Home/ Skills/ ai-content-editing

general ai-content-editing

ai-content-editing

This skill should be used when the user asks to "edit AI-generated content", "quality check AI content", "review AI writing", "fix AI-generated text", "make AI content sound human", "improve AI-written content", "AI content quality control", "humanize AI content", or any variation of editing, reviewing, improving, or quality-controlling content generated by AI tools like ChatGPT, Claude, or other LLMs for B2B SaaS publishing.

Download .md

AI Content Editing

AI-generated content is a first draft, never a final draft. LLMs produce fluent, grammatically correct text that reads like a competent but generic textbook. Publishing it unedited signals "no human expert was involved" to readers, Google, and AI search engines. AI content editing transforms raw LLM output into content that's authoritative, specific, and distinct.

The goal is not to remove every trace of AI. The goal is to add what AI cannot: proprietary data, expert judgment, specific examples from real experience, and a distinct point of view.

The AI Content Editing Framework

Edit AI content in five passes. Each pass targets a different layer of quality.

Pass	Focus	What you're fixing	Time
1. Accuracy	Facts, claims, numbers	Wrong information, hallucinations, outdated data	10-15 min
2. Specificity	Vague → specific	Generic statements that could apply to any company	10-15 min
3. Voice	Tone and style	AI-default voice → brand voice or expert voice	10-15 min
4. Structure	AEO readiness	Format for AI search extraction	5-10 min
5. Originality	Unique value	Add proprietary data, POV, or insider knowledge	15-30 min

Total: 50-85 minutes per 1,500-word piece. This is the minimum investment to transform AI output into publishable content.

Pass 1: Accuracy Check

AI hallucinations publish false information under your brand. This is the most important pass.

What to verify

Claim type	How to verify	Common AI errors
Statistics and numbers	Check original source	AI invents plausible-sounding statistics. "Studies show 73% of..." with no study
Product features	Check product website/docs	AI describes features that don't exist or describes them incorrectly
Pricing	Check current pricing page	AI uses outdated pricing or makes up price points
Company information	Check company website	AI confuses companies, invents founding dates, misidentifies founders
Process steps	Verify against documentation	AI invents plausible but incorrect process steps
Tool integrations	Check integration docs	AI claims integrations exist when they don't
Historical claims	Check primary sources	AI invents historical events or attributes quotes incorrectly

Accuracy rules

Never publish a statistic you can't source. If the AI claims "67% of sales teams do X" and you can't find the source, delete the statistic. Replace with a factual claim you can verify
Delete every unsourced "according to" and "studies show." If no study is cited, the AI invented it
Check every product name, feature name, and company name. AI frequently misspells product names or confuses similar products
Verify all URLs. AI generates plausible-looking URLs that don't exist

Pass 2: Specificity Edit

AI defaults to generic, safe statements. Specific content outperforms generic content in both SEO and AEO.

The specificity test

For every sentence, ask: "Could this sentence appear on any competitor's website unchanged?" If yes, it's too generic. Rewrite.

Generic (AI default)	Specific (edited)
"Many companies struggle with this challenge."	"B2B SaaS teams with 5-20 SDRs typically hit this problem when monthly lead volume exceeds 500."
"This can lead to significant improvements."	"Teams using this approach see 23% higher reply rates within the first 30 days, based on data from 200+ sequences."
"There are several tools available."	"The three most common tools are Apollo ($49/mo), ZoomInfo ($14,995/yr), and Clearbit (usage-based pricing starting at $99/mo)."
"Best practices suggest doing this regularly."	"Update comparison pages monthly. Pricing pages within 24 hours of any price change."

Specificity rules

Replace every "many" with a number. "Many companies" → "67% of B2B SaaS companies" or "companies with 50-200 employees"
Replace every "significant" with a measurement. "Significant improvement" → "23% increase" or "from 3 days to 4 hours"
Name tools, companies, and people. "Popular CRM" → "HubSpot, Salesforce, or Pipedrive"
Add time frames. "Regularly" → "monthly" or "every 2 weeks"
Add qualifying context. "Small businesses" → "B2B SaaS companies with $1-5M ARR and 10-50 employees"

Pass 3: Voice Edit

AI has a recognizable voice: formal, hedged, overly balanced, and devoid of personality. Edit for distinctiveness.

AI voice markers to eliminate

AI marker	Example	Fix
Hedge words	"It's important to note that...", "It's worth considering..."	Delete. Start with the actual point
False balance	"Both approaches have their merits..."	Take a position. "Use X when [condition]. Use Y otherwise."
Filler introductions	"In today's rapidly evolving landscape..."	Cut entirely. Start with the first substantive sentence
Passive voice default	"It has been observed that..."	Active voice: "We see..." or "[Subject] does..."
Em-dash overuse	"This approach — which many teams find effective — can..."	Restructure to two sentences or use parentheses sparingly
Excessive connectors	"Moreover, furthermore, additionally, it's also important..."	Cut. Each paragraph stands on its own
Summary conclusions	"In conclusion, we've covered..."	Cut. The content speaks for itself
Rhetorical questions	"But what makes this so important?"	Delete. State the answer directly
Superlatives	"One of the most powerful ways to..."	Delete "one of." Use concrete evidence instead

Brand voice calibration

Dimension	AI default	B2B SaaS target
Formality	Formal, academic	Conversational but competent. Peer-to-peer
Confidence	Hedged, cautious	Direct, opinionated. "Do X" not "Consider X"
Length	Verbose, padded	Terse. Cut 30% of AI word count
Perspective	Third-person, detached	First-person plural ("we've seen") or second-person ("you should")
Personality	None	Distinct POV, occasional strong opinions

Pass 4: Structure Edit (AEO)

Format the edited content for AI search extraction.

Structural checks

[ ] First 50 words contain a direct answer to the page's primary question
[ ] All H2s are question-shaped
[ ] Every comparison is in a table, not prose
[ ] Every paragraph leads with its key claim (inverted pyramid)
[ ] No paragraph exceeds 4 sentences
[ ] Lists are formatted as numbered or bulleted lists, not inline prose
[ ] FAQPage schema applied if page has Q&A sections

Pass 5: Originality Edit

This is the pass that transforms AI content from "competent" to "worth citing." Add what AI cannot generate on its own.

What to add

Addition type	Example	Impact
Proprietary data	"Across our 500+ customers, we see average reply rates of 3.1% for cold email"	Very high — unique, citable data
Expert judgment	"In our experience, teams that skip the ICP step waste 40% of their first quarter"	High — human expertise AI can't replicate
Named case examples	"When Acme Corp switched to this approach, they cut ramp time from 90 to 45 days"	High — specific, verifiable
Contrarian takes	"Most guides tell you to personalize every email. That's wrong at scale. Here's why:"	High — distinctive POV
Insider knowledge	"The feature most teams miss in HubSpot is the secondary contact scoring property"	Medium-high — specific domain knowledge
Visual data	Original charts, screenshots, diagrams with text descriptions	Medium — adds depth

Originality rules

Every page must contain at least one thing AI couldn't have written. Proprietary data, a specific customer example, an expert opinion, or a contrarian take
If you have no original insight to add, don't publish. AI-only content with no unique value adds nothing to the web and won't rank in either Google or AI search
Proprietary data is the ultimate differentiator. AI can't fabricate your customer data, your benchmark data, or your usage data. This is what gets cited

The AI Content Quality Scorecard

Score every AI-edited piece before publishing:

Criterion	0 (Fail)	1 (Acceptable)	2 (Strong)
Accuracy	Unchecked claims or known errors	All claims verified, sources cited	Verified with primary sources, links included
Specificity	Generic statements dominate	Mix of generic and specific	Every claim includes a number, name, or detail
Voice	Reads like unedited AI	Some AI markers remain but mostly brand voice	Distinct voice, no AI markers
Structure	No AEO formatting	Basic structure (H2s, some tables)	Full AEO compliance (answer-first, tables, schema)
Originality	No unique content — AI could've written it all	1-2 original insights added	Proprietary data, expert POV, or contrarian take that defines the piece

Score 8-10: Publish. 5-7: More editing needed. Below 5: Rewrite or don't publish.

Pre-Publish Checklist

Before publishing any AI-edited content:

[ ] All statistics verified against original sources
[ ] All product names, features, and pricing confirmed current
[ ] No unsourced "studies show" or "according to research" claims
[ ] Generic statements replaced with specific facts
[ ] AI voice markers eliminated (hedge words, filler intros, false balance)
[ ] Word count cut by 20-30% from AI original
[ ] At least one proprietary data point, expert insight, or contrarian take added
[ ] AEO structure applied (answer-first, question H2s, tables, schema)
[ ] Author byline added with real name and credentials
[ ] Page scored 8+ on the quality scorecard
[ ] Content reads as if a knowledgeable human wrote it, not an AI

Anti-Pattern Check

Publishing AI content with only grammar edits → Grammar is the one thing AI gets right. The problems are accuracy, specificity, voice, and originality. A grammar-only edit misses everything that matters
Adding "written by AI, edited by humans" disclaimers → This signals low quality to readers and search engines. Instead, invest the editing time to make the content genuinely good. Let quality speak for itself
Using AI to edit AI → Running AI output through another AI pass smooths prose but doesn't add accuracy, specificity, or originality. Human expertise is the irreplaceable ingredient
Skipping the accuracy pass → A single hallucinated statistic published under your brand can destroy credibility. Accuracy checking is non-negotiable on every piece
Publishing because the AI wrote it fast → Speed doesn't justify low quality. If a piece doesn't score 8+ after editing, it shouldn't be published. A thin, generic page hurts your brand more than no page at all
No original insight in any piece → If every piece on your site could have been written by anyone's AI, you have no content moat. Prioritize adding proprietary data and expert POV in every piece

Want agents that use skill files like this?

We customize skill files for your brand voice and methodology, then run content agents against them.

Book a call

# AI Content Editing

The goal is not to remove every trace of AI. The goal is to add what AI cannot: proprietary data, expert judgment, specific examples from real experience, and a distinct point of view.

## The AI Content Editing Framework

Edit AI content in five passes. Each pass targets a different layer of quality.

| Pass | Focus | What you're fixing | Time |
|------|-------|--------------------|----- |
| 1. Accuracy | Facts, claims, numbers | Wrong information, hallucinations, outdated data | 10-15 min |
| 2. Specificity | Vague → specific | Generic statements that could apply to any company | 10-15 min |
| 3. Voice | Tone and style | AI-default voice → brand voice or expert voice | 10-15 min |
| 4. Structure | AEO readiness | Format for AI search extraction | 5-10 min |
| 5. Originality | Unique value | Add proprietary data, POV, or insider knowledge | 15-30 min |

**Total: 50-85 minutes per 1,500-word piece.** This is the minimum investment to transform AI output into publishable content.

---

## Pass 1: Accuracy Check

AI hallucinations publish false information under your brand. This is the most important pass.

### What to verify

| Claim type | How to verify | Common AI errors |
|-----------|---------------|-----------------|
| Statistics and numbers | Check original source | AI invents plausible-sounding statistics. "Studies show 73% of..." with no study |
| Product features | Check product website/docs | AI describes features that don't exist or describes them incorrectly |
| Pricing | Check current pricing page | AI uses outdated pricing or makes up price points |
| Company information | Check company website | AI confuses companies, invents founding dates, misidentifies founders |
| Process steps | Verify against documentation | AI invents plausible but incorrect process steps |
| Tool integrations | Check integration docs | AI claims integrations exist when they don't |
| Historical claims | Check primary sources | AI invents historical events or attributes quotes incorrectly |

### Accuracy rules

- **Never publish a statistic you can't source.** If the AI claims "67% of sales teams do X" and you can't find the source, delete the statistic. Replace with a factual claim you can verify
- **Delete every unsourced "according to" and "studies show."** If no study is cited, the AI invented it
- **Check every product name, feature name, and company name.** AI frequently misspells product names or confuses similar products
- **Verify all URLs.** AI generates plausible-looking URLs that don't exist

---

## Pass 2: Specificity Edit

AI defaults to generic, safe statements. Specific content outperforms generic content in both SEO and AEO.

### The specificity test

For every sentence, ask: "Could this sentence appear on any competitor's website unchanged?" If yes, it's too generic. Rewrite.

| Generic (AI default) | Specific (edited) |
|---------------------|-------------------|
| "Many companies struggle with this challenge." | "B2B SaaS teams with 5-20 SDRs typically hit this problem when monthly lead volume exceeds 500." |
| "This can lead to significant improvements." | "Teams using this approach see 23% higher reply rates within the first 30 days, based on data from 200+ sequences." |
| "There are several tools available." | "The three most common tools are Apollo ($49/mo), ZoomInfo ($14,995/yr), and Clearbit (usage-based pricing starting at $99/mo)." |
| "Best practices suggest doing this regularly." | "Update comparison pages monthly. Pricing pages within 24 hours of any price change." |

### Specificity rules

- **Replace every "many" with a number.** "Many companies" → "67% of B2B SaaS companies" or "companies with 50-200 employees"
- **Replace every "significant" with a measurement.** "Significant improvement" → "23% increase" or "from 3 days to 4 hours"
- **Name tools, companies, and people.** "Popular CRM" → "HubSpot, Salesforce, or Pipedrive"
- **Add time frames.** "Regularly" → "monthly" or "every 2 weeks"
- **Add qualifying context.** "Small businesses" → "B2B SaaS companies with $1-5M ARR and 10-50 employees"

---

## Pass 3: Voice Edit

AI has a recognizable voice: formal, hedged, overly balanced, and devoid of personality. Edit for distinctiveness.

### AI voice markers to eliminate

| AI marker | Example | Fix |
|-----------|---------|-----|
| Hedge words | "It's important to note that...", "It's worth considering..." | Delete. Start with the actual point |
| False balance | "Both approaches have their merits..." | Take a position. "Use X when [condition]. Use Y otherwise." |
| Filler introductions | "In today's rapidly evolving landscape..." | Cut entirely. Start with the first substantive sentence |
| Passive voice default | "It has been observed that..." | Active voice: "We see..." or "[Subject] does..." |
| Em-dash overuse | "This approach — which many teams find effective — can..." | Restructure to two sentences or use parentheses sparingly |
| Excessive connectors | "Moreover, furthermore, additionally, it's also important..." | Cut. Each paragraph stands on its own |
| Summary conclusions | "In conclusion, we've covered..." | Cut. The content speaks for itself |
| Rhetorical questions | "But what makes this so important?" | Delete. State the answer directly |
| Superlatives | "One of the most powerful ways to..." | Delete "one of." Use concrete evidence instead |

### Brand voice calibration

| Dimension | AI default | B2B SaaS target |
|-----------|-----------|-----------------|
| Formality | Formal, academic | Conversational but competent. Peer-to-peer |
| Confidence | Hedged, cautious | Direct, opinionated. "Do X" not "Consider X" |
| Length | Verbose, padded | Terse. Cut 30% of AI word count |
| Perspective | Third-person, detached | First-person plural ("we've seen") or second-person ("you should") |
| Personality | None | Distinct POV, occasional strong opinions |

---

## Pass 4: Structure Edit (AEO)

Format the edited content for AI search extraction.

### Structural checks

- [ ] First 50 words contain a direct answer to the page's primary question
- [ ] All H2s are question-shaped
- [ ] Every comparison is in a table, not prose
- [ ] Every paragraph leads with its key claim (inverted pyramid)
- [ ] No paragraph exceeds 4 sentences
- [ ] Lists are formatted as numbered or bulleted lists, not inline prose
- [ ] `FAQPage` schema applied if page has Q&A sections

---

## Pass 5: Originality Edit

This is the pass that transforms AI content from "competent" to "worth citing." Add what AI cannot generate on its own.

### What to add

| Addition type | Example | Impact |
|--------------|---------|--------|
| Proprietary data | "Across our 500+ customers, we see average reply rates of 3.1% for cold email" | Very high — unique, citable data |
| Expert judgment | "In our experience, teams that skip the ICP step waste 40% of their first quarter" | High — human expertise AI can't replicate |
| Named case examples | "When Acme Corp switched to this approach, they cut ramp time from 90 to 45 days" | High — specific, verifiable |
| Contrarian takes | "Most guides tell you to personalize every email. That's wrong at scale. Here's why:" | High — distinctive POV |
| Insider knowledge | "The feature most teams miss in HubSpot is the secondary contact scoring property" | Medium-high — specific domain knowledge |
| Visual data | Original charts, screenshots, diagrams with text descriptions | Medium — adds depth |

### Originality rules

- **Every page must contain at least one thing AI couldn't have written.** Proprietary data, a specific customer example, an expert opinion, or a contrarian take
- **If you have no original insight to add, don't publish.** AI-only content with no unique value adds nothing to the web and won't rank in either Google or AI search
- **Proprietary data is the ultimate differentiator.** AI can't fabricate your customer data, your benchmark data, or your usage data. This is what gets cited

---

## The AI Content Quality Scorecard

Score every AI-edited piece before publishing:

| Criterion | 0 (Fail) | 1 (Acceptable) | 2 (Strong) |
|-----------|----------|----------------|------------|
| Accuracy | Unchecked claims or known errors | All claims verified, sources cited | Verified with primary sources, links included |
| Specificity | Generic statements dominate | Mix of generic and specific | Every claim includes a number, name, or detail |
| Voice | Reads like unedited AI | Some AI markers remain but mostly brand voice | Distinct voice, no AI markers |
| Structure | No AEO formatting | Basic structure (H2s, some tables) | Full AEO compliance (answer-first, tables, schema) |
| Originality | No unique content — AI could've written it all | 1-2 original insights added | Proprietary data, expert POV, or contrarian take that defines the piece |

**Score 8-10:** Publish. **5-7:** More editing needed. **Below 5:** Rewrite or don't publish.

---

## Pre-Publish Checklist

Before publishing any AI-edited content:

- [ ] All statistics verified against original sources
- [ ] All product names, features, and pricing confirmed current
- [ ] No unsourced "studies show" or "according to research" claims
- [ ] Generic statements replaced with specific facts
- [ ] AI voice markers eliminated (hedge words, filler intros, false balance)
- [ ] Word count cut by 20-30% from AI original
- [ ] At least one proprietary data point, expert insight, or contrarian take added
- [ ] AEO structure applied (answer-first, question H2s, tables, schema)
- [ ] Author byline added with real name and credentials
- [ ] Page scored 8+ on the quality scorecard
- [ ] Content reads as if a knowledgeable human wrote it, not an AI

---

## Anti-Pattern Check

- Publishing AI content with only grammar edits → Grammar is the one thing AI gets right. The problems are accuracy, specificity, voice, and originality. A grammar-only edit misses everything that matters
- Adding "written by AI, edited by humans" disclaimers → This signals low quality to readers and search engines. Instead, invest the editing time to make the content genuinely good. Let quality speak for itself
- Using AI to edit AI → Running AI output through another AI pass smooths prose but doesn't add accuracy, specificity, or originality. Human expertise is the irreplaceable ingredient
- Skipping the accuracy pass → A single hallucinated statistic published under your brand can destroy credibility. Accuracy checking is non-negotiable on every piece
- Publishing because the AI wrote it fast → Speed doesn't justify low quality. If a piece doesn't score 8+ after editing, it shouldn't be published. A thin, generic page hurts your brand more than no page at all
- No original insight in any piece → If every piece on your site could have been written by anyone's AI, you have no content moat. Prioritize adding proprietary data and expert POV in every piece