AI Slop Typography Audit

AI-Slop Typography Audit

Acknowledgement: Shared by Peter Bamuhigire, techguypeter.com.

Use When

• Reviewing an existing artifact (site, DOCX/PPTX/PDF, UI screen) for typographic slop.
• A deliverable "feels AI-generated" and you need to find and fix the tells.
• Gate-checking a product before it ships under the Chwezi name.

Do Not Use When

• Choosing fonts from scratch — use font-selection-and-pairing.
• The artifact has no type, or fonts are fixed by a binding brand guideline.

Required Inputs

• The artifact or its source (CSS/HTML, the document file, screenshots, or a style spec).
• The intended context and design voice (so the remediation maps to the right font category).

Workflow

1. Detect primary typefaces. Read the CSS font-family stacks / document style / theme. Flag any banned face from doctrine/references/ai-slop-banned-fonts.md used as primary, and flag bare system stacks used alone.
2. Detect secondary slop. Flag Space Grotesk/Poppins/Montserrat/Nunito and standalone Source Sans used as the distinctive choice.
3. Detect structural slop: one font for everything; mid-weights only (no extremes); near-identical size steps; wide-tracked body; pure-black body text.
4. Score each finding (primary-font slop is critical; structural slop is major/minor).
5. Remediate deliberately: propose a non-slop pairing + scale for the artifact's category via font-selection-and-pairing; state the replacement and reason.
6. Report findings → fixes, with before/after.

Anti-Patterns

• Swapping Inter for another single sans and calling it fixed.
• Removing a banned font but keeping monotype, timid weights, or wide-tracked body.
• Calling a face slop from personal taste alone; cite the current list or a documented convergence tell.
• Recommending an unlicensed or unavailable replacement; route the brief through selection and scan.
• Auditing only headings; inspect body, controls, data, errors, and responsive states too.

Outputs

• A findings list (each tagged critical/major/minor with the rule it breaks) and a stated, non-slop remediation ready to apply.

Examples

• Use the worked example in examples/ as the required findings-to-remediation model.

Quality Standards

• Tie every finding to visible evidence and a current doctrine rule; taste alone is insufficient.
• Inspect display, body, data, and state text at representative sizes and widths.
• Block release for a banned primary face or faux typography until remediated and re-audited.

Decision Rules

Capability Contract

Read and visual inspection are required. Review remains read-only unless remediation is requested. Search may inspect styles; rendering is preferred. Current-definition research routes to the doctrine-refresh skill.

Degraded Mode

Without styles, audit the render and mark causes unverified. Without a render, return a checklist and do not issue a pass. If doctrine currency is uncertain, stop and route to refresh.

• examples/typography-audit-filled.md — a filled audit of a sample SaaS landing-page stylesheet (Inter body + Geist display + Space Grotesk accent over a bare system stack): each banned and secondary-slop face detected with the rule it breaks, scored critical/major/minor, then one deliberate non-slop remediation (Bricolage Grotesque 800 -> Hanken Grotesk 400, 03 Modern Product / Grotesque) mapped finding-by-finding with the before/after CSS.

References

• doctrine/references/ai-slop-banned-fonts.md, pairing-principles.md, type-scale-and-spacing.md.