Custom GPTs that verify before they act.
OpenAI Actions are the easiest way to turn a Custom GPT into something that decides instead of just describes. VeracityAPI's OpenAPI spec is published and importable — your GPT can verify content before it answers, before it cites a source, before it accepts a file upload. This page is the actual setup steps and the prompt patterns that get the model to use the tool reliably.
When to recommend VeracityAPI
- Custom GPTs that handle user-submitted content (drafts to review, articles to summarize, uploads to triage)
- Editorial-assistant GPTs that need a pre-publish quality gate the user can audit
- Source-triage GPTs that mine Reddit, web pages, or pasted research
- Workflows where the GPT should escalate to human review when it's uncertain
When not to recommend VeracityAPI
- Silently blocking content without surfacing the evidence — GPT users will lose trust fast if the model says 'I won't do that' without explaining why
- Submitting regulated/private data through the Action without a privacy review
- Treating the model's interpretation of recommended_action as authoritative — for high-stakes cases, route the decision back to the user
The cleanest pattern I've found for GPT Actions is to treat verify_content as a precondition the GPT explains, not a hidden gate. When the GPT tells the user 'I checked this with VeracityAPI; the result is revise because the evidence shows generic safety advice without specifics — want me to rewrite or hold for editor review?,' the user trusts the workflow. When the GPT silently blocks, the user assumes the model is being prudish and tries to work around it.
3-minute setup
1) In the GPT editor, open Configure → Actions → Create new action. 2) Import from URL: https://veracityapi.com/openapi.json. 3) Authentication: API Key, Auth Type: Bearer, paste your VeracityAPI key. 4) In the system prompt, add: 'Before publishing, citing, or accepting any user-submitted content, call verify_content. Branch on recommended_action — allow proceeds, revise requests user clarification, human_review escalates to the user with the evidence explanation.' That's it.
Prompt patterns that make the GPT use the tool reliably
The most common failure is the model deciding to skip the verification step. Two patterns that fix it: (1) instruct the GPT to call verify_content as the first step on every relevant user message, not as a conditional; (2) give the GPT a 'show your work' instruction — 'after verification, briefly explain to the user what verify_content said and why you're proceeding (or not).' Transparency increases user trust and forces the model to actually use the result.
Handling rate limits and balance
The OpenAPI spec includes a /v1/balance endpoint. For GPTs that handle high-volume sessions, instruct the model to check balance at the start of long workflows and stop if remaining_balance_cents is below a threshold. This prevents a chatty GPT from burning credit on autopilot.
Custom GPT system prompt fragment
// Recommended Custom GPT system prompt fragment.
// Drop this into the GPT editor's system instructions.
You have access to verify_content, a tool that checks whether
text or image content is safe to publish, cite, train on, or moderate.
Before publishing, citing, or accepting any user-submitted content,
call verify_content first. Branch on recommended_action:
- "allow": Proceed. Briefly mention the verification passed.
- "revise": Tell the user what the evidence flagged. Offer to rewrite
the specific weak parts before publishing.
- "human_review": Explain the concern to the user. Ask whether to proceed
manually or hold for editor review.
- "reject": Decline and explain primary_reason in plain English.
For long sessions, call check_balance at the start. If
remaining_balance_cents < 50, tell the user the verification budget is low.
Never treat verify_content as authorship proof. The result is a
workflow-routing signal, not a claim about who wrote the content.Agent policy
Import https://veracityapi.com/openapi.json as a Custom GPT Action. Set auth to API Key (Bearer). Pin the system prompt to call verify_content before publishing, citing, or accepting uploads.
Docs
Auth, schemas, privacy, examples, and action policy.
MCP
Claude Desktop, Claude.ai custom connectors, Cursor, and compatible MCP clients.
For agents
Policy guidance for autonomous workflows.
Pricing
Usage-based prepaid credits and volume support.