Social media caption pre-flight

Business value

• Improves caption specificity before posts go live.
• Reduces generic social copy that gets scrolled past or deprioritized.
• Creates a cheap QA step across high-volume short-form workflows.

Agent job to be done

Act as a caption quality reviewer. Preserve the creative idea, but block or rewrite captions that lack concrete details, platform-native specificity, or credible hook context.

format: captionintended_use: publishdomain: social media / travel safety

When to call VeracityAPI

Run after final caption draft and hashtag generation, before scheduling or publishing.

What text to submit

Caption text, hook, on-screen text, CTA, and optionally first comment. Do not include unrelated transcript unless the caption depends on it.

Decision policy

• allow: publish when low risk and evidence is empty or minor.
• revise: rewrite when slop_risk >= 0.40 or recommended_fixes mention adding concrete details.
• human_review: hold when high risk, especially for safety/scam claims or sensitive destination advice.
• local override: require at least one named place, mechanism, customer detail, or concrete claim before publishing.

Automation recipe

• Scheduler creates a post package: video asset, caption, hashtags, platform target.
• Agent scores caption with format=caption and intended_use=publish.
• If revise, caption agent rewrites using evidence spans and brand voice constraints.
• Rescore once. If still high, send to human review or publish with a safer minimal caption.

Evidence spans agents should inspect

• generic hooks like ‘don’t make this mistake’ without specifics
• broad claims about places or people
• caption bloat that does not add context
• salesy or inauthentic phrasing

Policy pseudocode

if (result.recommended_action === "allow") continueWorkflow();
if (result.recommended_action === "revise") rewriteWith(result.evidence, result.recommended_fixes);
if (result.recommended_action === "human_review") queueForHumanReview(result);
if (result.recommended_action === "reject") discardOrRebuild();

KPIs to track

• caption revise rate
• average reach per post before/after
• saves/shares/comments per impression
• percentage of posts requiring human review
• caption rewrite latency

What can go wrong

• Social algorithms are not directly measured by VeracityAPI.
• Short captions may produce lower confidence; pair score with local heuristics like named-place checks.
• Do not optimize away brand personality just to reduce risk.

Cost and latency notes

Analyze only is $0.005 per 1,000 characters; Analyze + revise with auto_revise=true is $0.010 per 1,000 characters. Both round up to the nearest 1,000 characters. Short captions/emails usually cost $0.005; longer pages or chapters scale linearly by length. Current v0.1 latency is LLM-bound, so batch/concurrent orchestration is recommended for high-volume pipelines.