analysis." Under European cookie rules, strictly necessary cookies are one of the narrow cases where you do not need the visitor's consent first.
Not every regulator has accepted that framing. In a 2023 decision against a scooter rental company, France's data protection authority, the CNIL, addressed the firm's use of Google reCAPTCHA when users created accounts and found it had failed to obtain the required consent, as the law firm Hunton Andrews Kurth reported at the time [6]. That was one regulator, one company and one set of facts. It does not make reCAPTCHA unlawful everywhere, and I am not aware of an Irish ruling on it specifically.
What it does mean is that reCAPTCHA is not a neutral, consent-free component you can bolt on without thinking. If you have gone to the trouble of running privacy-first analytics so your site does not need a consent banner, putting Google's script back onto every page reopens a question you thought you had closed. Should you keep it, mention it in your privacy policy and ask your solicitor whether your consent setup needs to cover it. That is a judgement for someone qualified to make it, not for a form plugin's default settings.

The key that sits in someone else's account
This is where people get caught out. Every reCAPTCHA installation runs on a pair of keys tied to a Google account. Google has now moved the older keys into its Cloud platform. Its migration overview [3] sets out the timeline: from the fourth quarter of 2025 it began automatically migrating Classic keys, provisioning a Google Cloud project for each and emailing the key owner an invitation to accept ownership, with the automated migration due to complete in the first quarter of 2026. Existing keys keep working without code changes. To create, delete or modify a key, though, someone has to accept that project, and configuration now happens in the Google Cloud console.
Now ask the obvious question. Whose email address received that invitation? On a lot of small business sites, the honest answer is the freelancer, agency or relative who set the form up years ago. In the Kerry composite, the keys lived in a Gmail account belonging to the person who refreshed the site in the spring. The owner could not see the scores, could not change the threshold and could not have said what the threshold was. A piece of their customer pipeline depended on an account they had never logged into.
On cost, the same overview says the free tier covers 10,000 assessments a month, with billing required above that. A small business contact form will not come close, so cost is not the risk here. Control is. Treat a reCAPTCHA key like any other dependency: verify who owns it, and if the answer is someone outside the business, fix that before you need to change anything.
What actually stops form spam
Step back from the tool and describe the job. A sound approach to contact form spam should pass five tests:
- The visitor never has to prove anything. Real customers should not be asked to solve a puzzle before they can ask you for a quote.
- Nothing is lost silently. A message that looks suspicious should be held somewhere you can review, not deleted.
- Protection is layered. No single check carries the whole load, so one false positive does not cost you a customer.
- It runs where it is needed. Protection for a contact form belongs on the contact form, not on every page.
- The business controls it. Settings and logs sit in accounts the owner can open.
Measured against that list, most small business sites can drop the puzzle entirely and use a combination of quieter measures.
A honeypot field
A honeypot is a form field hidden from people but visible to bots, which tend to fill in every box they find. Anything that arrives with that field completed is discarded. Customers never see it. The W3C note [5] puts the advice in a single line: if a honeypot suffices, use a honeypot until evidence of robotic attacks dictates something else. Most major WordPress form plugins include one, often as a single setting.
The same bots rarely stop at the contact form. If your site still accepts blog comments, they are using that door too, and there is a reasonable case for closing comments on a business website entirely.
A time check
Bots complete forms in a fraction of a second. People take a while to type out what they need. Rejecting submissions finished faster than any human could manage catches a surprising amount of automated junk, again without asking the visitor to do anything.
A spam filter with a review folder
A spam-filtering plugin reads the message itself and judges it on content and sender patterns. The setting that matters most is what happens to a suspected spam message. It should go to a folder in your WordPress dashboard, not into the void. Configure the form to store every submission in the database as well as emailing it, and a false positive becomes a quick weekly check rather than a lost customer.
A challenge only as a last resort
If a particular form genuinely is under sustained attack, privacy-focused CAPTCHA alternatives exist that verify visitors without image puzzles, and many can be set to step in only when other signals look wrong. Should you decide reCAPTCHA is still the right tool, load it on the contact page only, not across the whole site.
What the server handles before WordPress does
Underneath all of this sits the hosting layer, and it matters more than most owners realise. Much of the automated traffic hitting small business sites is not aimed at the contact form at all. It is hammering the login page and probing for known weaknesses. On Web60, that traffic meets server-level security hardening and fail2ban intrusion prevention, which blocks repeat offenders at the server, alongside automatic malware scanning. In practice, your form plugin deals with a smaller and less hostile crowd, and your site is not spending its resources answering bots that were never going to become customers.
It is the same layered thinking set out in our complete guide to WordPress security and backups, applied one level further down, and it is part of the server-level protection that comes with every Web60 site.
One thing none of this will do is stop a person. Server-level protection looks at traffic patterns, not at what a message says. Someone paid to paste SEO pitches into contact forms by hand will pass a honeypot, beat a time check and never trip fail2ban, because to every one of those layers they look exactly like a customer. A content filter will catch some of them. Inevitably, a few will reach your inbox. That is the honest floor for any public contact form, and it is the price of being reachable.
Changing it without losing a week of enquiries
Make the change in a staging environment first, so a misconfigured form never touches production. Deploy it, then verify it the way a customer would: send a test enquiry from your own phone on mobile data, not your office broadband, and confirm it arrives in your inbox and in the stored submissions. If anything looks wrong, roll back and try again before your customers find the conflict for you.

When a heavier tool is the right call
None of this means reCAPTCHA, or the wider bot-management products built around it, are useless. They were designed for a different scale of problem.
If you run a busy online shop whose checkout is being used for card testing, or whose customer logins face credential-stuffing attempts in their thousands, you need risk scoring tuned on vast amounts of traffic data and someone whose job it is to watch it. That workload genuinely suits enterprise bot-management services and the premium managed hosts that bundle a commercial web application firewall with a dedicated security team. For that business, the occasional false positive is an acceptable trade against fraud losses, and it will usually have the analytics to measure the trade properly.
A home-adaptations installer receiving a steady handful of enquiries a week is not that business. Neither is a dental practice, a café taking group bookings or a solicitor's office. For them, the puzzle costs more than it saves.
Back at the kitchen table
In the composite, the fix took an afternoon. The reCAPTCHA came off. A honeypot and a time check went on. Every submission was set to save in the dashboard as well as arriving by email, with suspected spam held for review rather than binned. Removing reCAPTCHA also removed the question of whose Google account held the keys.
The spam did not vanish. A few junk messages still turn up each week, most of them caught and parked in the review folder. The enquiries came back, though, including from customers who had clearly tried before. That is the trade worth making.
The question to ask about your own form
Spam is irritating. Losing a customer at the moment they decided to contact you is expensive, and far harder to see. Any protection on your contact form deserves to be judged on both counts, not only on how clean your inbox looks.
So open your own site on your phone, away from your office Wi-Fi, and fill in the form as a customer would. Notice whether it asks you to prove anything. Find out where submissions go when they are flagged, and whose account holds the keys. If you are starting from scratch, a Web60 site gives you full WordPress, the plugin choice to set this up properly and server-level protection underneath, all for €60 a year. Who your form turns away is now a decision you can make deliberately rather than one a default setting makes for you.
Frequently Asked Questions
Do I need a CAPTCHA on my business contact form?
Most small business websites do not. A hidden honeypot field, a minimum completion time and a spam filter that holds suspicious messages for review will stop the bulk of automated spam without asking customers to do anything. Keep a challenge in reserve for a form that is under sustained attack, and load it on that page only.
Does Google reCAPTCHA need cookie consent in Ireland?
It is not settled. Google describes its _GRECAPTCHA cookie as necessary for risk analysis, which would place it outside the consent requirement, but France's data protection authority, the CNIL, found in a 2023 decision that a company using reCAPTCHA had failed to obtain the required consent. We are not aware of an Irish ruling on reCAPTCHA specifically. If you use it, disclose it in your privacy policy and ask your solicitor whether your consent setup should cover it.
Why does my contact form still get spam with reCAPTCHA?
Two reasons. Automated tools have become very good at solving CAPTCHAs, and academic research published at USENIX Security 2023 found bots outperforming humans on both speed and accuracy across many common types. Some form spam is also sent by people typing by hand, who pass any human test because they are human. A content filter catches some of that, but none of it disappears entirely.
What is a honeypot field on a contact form?
A honeypot is a form field hidden from people but visible to bots, which tend to fill in every box they find. If a submission arrives with that field completed, the form treats it as spam and discards or holds it. Visitors never notice it, and most major WordPress form plugins offer one as a simple setting.
Is Google reCAPTCHA still free?
For a typical small business, yes. Google now runs reCAPTCHA keys through Google Cloud projects, and its migration documentation says the free tier covers 10,000 assessments a month, far more than a small business contact form will use. Above that, billing has to be enabled on the Google Cloud project.
Who owns my website's reCAPTCHA keys?
Whichever Google account created them. When Google automatically migrated older keys into Google Cloud projects, it emailed the key owner an invitation to accept the new project, and creating, deleting or changing keys now requires that access. If a former developer, agency or relative set up your form, the keys may sit in their account. Verify this, then either take ownership or replace the protection with measures you control.
Sources
Ian oversees Web60's hosting infrastructure and operations. Responsible for the uptime, security, and performance of every site on the platform, he writes about the operational reality of keeping Irish business websites fast, secure, and online around the clock.
More by Ian O'Reilly →Ready to get your business online?
Describe your business. AI builds your website in 60 seconds.
Build My Website Free →More from the blog
Payment Test Mode Left On: The Staging Push That Took No Money
Payment test mode left on after a staging push leaves a shop that looks open but takes no money. Where the switch hides and what to verify after every change.
ChatGPT WordPress Fixes: Five Myths That Break Business Websites
ChatGPT WordPress fixes look easy: ask, copy, paste. Five myths about AI code and plugin advice, and the safe way to try a fix before customers see it.
