Last updated: October 15, 2026
A useful ChatGPT form-analysis prompt defines the decision, period, columns, output, and evidence. Choose one task such as theme classification, request counts and traceable IDs, and review the result. Do not let the model invent operational columns missing from a CSV.
Prepare the Input First
MCP CSV/Excel exports contain form-answer columns and submission time, not native response IDs, status, owner, notes, or tags. Export covers the whole form; filter a controlled working copy afterward if needed. For ID/status analysis, retrieve supported response details separately rather than inventing CSV columns or joining by raw contact information. Email and phone values returned through MCP are masked; never reuse a masked value as a recipient or CRM identity key. Native Sheets automatic sync likewise writes submission time and form fields; additional owner/status columns belong in a separate working ledger. Free owner notifications are a six-hour digest, exempt from the email quota. They are different from quota-counted custom/respondent email and from a chat-generated backlog report.
The schemas and owner/deadline examples below describe an operating model, not automatic response columns. Native FORMLOVA status accepts only new, in_progress, resolved, and spam. Keep ownership, waiting stages, SLA timestamps, and AI suggestions in documented notes/tags, purpose-built form fields where appropriate, or a separate ledger. A needs_review flag is a review convention, not a fifth native status. When tags are updated, read and preserve the existing array. Confirm each approved status change by response ID and retrieve the record again.
Use the smallest relevant dataset. A practical structure is:
Proposed analysis or ledger schema; fields must exist or remain unknown. This is not an executed output or a native CSV schema.
response_id
submitted_at
form_type
category
score
message
status
owner
first_response_due_at
followup_permission
source
Explain the columns in the prompt. Preserve response IDs, because analysis without traceability cannot drive work.
Before sending the data:
- remove columns the task does not require;
- mask names, email addresses, phone numbers, and other direct identifiers where possible;
- replace account or company details with controlled labels;
- define the date range and timezone;
- state how tests, duplicates, and sales pitches are represented;
- keep the original response in the authorized source of truth.
Try one theme-classification contract
This is a proposed exercise, not an executed result. Begin with a draft or fictional data; use real responses only after defining the form, period, and allowed data.
Draft an unpublished survey with a 1–5 score, reason, and contact permission. For anonymous fictional responses I provide, return up to five defined themes, counts, source IDs, paraphrases, unclassified items, and review questions. Confirm actual columns and record count before using real data. Do not disclose personal details, infer causes, or send messages.
Check multi-theme counting and missing responses. If using a CSV-only file, assign working row IDs without claiming they are native FORMLOVA response IDs, and reconcile batch totals before accepting a conclusion.
Free covers form creation, publishing, collection, response search/status, export, and basic analytics. AI reasoning here runs in your connected client under that provider’s terms and any separate fees. Custom auto-reply, reminders, imports, detailed analytics, file uploads, and native automatic Sheets sync require Standard or above; bulk email and sequences require Premium or above. External-MCP recipes can be run from Free when the necessary external connections, permissions, and provider conditions are met. They do not start background monitoring merely by being copied into chat.
Create a survey designed for analysis
Start with the draft and review exercise above on Free, then add paid automation only when the operation needs it.
Sign up free and draft this form
After signup and email verification, connect a supported AI client and use the request above. Review its draft before publishing; the AI provider’s availability, terms, and fees also apply.
Use This Reusable Guardrail Block
Add this block to the end of each prompt.
Proposed analysis or ledger schema; fields must exist or remain unknown. This is not an executed output or a native CSV schema.
Guardrails:
- Use only the supplied records and field definitions.
- Preserve response_id in every record-level output.
- Do not expose names, email addresses, phone numbers, or other contact details.
- Do not invent a missing owner, deadline, score, or outcome.
- Use needs_review when evidence is ambiguous.
- Separate observed facts from inferences.
- Do not send messages, assign owners, change status, delete, or exclude records.
- Do not generalize from a small sample without stating the limitation.
This does not replace your organization’s privacy or review process. It makes the intended model behavior explicit.
Prompt 1: Create a First-Pass Summary
Use this prompt to understand the dataset before asking narrower questions.
Proposed analysis or ledger schema; fields must exist or remain unknown. This is not an executed output or a native CSV schema.
You are assisting with form response analysis.
Scope:
- Form type: [contact / booking / application / event / survey]
- Date range: [start] through [end]
- Timezone: [timezone]
Column definitions:
- response_id: stable identifier
- category: respondent-selected category
- score: optional rating
- message: free-text response
- status: new, in_progress, resolved, or spam
- owner: current accountable role or person; may be empty
Return:
1. A three-sentence summary of the response set.
2. Up to five major categories or themes with counts.
3. Urgent, low-score, or unhandled records in a separate table.
4. Potential tests, duplicates, or sales pitches in a review table.
5. Three questions a person should investigate next.
6. Data limitations.
[Add the reusable guardrail block.]
Use the output as an index, not a final report. Review exceptions and definitions before accepting trend language.
Prompt 2: Classify Open-Ended Responses Into Themes
Classify the supplied open-ended responses into a maximum of seven themes.
Requirements:
- Define each theme in one sentence.
- Assign one primary theme and optional secondary theme per response.
- Keep uncommon but high-impact responses visible.
- Do not create an “Other” theme larger than any named theme without explaining why.
- Return a crosswalk table: response_id, primary_theme, secondary_theme,
evidence_summary, confidence, needs_review.
- Return a theme summary: theme, count, share, representative meaning,
possible next action.
- Paraphrase examples; do not quote personal details.
[Add the reusable guardrail block.]
Theme counts are descriptive. They do not establish statistical significance or explain cause. For survey-specific interpretation, use Survey Analysis AI.
Prompt 3: Find Low Scores and Follow-Up Candidates
Proposed analysis or ledger schema; fields must exist or remain unknown. This is not an executed output or a native CSV schema.
Review the supplied responses for low-score or follow-up candidates.
Candidate rules:
- score is [threshold] or lower; or
- message describes strong dissatisfaction, cancellation, inability to use the service,
an approaching deadline, or a request for contact; and
- followup_permission permits contact, when that field is required.
Return:
| response_id | trigger | evidence | status | owner | contact_allowed |
| priority_candidate | human_check | next_action_candidate |
Rules:
- A low score does not automatically authorize contact.
- Do not draft or send a response.
- Put unclear consent or meaning in needs_review.
- State the number of records reviewed and selected.
[Add the reusable guardrail block.]
This prompt determines reading order. It does not decide refunds, remediation, or the substance of a reply.
Prompt 4: Classify Inquiry Text
Proposed analysis or ledger schema; fields must exist or remain unknown. This is not an executed output or a native CSV schema.
Classify each contact-form inquiry.
Allowed values:
- category: sales | billing | support | recruiting | press | other
- priority: high | normal | low | needs_review
- owner_candidate: sales | finance | support | recruiting | communications | unknown
- message_type: legitimate | sales_pitch | spam | test | needs_review
Return:
response_id, category, priority, owner_candidate, message_type,
evidence_summary, confidence, next_action_candidate.
Do not treat promotional language as proof of spam.
Do not infer urgency from tone alone.
Do not assign a named person or take an action.
[Add the reusable guardrail block.]
For production label design and evaluation, continue to AI Inquiry Classification.
Prompt 5: Build an Unhandled-Response Queue
Use this only when the required native status and ledger timestamps have actually been supplied. A plain export cannot establish ownership, overdue work, or SLA compliance. For CSV-only analysis, use theme/score prompts and working row IDs instead.
Proposed analysis or ledger schema; fields must exist or remain unknown. This is not an executed output or a native CSV schema.
Create a review queue from the supplied open responses.
Queue conditions:
- status is new; or
- owner is empty and status IN [new, in_progress]; or
- first_response_due_at has passed without first_human_response_at; or
- next_review_at has passed for a waiting response; or
- priority is high and status IN [new, in_progress].
Order:
1. high-priority overdue
2. high-priority unassigned
3. due within the current operating window
4. normal overdue
5. waiting review overdue
6. remaining new responses
Return:
response_id, queue_reason, deadline, elapsed_time, owner,
evidence_summary, next_action_candidate, needs_review.
Use timestamps from the data; do not calculate business hours unless a calendar rule is supplied.
[Add the reusable guardrail block.]
The status and SLA model behind this prompt is explained in Unhandled Form Responses.
Prompt 6: Analyze a Large CSV in Batches
Do not ask one prompt to solve every analytical job over a large file. First create a manifest and batch plan.
Inspect the CSV schema and propose a batching plan without performing the final analysis.
Return:
1. row count and column list;
2. missing-value summary for analysis fields;
3. proposed batches by date, form, or segment;
4. exception batches for low scores, unhandled work, and free text;
5. fields to remove or mask before analysis;
6. a stable output schema that can be merged across batches.
Do not infer missing values. Do not include personal data examples.
For each batch, reuse the same labels and schema. Then run a synthesis prompt:
Merge these batch summaries.
Do not add batch percentages directly; recompute from supplied counts.
Identify themes that appear across batches and themes limited to one segment.
Preserve all limitations and review flags.
Return reconciled totals before conclusions.
Prompt 7: Draft a Short Internal Report
Proposed analysis or ledger schema; fields must exist or remain unknown. This is not an executed output or a native CSV schema.
Convert the validated analysis into a one-page internal report.
Audience: [team or role]
Decision: [what the audience must decide]
Structure:
1. Scope and definitions
2. Three-line conclusion
3. Intake and operational health metrics
4. Important exceptions, with response IDs
5. Major themes and supporting counts
6. Up to three recommended actions, each with owner role and review date
7. Limitations and unresolved questions
Do not claim causality or confirmed downstream conversion unless the supplied
data includes a reconciled identifier and the method is stated.
Do not include personal contact details.
The framework for choosing report metrics is in Form Analysis.
Prompt 8: Draft a Reply Without Sending It
Response analysis and response sending must remain separate.
Draft a reply to the supplied response.
Requirements:
- Do not send it.
- Use [VERIFY] for facts, dates, prices, promises, or policies that need confirmation.
- Do not repeat personal details unless required for clarity.
- Do not promise an outcome that is not present in the approved facts.
- Match the requested tone: [tone].
- End with a checklist of facts a person must confirm before sending.
Human review is essential for complaints, refunds, contracts, hiring, security, health, legal, or financial matters.
Improve Weak Outputs Systematically
If the answer is generic, do not merely ask the model to “be more specific.” Diagnose the missing contract.
| Weak output | Likely missing element | Fix |
|---|---|---|
| Broad summary | Decision and scope | State audience, dates, and decision |
| Unstable categories | Controlled vocabulary | Define allowed labels |
| Unsupported priority | Evidence rule | Require trigger and response ID |
| Invented owner | Unknown state | Allow unknown and needs_review |
| Personal data in report | Output restriction | Require redaction and IDs |
| False trend | Sample limitation | Require counts, denominator, and caveat |
The most reliable prompt improvements are explicit instructions, examples of the desired output shape, and a clear separation between analysis and actions. See OpenAI’s current official prompt engineering guidance for general prompting principles; verify the current product documentation for the exact model and interface you use.
Use FORMLOVA When the Prompt Becomes a Routine
Copy-and-paste analysis is useful for one-off work. It becomes fragile when the same CSV preparation, field definitions, redaction, labels, and report format must be repeated every week.
FORMLOVA connects response retrieval and operational state to a user-controlled AI client through MCP. The model reasons in that client session; FORMLOVA does not need a hidden server-side general-analysis step. Reusable routes include the AI Response Report, Open-Text Theme Report, and Inquiry Weekly Status Report.
A Workflow page supplies a reusable chat prompt. Run it in an active client session and inspect the requested tools and plan requirements before execution; it does not install a per-response listener or schedule by itself. A recurring report/reminder requires supported Standard-or-above settings or a separately configured scheduler; external Slack/Sheets MCP access has its own connection and provider conditions.
For the operating model behind all of these prompts, read Chat With Form Responses. To try them against a managed response set, start with FORMLOVA and use the setup guide.
For the broader system around response notifications, exports, reports, and follow-up, continue to the FORMLOVA Form Automation Guide.
Disclosure and Verification
- FORMLOVA product and MCP statements were checked against the local SPEC and implementation on August 13, 2026.
- General prompt-design terminology was checked against OpenAI’s official prompt engineering documentation. This article intentionally makes no claim about current ChatGPT plans, file limits, retention settings, or UI availability.
Rewrite verification (October 15, 2026): the form operations, plan boundaries, export schema, and proposed exercises above were checked against the current repository. Earlier dated research below or above remains historical; the fictional examples are not measured outcomes.
