Training & Accuracy
FormSentry uses AI to classify form submissions. You can improve accuracy for your specific forms by providing training examples.
How Classification Works
Section titled “How Classification Works”When a submission arrives, FormSentry classifies it using:
- Form context — your form’s name, description, and expected fields
- Training examples — submissions you’ve marked as legitimate or spam
- AI analysis — the AI model considers all context to make a decision
Training examples are the most effective way to improve accuracy. They teach the AI what your specific form considers legitimate vs spam.
Adding Training Examples
Section titled “Adding Training Examples”From the Dashboard
Section titled “From the Dashboard”- Go to Submissions in your FormSentry dashboard
- Find a submission that was misclassified
- Click the submission to view details
- Click Mark as Legitimate or Mark as Spam to reclassify it
The reclassified submission automatically becomes a training example. Future submissions similar to it will be classified according to your correction.
Training examples are scoped per form — examples added to one form do not affect other forms.
Tips for Good Training Data
Section titled “Tips for Good Training Data”- Add examples of both types — the AI needs to see what you consider legitimate AND what you consider spam for your form
- Include edge cases — borderline submissions are the most valuable training examples
- Be consistent — if similar submissions are marked differently, the AI will be uncertain
- Start with 3-5 examples per category — even a small number of examples significantly improves accuracy
- Add more as you go — continue marking misclassified submissions over time
Up to 5 legitimate and 5 spam examples are considered as context per request. If you have more, the most recent are used.
How Training Improves Accuracy
Section titled “How Training Improves Accuracy”Without training data, the AI relies on general spam signals and your form’s name and description. This works well for obvious spam but may miss form-specific nuances.
With training data, the AI learns your form’s specific patterns:
- What topics are on-topic vs off-topic for your form
- What writing styles your legitimate users have
- What types of submissions you consider spam
Example: A technical support form for a software product might receive a well-written message about a competitor’s product — without training, the AI might classify this as legitimate. After adding it as a spam training example, similar off-topic submissions are correctly flagged.
Form Context
Section titled “Form Context”Your form’s name and description also influence classification. The AI receives the form name, description, and list of expected fields with each request. Be specific:
| Description | Accuracy Impact |
|---|---|
| ”Contact form” | Low — too vague for the AI to determine relevance |
| ”Technical support for Bright Solutions software products” | High — gives the AI clear context about what’s relevant |
See Form Configuration for more on setting up form context.
Confidence Scores
Section titled “Confidence Scores”The confidence field in the API response indicates how certain the AI is:
| Score | Meaning |
|---|---|
| 0.9 — 1.0 | Certain |
| 0.7 — 0.89 | High confidence |
| 0.5 — 0.69 | Moderate — may benefit from training examples |
| 0.3 — 0.49 | Low confidence |
| 0.0 — 0.29 | Uncertain — training examples strongly recommended |
When your plan’s AI limit is reached, FormSentry falls back to rule-based detection, which always returns a confidence of 0.7.
If you see many submissions in the 0.5—0.7 range, adding training examples for those types of submissions will push accuracy higher.
Exact Matches
Section titled “Exact Matches”If a submission exactly matches a training example (same content in all fields), it is instantly classified without calling the AI model. This provides:
- Instant response times
- 100% confidence score