Home/ Skills/ pseo-ai-content-enrichment

general pseo-ai-content-enrichment

pseo-ai-content-enrichment

This skill should be used when the user asks to "enrich pSEO with AI", "use AI for programmatic content", "AI content for pSEO pages", "enrich programmatic pages with AI", "generate pSEO content with AI", "AI-assisted pSEO", "add AI content to programmatic pages", "scale content enrichment with AI", or any variation of using AI to enrich, generate, or improve content for programmatic SEO pages at scale.

Download .md

pSEO AI Content Enrichment

AI content enrichment is the process of using LLMs to add unique, valuable content to programmatic SEO pages at scale. Raw data produces thin pages. AI enrichment transforms structured data into pages with unique descriptions, per-entry analysis, specific FAQ sections, and contextual content — without the cost of manually writing 500 pages.

The key distinction: AI is the production tool, not the data source. The data must come from verified sources. AI transforms that data into readable, structured content.

The AI Enrichment Pipeline

Stage	Input	AI action	Output	Human check
1. Description generation	Entity name + data fields	Generate a 100-200 word unique description per entry	Per-entry description	20% sample review
2. Comparison context	Entity data + competitor data	Generate "how it compares" section per entry	Per-entry comparison paragraph	20% sample review
3. Pros/cons generation	Entity features + reviews	Generate 3-5 pros and 2-3 cons per entry	Per-entry pros/cons list	20% sample review
4. FAQ generation	Entity data + PAA research	Generate 3-5 entry-specific FAQ Q&A pairs	Per-entry FAQ section	20% sample review
5. Expert summary	All generated content + data	Generate a 2-3 sentence "our take" per entry	Per-entry expert callout	50% review (higher stakes)

Prompt Engineering for pSEO

The master prompt structure

Every AI enrichment task needs a master prompt that produces consistent, high-quality output across hundreds of entries.

Prompt template:

You are writing content for a [page type] about [entity type].

Context:
- Entity name: {name}
- Category: {category}
- Key data: {data_field_1}, {data_field_2}, {data_field_3}

Task: Write a [content element] for this specific entity.

Rules:
- [Word count constraint]
- [Specificity requirement: must reference the entity's specific data]
- [Banned phrases list]
- [AEO formatting requirement]
- [Uniqueness requirement: must differ from generic category description]

Output format:
[Exact format specification]

Prompt rules

Rule	Why	Example
Include entity-specific data in the prompt	Forces AI to use real data, not generic text	"Pricing: $49/month. Integrations: Salesforce, HubSpot, Slack."
Set word count constraints	Prevents verbosity and ensures consistency	"Write exactly 100-150 words."
Ban generic phrases	Prevents AI from producing interchangeable descriptions	"Never use: 'comprehensive solution', 'powerful platform', 'cutting-edge'"
Require specific comparisons	Forces differentiation	"Compare to [top 2 competitors by name] on at least one dimension"
Specify the output format	Ensures consistent structure	"Output as: 3-5 bullet points, each starting with a bold label"
Include audience context	Focuses the content	"Written for a B2B SaaS buyer evaluating CRM tools for a 50-person team"

Example prompts by content element

Per-entry description:

Write a 100-150 word description of {tool_name} for a B2B buyer evaluating {category} tools.

Data: Pricing starts at {price}. Key features: {features}. Best for: {best_for}. Integrations: {integrations}.

Rules:
- First sentence must define what {tool_name} does in under 25 words
- Include at least one specific number (pricing, user count, or metric)
- Compare to one named competitor on one dimension
- Never use: "comprehensive", "cutting-edge", "powerful", "seamless"
- Write in present tense, second person

Per-entry FAQ:

Generate 4 FAQ questions and answers about {tool_name} specifically.

Data: {all_data_fields}

Rules:
- Questions must be specific to {tool_name}, not generic category questions
- At least one question about pricing, one about integrations, one about limitations
- Each answer: 2-3 sentences, includes a specific fact from the data
- Never answer with "Contact sales" or "It depends"

Quality Control at Scale

The 20% review rule

Human-review at least 20% of AI-generated content before publishing any batch.

Review focus	What to check	Fail criteria
Accuracy	Do the facts match the source data?	Any factual error = fail
Specificity	Is the content specific to this entry or generic?	Could apply to 5+ other entries unchanged = fail
Uniqueness	Is this content different from other pages?	> 70% similar to another generated page = fail
Readability	Does it read naturally?	Obvious AI patterns (hedge words, filler, em-dashes) = edit
AEO compliance	Answer in first sentence? Extractable claims?	No direct answer = edit

Batch quality workflow

Step	Action	Decision
1	Generate content for entire batch (100-500 entries)	—
2	Automated checks: word count, uniqueness, banned phrases	Auto-reject failures
3	Human review 20% random sample	If > 10% of sample fails → fix prompt, regenerate entire batch
4	Fix individual failures	—
5	Publish in staggered batches	—
6	Monitor indexation and quality signals post-publish	If indexation < 80% → diagnose and fix

What AI Should NOT Do in pSEO

AI should NOT	Why	What to do instead
Invent data (pricing, features, stats)	Hallucinated facts damage credibility	Feed verified data to the prompt. AI formats it, doesn't create it
Write the entire page from scratch	Produces generic content without unique data	AI enriches a data-driven template, it doesn't replace the data
Generate content without entity-specific data	Output will be generic and interchangeable	Always include entity-specific data fields in every prompt
Self-assess quality	AI can't reliably judge its own output quality	Human review is non-negotiable
Replace expert judgment	AI can't provide genuine expert opinions	Label AI-generated takes as overview, not expert opinion

Pre-Enrichment Checklist

[ ] Verified data source built with all required fields per entry
[ ] Master prompts written and tested on 10 sample entries
[ ] Prompt includes entity-specific data (not just name)
[ ] Banned phrase list defined and included in prompts
[ ] Word count constraints set for each content element
[ ] AEO formatting requirements included in prompts
[ ] Automated quality checks built (word count, similarity, banned phrases)
[ ] 20% human review process defined with fail criteria
[ ] Batch regeneration process ready (if sample fails > 10%)
[ ] Post-publish monitoring plan set (indexation, quality signals)

Anti-Pattern Check

Prompting AI with just the entity name → "Write about HubSpot" produces generic content. "Write about HubSpot given: pricing $20/seat, 150K customers, integrates with Salesforce/Slack/Gmail, best for SMBs" produces specific, useful content. Always include data
No banned phrase list → AI defaults to "comprehensive", "powerful", "cutting-edge", "seamless" across every page. Ban these and 20+ similar words to force specificity
Skipping human review → AI hallucinates facts, produces near-duplicates, and occasionally generates nonsense. The 20% review catches systemic issues before 500 bad pages go live
Using the same prompt for every content element → Description prompts, FAQ prompts, and comparison prompts need different structures. One master prompt for everything produces inconsistent quality
AI generates data it wasn't given → If the prompt doesn't include pricing and the AI outputs "$49/month" — that's a hallucination. Instruct AI to only use data provided in the prompt. Never invent facts
No similarity check across generated content → AI may produce very similar descriptions for similar entities. Run pairwise similarity checks. If two pages are > 70% similar, regenerate one with a differentiation-focused prompt

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

# pSEO AI Content Enrichment

The key distinction: AI is the production tool, not the data source. The data must come from verified sources. AI transforms that data into readable, structured content.

## The AI Enrichment Pipeline

| Stage | Input | AI action | Output | Human check |
|-------|-------|-----------|--------|-------------|
| 1. Description generation | Entity name + data fields | Generate a 100-200 word unique description per entry | Per-entry description | 20% sample review |
| 2. Comparison context | Entity data + competitor data | Generate "how it compares" section per entry | Per-entry comparison paragraph | 20% sample review |
| 3. Pros/cons generation | Entity features + reviews | Generate 3-5 pros and 2-3 cons per entry | Per-entry pros/cons list | 20% sample review |
| 4. FAQ generation | Entity data + PAA research | Generate 3-5 entry-specific FAQ Q&A pairs | Per-entry FAQ section | 20% sample review |
| 5. Expert summary | All generated content + data | Generate a 2-3 sentence "our take" per entry | Per-entry expert callout | 50% review (higher stakes) |

---

## Prompt Engineering for pSEO

### The master prompt structure

Every AI enrichment task needs a master prompt that produces consistent, high-quality output across hundreds of entries.

**Prompt template:**
```
You are writing content for a [page type] about [entity type].

Context:
- Entity name: {name}
- Category: {category}
- Key data: {data_field_1}, {data_field_2}, {data_field_3}

Task: Write a [content element] for this specific entity.

Rules:
- [Word count constraint]
- [Specificity requirement: must reference the entity's specific data]
- [Banned phrases list]
- [AEO formatting requirement]
- [Uniqueness requirement: must differ from generic category description]

Output format:
[Exact format specification]
```

### Prompt rules

| Rule | Why | Example |
|------|-----|---------|
| Include entity-specific data in the prompt | Forces AI to use real data, not generic text | "Pricing: $49/month. Integrations: Salesforce, HubSpot, Slack." |
| Set word count constraints | Prevents verbosity and ensures consistency | "Write exactly 100-150 words." |
| Ban generic phrases | Prevents AI from producing interchangeable descriptions | "Never use: 'comprehensive solution', 'powerful platform', 'cutting-edge'" |
| Require specific comparisons | Forces differentiation | "Compare to [top 2 competitors by name] on at least one dimension" |
| Specify the output format | Ensures consistent structure | "Output as: 3-5 bullet points, each starting with a bold label" |
| Include audience context | Focuses the content | "Written for a B2B SaaS buyer evaluating CRM tools for a 50-person team" |

### Example prompts by content element

**Per-entry description:**
```
Write a 100-150 word description of {tool_name} for a B2B buyer evaluating {category} tools.

Data: Pricing starts at {price}. Key features: {features}. Best for: {best_for}. Integrations: {integrations}.

Rules:
- First sentence must define what {tool_name} does in under 25 words
- Include at least one specific number (pricing, user count, or metric)
- Compare to one named competitor on one dimension
- Never use: "comprehensive", "cutting-edge", "powerful", "seamless"
- Write in present tense, second person
```

**Per-entry FAQ:**
```
Generate 4 FAQ questions and answers about {tool_name} specifically.

Data: {all_data_fields}

Rules:
- Questions must be specific to {tool_name}, not generic category questions
- At least one question about pricing, one about integrations, one about limitations
- Each answer: 2-3 sentences, includes a specific fact from the data
- Never answer with "Contact sales" or "It depends"
```

---

## Quality Control at Scale

### The 20% review rule

Human-review at least 20% of AI-generated content before publishing any batch.

| Review focus | What to check | Fail criteria |
|-------------|---------------|--------------|
| Accuracy | Do the facts match the source data? | Any factual error = fail |
| Specificity | Is the content specific to this entry or generic? | Could apply to 5+ other entries unchanged = fail |
| Uniqueness | Is this content different from other pages? | > 70% similar to another generated page = fail |
| Readability | Does it read naturally? | Obvious AI patterns (hedge words, filler, em-dashes) = edit |
| AEO compliance | Answer in first sentence? Extractable claims? | No direct answer = edit |

### Batch quality workflow

| Step | Action | Decision |
|------|--------|----------|
| 1 | Generate content for entire batch (100-500 entries) | — |
| 2 | Automated checks: word count, uniqueness, banned phrases | Auto-reject failures |
| 3 | Human review 20% random sample | If > 10% of sample fails → fix prompt, regenerate entire batch |
| 4 | Fix individual failures | — |
| 5 | Publish in staggered batches | — |
| 6 | Monitor indexation and quality signals post-publish | If indexation < 80% → diagnose and fix |

---

## What AI Should NOT Do in pSEO

| AI should NOT | Why | What to do instead |
|--------------|-----|-------------------|
| Invent data (pricing, features, stats) | Hallucinated facts damage credibility | Feed verified data to the prompt. AI formats it, doesn't create it |
| Write the entire page from scratch | Produces generic content without unique data | AI enriches a data-driven template, it doesn't replace the data |
| Generate content without entity-specific data | Output will be generic and interchangeable | Always include entity-specific data fields in every prompt |
| Self-assess quality | AI can't reliably judge its own output quality | Human review is non-negotiable |
| Replace expert judgment | AI can't provide genuine expert opinions | Label AI-generated takes as overview, not expert opinion |

---

## Pre-Enrichment Checklist

- [ ] Verified data source built with all required fields per entry
- [ ] Master prompts written and tested on 10 sample entries
- [ ] Prompt includes entity-specific data (not just name)
- [ ] Banned phrase list defined and included in prompts
- [ ] Word count constraints set for each content element
- [ ] AEO formatting requirements included in prompts
- [ ] Automated quality checks built (word count, similarity, banned phrases)
- [ ] 20% human review process defined with fail criteria
- [ ] Batch regeneration process ready (if sample fails > 10%)
- [ ] Post-publish monitoring plan set (indexation, quality signals)

---

## Anti-Pattern Check

- Prompting AI with just the entity name → "Write about HubSpot" produces generic content. "Write about HubSpot given: pricing $20/seat, 150K customers, integrates with Salesforce/Slack/Gmail, best for SMBs" produces specific, useful content. Always include data
- No banned phrase list → AI defaults to "comprehensive", "powerful", "cutting-edge", "seamless" across every page. Ban these and 20+ similar words to force specificity
- Skipping human review → AI hallucinates facts, produces near-duplicates, and occasionally generates nonsense. The 20% review catches systemic issues before 500 bad pages go live
- Using the same prompt for every content element → Description prompts, FAQ prompts, and comparison prompts need different structures. One master prompt for everything produces inconsistent quality
- AI generates data it wasn't given → If the prompt doesn't include pricing and the AI outputs "$49/month" — that's a hallucination. Instruct AI to only use data provided in the prompt. Never invent facts
- No similarity check across generated content → AI may produce very similar descriptions for similar entities. Run pairwise similarity checks. If two pages are > 70% similar, regenerate one with a differentiation-focused prompt