Last updated: 2026-04-17
This post is an update announcement from FORMLOVA. It covers what changed at a high level -- see How to Use Sales Email Detection for a full usage walkthrough, and Why We Built Sales Email Detection for the rationale and design philosophy. The author is on the FORMLOVA team.
FORMLOVA now automatically classifies form responses and flags sales emails with AI. It is available on every plan, at no extra cost.
As far as we can tell, this is a first for form services. CAPTCHA-style bot protection has been standard across the industry for a long time, but human-written sales emails slip right through it. That is the gap we are closing.
What changed
- After each response is submitted, AI classifies the content into three labels: legitimate, sales, and suspicious
- Labels and scores show up directly in the dashboard's response list
- Saying "analyze without sales emails" in chat removes the sales-labeled rows from any analysis
- You can manually correct a label if the AI got it wrong, and your correction will not be overwritten
- When you publish a form that contains text inputs, FORMLOVA asks once whether to enable detection -- one click to opt in or skip
| Item | Detail |
|---|---|
| Available plans | Free / Standard / Premium |
| Extra cost | None (FORMLOVA absorbs the LLM cost) |
| Target forms | Forms with text input fields |
| Out of scope | Paid-event forms (Stripe Connect) |
Why we built it
I have spent a lot of time working on client inquiry forms, and I have seen situations where 8 out of 10 incoming messages were sales pitches. When you calculate CVR from paid-ad traffic, failing to exclude those pitches makes your cost-efficiency numbers meaningless.
Today, the standard way to strip them out is visual, one-by-one triage. Do it carefully and you burn hours; do it roughly and your numbers lie. Either way, you lose.
The goal of this release is to collapse that step on the receiving side of the form, so operators do not have to choose between accuracy and time. The full story is in Why We Built Sales Email Detection.
How to use it
This section covers the essentials. For a full walkthrough see How to Use Sales Email Detection.
1. Turn it on when publishing a form
When you publish a form that includes text fields (text, textarea, email, URL, phone), the FORMLOVA MCP server always asks whether to enable sales email detection. Pick "enable" or "skip" based on how the form will be used.
2. Check labels in the dashboard
The response list in the dashboard shows each reply's label (legitimate / sales / suspicious) and its score (0-100). Filter by label to see only sales entries or only the uncertain ones.
3. Exclude sales emails from analysis
Example chat:
Give me this month's CVR, excluding sales emails.
That one line returns a CVR figure calculated after dropping the sales-labeled responses. The same control is available as the exclude_sales parameter on the MCP tools get_responses and export_responses.
4. Correct mistakes manually
AI judgments are not perfect. A legitimate inquiry can occasionally get flagged as sales, and vice versa. Editing the label in the dashboard locks it in -- subsequent automatic classifications will not overwrite it.
The stance is simple: AI proposes, humans decide. Label changes are recorded in the audit log as well.
How it works under the hood
- Classification runs asynchronously after the response is saved. It does not add any latency to the submission the respondent sees.
- If classification fails or times out, the form submission itself is never broken -- the response is stored without a label.
- Each classification costs roughly $0.0002. We absorb that cost, and there is no metering on the user side.
- To mitigate prompt injection, system messages and the respondent's content are kept in separate roles when sent to the model.
- Email addresses in the response content are masked down to their domain portion before anything reaches the model.
What comes next
This release classifies across three buckets -- sales, legitimate, and uncertain. The same machinery naturally extends to finer intent classification within the legitimate bucket (for example: evaluation, information gathering, support, partnership).
Once intent is known, pairing FORMLOVA's MCP server with other services' MCP servers lets operators route responses conditionally in a single chat sentence: "send evaluation responses to sales Slack," "push information-gathering responses into the CRM and stop there." That composability is hard to reproduce with a form service on its own.
This release is the first step in that direction.
FAQ
Are responses flagged as sales automatically deleted?
No. The label is metadata. The response itself is stored and retained as usual. You can change the label at any time.
Does classification run on paid-event forms?
No. Forms that accept payments via Stripe Connect are excluded. Spammers rarely spend money to send sales pitches, so running classification there is a poor cost-accuracy trade-off.
Which model do you use?
Claude Haiku 4.5 via OpenRouter. It is a lightweight, fast, low-cost model; most of the accuracy work happens on the prompt side.
How accurate is it?
It is improving continuously. The prompt deliberately instructs the model to default to "legitimate" whenever it is unsure, so false sales flags on legitimate inquiries are kept low. If you spot a mistake, edit the label -- your correction will not be overwritten by later runs.
Can I turn it on for existing forms?
Yes. Use the dashboard toggle or ask in chat ("turn on sales email detection for this form"). It takes effect for new responses going forward.
Summary
- Sales emails in form responses can now be detected automatically by AI
- Available on every plan, no extra cost
- The goal: cleaner CVR numbers and less manual triage
- One step at publish time turns it on
- AI proposes, humans decide -- you can always correct the label
Related articles:
- How to Use Sales Email Detection -- enabling, dashboard, exclusion, workflows
- Why We Built Sales Email Detection -- background and design philosophy
The fastest way to see it is to try it on one of your own forms.
