Pre-Publish Audit Rules Explained

Why we audit before publishing

AI engines treat content from sites that look promotional or unsourced as low-trust. They cite high-trust pages and ignore promotional ones. Citadex's pre-publish audit exists to keep every article you publish through us on the high-trust side of that line — because a published article that gets ignored by ChatGPT is worse than no article at all.

The audit runs automatically on every article before the Publish button activates. If any rule fails, you get a clear violation list and the article cannot be published until the violations are resolved.

The four rules

1. No fabricated statistics without attribution

What it checks: Any sentence containing a statistic (a number followed by %, $, "users", "customers", year-over-year claims, market-share figures, etc.) must either:

• Cite a verifiable source (named report, study, public dataset), or
• Be tagged as first-party data with a link to where the data is documented on your site

Why: Made-up stats are the single biggest reason AI engines decline to cite an article. "73% of B2B buyers..." without a source is a red flag that propagates across the entire piece.

How to fix: Either add a citation in markdown link form right after the stat, or remove the stat. If the stat is real but private (e.g. your internal customer count), publish it to a public page on your site first and link to that.

2. No more than two brand mentions in the body

What it checks: Direct mentions of your brand name in the article body (excluding the title, byline, and footer). Limit: 2.

Why: Articles that mention the brand five or six times read as promotional content even when they're factually useful. AI engines and human readers both downweight them. Strategic articles that establish category authority typically mention the brand 0–2 times in the body and let the byline carry attribution.

How to fix: Replace extra mentions with category language ("an AEO platform", "our tool", "the dashboard") or remove them. If you're writing a deliberate brand-comparison article where mentions are unavoidable, see Overriding the rules below.

3. No soft-sell CTAs in the body

What it checks: Phrases like "Sign up today", "Try our free trial", "Get started for free", "Schedule a demo", "Contact sales" appearing in the body of the article.

Why: CTAs in body content trigger AI engines' promotional classifiers. Even one soft-sell line can knock a quality article out of citation eligibility. CTAs belong in dedicated CTA blocks at the end of the post (and you can add those manually after publish — they're outside the audited body).

How to fix: Remove the CTA. If the surrounding paragraph needs a hook, replace with something educational ("This is one of the reasons we built [feature]") or factual.

4. Every data point must trace to a source

What it checks: Beyond stats (rule 1), claims like "according to industry research...", "experts agree...", "the leading platforms include..." need either a named source or removal. Vague-authority phrases without backing fail the audit.

Why: Vague authority claims read as filler. AI engines and human editors both downweight them.

How to fix: Name the source ("according to Gartner's 2025 AI report"), remove the claim, or rewrite it as a clearly first-party statement ("In our experience working with B2B brands, ...").

Severity levels

Each violation is tagged:

• Hard block (rules 1, 4) — publish is disabled until resolved
• Soft warning (rules 2, 3) — publish is disabled by default, but you can override with a one-line justification

Overriding the rules

For legitimate exceptions — typically explicit competitor comparison articles or product launch announcements that genuinely need to lead with the brand — click Override in the audit panel. You'll be asked to write a one-sentence reason ("this is a launch announcement"). The override is logged with the article for transparency.

Hard blocks (rules 1, 4) cannot be overridden. Cite the source or remove the claim.

When the audit gets it wrong

False positives happen. If the audit flags a statistic that does have a citation but the citation format wasn't recognized, or it flags a brand mention that's actually a quoted competitor name:

• Use Override if it's a soft warning
• Use the Report False Positive link in the violation card if it's a hard block — we adjust the detection rules based on these reports

How the audit interacts with Autopilot

When Autopilot is publishing on a schedule, the same audit runs. If an article fails:

• Soft warnings are auto-overridden with the note "Autopilot — scheduled publish" and logged
• Hard blocks pause the article and surface a notification — Autopilot won't publish unsourced content automatically

Frequently Asked Questions

Q: I have a comparison article that mentions our brand 8 times. How do I publish it?

Override the rule with the justification "competitor comparison article". The article is still published; the override is logged so reviewers know it was intentional.

Q: The audit flagged my own customer count as a fabricated stat.

That's rule 1 working — it can't verify your private data. Publish your customer count on a public page (about page, transparency page) and link to it from the article.

Q: Can I disable the audit for my account?

No, and we won't add this. The audit's value is that it's universal — disabling it for one account would mean some Citadex-generated content drifts back into promotional territory.

Q: What about articles I generate elsewhere and just want to score?

The audit only runs at publish time. If you're using Citadex only to score and not to publish, no audit fires. The Content Scorer in the AI Content Studio gives you the audit signals as advisory.

Q: Does the audit also check for spelling and grammar?

No. The audit is specifically about AEO trust signals, not editorial quality. Run your own editorial review separately.