Detect AI slop before it ships.
A drop-in linter for AI outputs. One API call returns the action your code routes on — allow, revise, human_review, reject — plus the evidence.
Think CI/CD checks, but for generated content.
Don’t build fragile probability thresholds. Just write your switch statement.
switch (result.recommended_action) {
case "allow": publish(); break;
case "revise": requestRevision(result.recommended_fixes); break;
case "human_review": queueForReview(result.evidence); break;
case "reject": block(); break;
}{
"recommended_action": "human_review",
"risk_level": "high",
"primary_reason": "unsupported_generic_claims",
"evidence_count": 3,
"confidence": "high"
}One API call. Four routing actions.
VeracityAPI returns a single field your code routes on. No probability threshold to tune. No "human or AI" coin flip. Four actions, evidence attached.
allowLow risk. Publish, cite, index, accept, or train. The default-permissive route.
reviseSend back to your agent or editor with evidence and recommended fixes. auto_revise:true does this in-line.
human_reviewQueue for moderation, source verification, or policy review. Uncertainty escalates.
rejectBlock, quarantine, discard. High-risk content does not reach users.
28 trust tools. 21 free.
The free ones run in your browser, as no-key demos, or as API-key-gated sample audits. The paid ones are one API call. Same pipeline.
21 tools. No signup to browse.
Browser-local checks, no-key demos, generators, and backend audit samples from the same canonical free-tool registry as /tools.
Veracity Text Linter
Paste a draft, get specificity, slop, and provenance feedback inline. Browser-only fast scan is free; deep semantic scan uses your API key.
Run check →Specificity Score Checker
Browser-local score for named specifics, numbers, source cues, concrete examples, and generic-pressure gaps.
Run check →Unicode Evasion Scanner
Find invisible characters, bidi controls, compatibility glyphs, and mixed-script lookalikes before text hits filters or agents.
Run check →Provenance Weakness Checker
Flag unsupported statistics, anonymous authority, overbroad claims, and text that needs better source cues.
Run check →Stylometric Fingerprint
Describe rhythm, lexical diversity, function-word density, punctuation profile, repetition, and readability complexity.
Run check →Pattern Pressure Checker
Replace naive burstiness with visible cliché, booster, em-dash, and formulaic-language pressure markers.
Run check →Rewrite Safety Checker
Compare original vs rewrite and catch changed numbers, URLs, quotes, entities, added claims, and removed caveats before publish.
Run check →Unsupported Claim Finder
Extract claims that need a source, screenshot, benchmark, citation, or human review — not a fake fact-checker.
Run check →RAG Source Grounding Checker
Check whether an AI/RAG answer stays grounded in supplied source text before agents cite or publish it.
Run check →Prompt Injection Pattern Detector
Scan user input, RAG chunks, scraped pages, and tool output for override, secret-extraction, jailbreak, and payload-smuggling patterns.
Run check →llms.txt Generator
Generate a clean llms.txt and llms-full.txt skeleton so answer engines and agents can find your docs.
Generate →AI Citability Checker
Score answer-engine readiness from specificity, visible sources, Q&A structure, freshness cues, and extractable sections.
Run check →PII + Invisible Character Cleaner
Find and redact emails, phones, Luhn-valid cards, API-key-shaped strings, JWTs, IPs, and hidden Unicode before LLM input.
Run check →Agent Publish Gate
Pre-publish gate for agent-written drafts: injection, PII, claims, grounding, disclosure, and citability checks.
Run check →CMS Publish Simulator
Preview CMS readiness for generated pages before they hit WordPress, Webflow, Shopify, or custom queues.
Run check →AI Disclosure Generator
Generate AI-assisted disclosure copy with human-review status, placement guidance, and schema notes.
Generate →Review Authenticity Checker
Triage review text for duplicate copy, incentives, hype pressure, no specifics, and absolutes without claiming fraud proof.
Run check →Testimonial Credibility Checker
Check testimonial attribution, consent, outcome claims, and concrete detail before publishing social proof.
Run check →AI Bot Policy Auditor
Authenticated backend audit for robots.txt, llms.txt, agents.json, and explicit AI crawler policy coverage.
Open audit →Sitemap Audit
Authenticated sitemap audit for HTTPS, host consistency, lastmod coverage, duplicates, and agent discovery readiness.
Open audit →Playground
Paste any text, see the live recommended_action JSON. No signup, no key, store_content=false forced server-side.
Open demo →Five endpoints. One pipeline.
Authenticated text surfaces share the same pipeline and expose action-routing where relevant. Balance returns account credit state for agent preflights. Start with $1.50 free credit, no card required.
Analyze
Text in. recommended_action out — allow, revise, human_review, or reject — plus evidence and recommended fixes.
Auto-revise
Set auto_revise:true. The API returns a fact-safe rewrite when routing to revise. Protected entities fail closed.
Deep Scan
Style-dimension scores, rewrite briefs, and semantic-risk flags from the Text Linter's authenticated tier.
POST /v1/analyze-style →Batch
Up to 25 items per call. Same per-item pricing. For RAG ingestion gates and backfills.
POST /v1/analyze-batch →Balance preflight
Free GET /v1/balance. Agents short-circuit cleanly when credit runs out mid-run.
Local or hosted. Claude, Cursor, anything MCP.
Local NPM wrapper or hosted Streamable-HTTP endpoint. Both expose the deterministic text tools plus analyze_text, analyze_batch, and balance checks; the local wrapper also includes broader verify/media helpers.
Try it.
No signup. Browser.Custom paste is primary: paste your own draft, review, caption, or source snippet. Demo calls force store_content=false, cap input at 4,000 characters, and are rate limited.
estimatedCostCents: 0.5 · $0.005 / 1k text chars
Recommended action: human_review. Replace generic warnings with named examples, sourceable details, and concrete workflow guidance.
Sign up free Want the full API output, SDKs, image analysis, and higher limits? Sign up free.
"should always stay alert"
Vague, universally applicable advice lacking specificity or actionable detail.
"Pickpockets are everywhere"
Sweeping generalization without supporting evidence or useful context.
"major European cities"
No named cities, neighborhoods, timeframes, or source details.
{
"analysis_id": "demo_01KRA1EQPDJ7N2KHBXCQMGZYFJ",
"modality": "text",
"content_trust_score": 0.22,
"specificity_risk": 0.78,
"provenance_weakness": 0.78,
"synthetic_risk": 0.72,
"slop_risk": 0.78,
"confidence": "medium",
"primary_reason": "unsupported_generic_claims",
"evidence": [
{
"type": "generic_phrasing",
"severity": "high",
"span": "should always stay alert",
"explanation": "Vague, universally applicable advice lacking specificity or actionable detail."
},
{
"type": "hedging_and_absolutes",
"severity": "high",
"span": "Pickpockets are everywhere",
"explanation": "Sweeping generalization without supporting evidence or useful context."
},
{
"type": "absence_of_specificity",
"severity": "medium",
"span": "major European cities",
"explanation": "No named cities, neighborhoods, timeframes, or source details."
}
],
"recommended_fixes": [
"Replace generic warnings with named examples, locations, and sourceable details.",
"Remove absolute claims unless they are supported by evidence.",
"Add concrete decision guidance for the intended workflow."
],
"risk_level": "high",
"recommended_action": "human_review",
"model_version": "v0.1",
"limitations": [
"Scores are probabilistic workflow risk signals, not proof of AI authorship or truth.",
"v0.1 uses an LLM-backed structured scoring pass; treat synthetic_risk as texture risk, not ground-truth authorship detection.",
"English-calibrated at MVP; non-English content should be treated as experimental."
]
}Does it actually work?
0.88 F1 on the seed-500.
Balanced five-slice text corpus, confusion matrix, and known limits live in /evals.
Haiku · temperature=0 · schema-constrained.
Deterministic given the input. Same pipeline behind every free tool.
See the pipeline →What we don't claimWorkflow signals, not forensic proof.
Not a binary "human or AI" detector. We say so on every page.
Methodology →Dogfooded on production publishing workflows for allow / revise / human_review decisions before publication. Honest internal dogfood, not a fake customer-logo strip.
Where it routes today.
RAG ingestion gate.
Score scraped, user-submitted, or partner content before it enters your index. Reject the bottom slice. Queue the uncertain middle.
See use cases →Pre-publish QA.
Run drafts through one call before publish. revise sends fixes back to the writer. human_review queues for an editor.
Agent output check.
Wrap your agent's final response. The four actions tell the orchestrator whether to ship, retry, escalate, or abort.
See use cases →Simple pricing.
$1.50 free credit · no card$0.005 / 1k chars
POST /v1/analyze — routing + evidence. The default surface.
$0.010 / 1k chars
auto_revise:true — adds a fact-safe rewrite when routing to revise.
$0.02 / image
POST /v1/analyze-image — synthetic-image workflow-risk triage.
Public demo remains free, no-key, capped, and rate limited. Authenticated agent preflight: GET /v1/balance.