The Perfect Prompt Formula for SEO Content (Step-by-Step)

SEO content prompts fail when they optimize for word count instead of search intent. Google rewards pages that answer the query completely — AI rewards prompts that define intent, structure, and trust signals before a single paragraph is written.

This is a step-by-step prompt formula — not one mega-prompt, but seven sequential prompts that mirror how strong human SEO editors work. Compatible with GPT-5.5, Claude Opus 4.8, and Gemini 3.1 Pro.

Pair this formula with the five-stage blog chain in our blog post templates article — this guide goes deeper on trust, snippets, and keyword placement per step.

The formula overview

1. Intent lock
2. Keyword + entity map
3. E-E-A-T guardrails
4. Competitive gap brief
5. Outline with snippet targets
6. Section draft (repeat per H2)
7. Optimization + human fact pass

Each step outputs input for the next. Skip a step, skip ranking potential.

Step 1 — Intent lock prompt

Before keywords, lock intent.

Primary keyword: [keyword]

SERP intent type: [informational / commercial / transactional / comparison]

Reader job-to-be-done: What must they be able to do after reading?

Misinterpretations to avoid: [e.g. not a product review, not beginner if SERP is advanced]

Output: 2-sentence intent statement + 3 must-answer questions.

Human check: open top 3 Google results — does intent statement match?

Claude Opus 4.8 nuance: paste 3 SERP titles and ask if your intent statement aligns or conflicts.

Step 2 — Keyword and entity map

Build semantic coverage without stuffing.

Using intent: [paste step 1]

Create a keyword map:

• Primary keyword (1)
• Secondary keywords (3-5)
• Related entities and concepts Google expects (people, tools, processes)
• Questions from People Also Ask (5)

Format as table: Term | Suggested placement (H2/intro/FAQ) | Notes

Do not assign density percentages — assign placement.

Example row: "structured output prompting" | H2 in section 4 | link to cluster 2 article

Gemini 3.1 Pro: paste 5 competitor page outlines and ask for entity gaps in one call.

Step 3 — E-E-A-T guardrails prompt

Trust rules before drafting.

Article topic: [topic]

Our credentials: [what we actually know/have tested]

Guardrails for the AI drafter:

• Mark any stat without source as [CITE NEEDED]
• Include at least 2 places for first-hand experience (I/we tested, our data shows)
• Do not invent case studies or client names
• Distinguish fact from recommendation
• YMYL check: [if health/finance/legal — stricter sourcing rules]

Output: bullet list of trust rules to prepend to every draft prompt in this project.

Positive framing: "Ground claims in provided sources" beats "don't hallucinate." See positive framing article.

Step 4 — Competitive gap brief

Prompt:

Top-ranking pages cover: [bullets from your SERP scan]

Gaps they miss: [your angles]

Our unique value: [data, methodology, contrarian take]

Produce a brief with: thesis, 3 differentiation bullets, sections competitors have that we must match, sections we add that they lack.

This is where human judgment is non-negotiable — AI cannot browse SERP for you reliably without tools.

Worked angle example: competitors list "10 prompting tips" generically; your gap is "model-class split — reasoning vs chat" backed by cluster 2 research.

Step 5 — Outline with snippet targets

Using brief: [paste]

Create outline:

• H1 (one option, keyword natural)
• H2s (6-8) — label each with snippet target: [paragraph / list / table / FAQ]
• FAQ (5-6) — question phrasing matched to PAA
• Suggested internal links: [placeholders]
• Meta description draft (155 chars)

Featured snippet rule: one H2 should be a direct 40-60 word answer block.

GPT-5.5 follows snippet labels literally — good for this step. Claude adds better H2 angle phrasing.

Step 6 — Section draft formula

Repeat per H2. Combine steps 2-3 guardrails into every call.

GUARDRAILS: [paste step 3 rules]

KEYWORD MAP row for this section: [paste relevant row]

Write H2: [title]

Snippet target: [list/table/paragraph]

Length: [word count]

Must include: [specific example or data you provide]

Answer-first structure. Short paragraphs. [CITE NEEDED] for unverified stats.

Use RTF structure from our framework article — Role, Task, Format, Context in each section call.

Step 7 — Optimization pass prompts

Keyword placement audit:

Review this draft. Confirm primary keyword [X] appears in: first 100 words, one H2, conclusion. Flag forced usage. Suggest 2 semantic variants to add naturally.

Readability pass:

Shorten sentences over 25 words. Split walls of text. Add one bulleted list if section exceeds 300 words.

Schema/FAQ check:

Are FAQ answers 50-80 words each, starting with direct answer? Revise if not.

Human fact pass (you, not AI): replace every [CITE NEEDED]; add screenshots, original data, author bio.

Extra templates for formula steps

Template N — YMYL intent lock (finance/health)

Primary keyword: [keyword]

YMYL category: [health/finance/legal]

Required: cite primary sources only; no treatment or investment advice — informational only.

Reader job-to-be-done: [specific, non-diagnostic]

Output: intent statement + 3 questions + list of claims that require licensed review.

Template O — Comparison article keyword map

Products: [A] vs [B] vs [C]

Intent: commercial comparison

Output table: Term | A mention | B mention | C mention | H2 placement

Include: pricing, integrations, support, limits — mark UNVERIFIED where you lack data.

Template P — Snippet-optimized definition block

Write a 45-55 word definition of [term] for featured snippet.

Start with "[Term] is..." Answer the question directly. No preamble. Suitable for H2: "What is [term]?"

CONTEXT: [paste step 1 intent]

Template Q — Optimization pass (GPT-5.5 batch)

Review full draft below. Return:

1. Keyword placement score (pass/fail per rule)
2. List of [CITE NEEDED] still present
3. Three sentences to cut as filler
4. Two internal link insertion points with anchor text suggestions

[paste draft]

Worked example: seven steps for "lost in the middle"

Step 1: Intent = informational; reader must fix long-prompt failures in production.

Step 2: Map "prompt length", "U-shaped attention", "bookend instructions" to H2s.

Step 3: Guardrails: cite Liu et al.; mark unverified model claims [CITE NEEDED].

Step 4: Gap = 2026 context window myth vs measured U-curve.

Step 5: Outline with list snippet for "symptoms" H2, paragraph snippet for definition H2.

Step 6: Seven section drafts via Claude; one H2 per call with guardrails prepended.

Step 7: GPT-5.5 keyword audit; human replaced two stats and added PromptMake screenshot.

Total production time: ~3 hours with formula vs ~6+ hours rewriting one-shot drafts. The formula cost is front-loaded in steps 1–4.

Before / after: formula vs one-shot

Before

"Write an SEO article about prompt engineering with good keywords and make it rank on Google."

Result: generic tips, no SERP alignment, invented statistics, no snippet structure.

After (step 6 only, with prior steps pasted)

GUARDRAILS: [paste]. KEYWORD MAP: "positive framing" → this H2. Write H2: "Why positive constraints beat negative rules." Snippet: list. Length: 400 words. Must include: before/after example from our testing.

Result: on-brief section ready for optimization pass.

Model notes by step

| Step | GPT-5.5 | Claude Opus 4.8 | Gemini 3.1 Pro | Gemini 3.5 Flash |

| 1 Intent | Good | Excellent nuance | Good with SERP paste | Fast draft |

| 2 Keyword map | Excellent tables | Good | Excellent multi-competitor | Good |

| 3 E-E-A-T rules | Good | Good | Good | Adequate |

| 4 Gap brief | Good | Best prose thesis | Best many-source | Skip for depth |

| 5 Outline | Best format | Best angles | Good | Good |

| 6 Sections | Good + edit | Best voice | Good technical | Adequate |

| 7 Optimize | Best audit | Good readability | Good | Fast checklist |

Snippet target cheat sheet

| Snippet type | H2 pattern | Prompt instruction |

| Paragraph | What is X? / Why does X? | 40-60 words, answer-first |

| List | How to X / X ways to | Numbered or bulleted, 4-8 items |

| Table | X vs Y / comparison | Markdown table, max 8 rows |

| FAQ | People Also Ask phrasing | 50-80 words, direct answer line 1 |

Label every H2 in step 5 — models follow labels better than implied structure.

One-page formula card

Intent → Keywords → Trust rules → Gap brief → Outline → Sections → Optimize → Human verify

Never: Intent → write 2000 words → hope

Print this card next to your stage prompts. Each arrow is a separate API call or chat turn.

Common failures

Skipping intent lock → article ranks for wrong query

No E-E-A-T guardrails → confident hallucinated stats

Optimization without human fact pass → AI slop with keywords

One-shot full article → generic structure

Keyword density targets instead of placement map

Snippet H2 written as vague topic label ("Overview") instead of question

Internal links omitted from outline — added as afterthought

Using same guardrails for YMYL and casual topics

Step 7 only on AI — human must own citations and experience

E-E-A-T signals to add in human pass (not AI)

Author byline with credentials

Original screenshots, data, or workflow photos

Date updated and methodology note

Links to primary sources replacing [CITE NEEDED]

First-person experience paragraphs AI marked as placeholders

Disclosure when AI assisted drafting — transparency supports trust on YMYL-adjacent topics

PromptMake workflow

/text → paste step 1 ingredients → chain through Improve for steps 5-6 → human owns steps 4 and 7.

3 free /text runs/day. Save step 3 guardrails as a reusable block for the whole content project.

Frequently asked questions

Can I combine steps to save time?

Steps 1+2 can merge for experienced editors. Never merge 3+6 — guardrails must precede every draft.

How is this different from the five-stage blog chain?

Five-stage = production workflow. Seven-step formula = SEO-specific depth (intent, entities, snippets, E-E-A-T). Use both.

Do I need different formulas for commercial vs informational?

Intent lock changes; steps 2-7 structure stays. Commercial needs comparison table snippet targets and UNVERIFIED pricing tags.

What's the minimum viable formula?

Steps 1, 3, 5, 6, 7. Skip keyword map only if topic is ultra-narrow and you know entities by heart.

How many section prompts for a 2,000-word post?

Typically 6-8 — one per H2 plus separate intro and conclusion prompts.

Should AI write meta description before or after draft?

Draft in step 5 outline; finalize in step 7 after keyword audit confirms phrasing.

Gemini for SEO formula?

3.1 Pro when step 4 includes many competitor pastes. GPT-5.5 for step 7 audit. Claude for step 6 prose.

How to handle [CITE NEEDED] at scale?

Export all tags with grep before publish. Batch-research or cut claims — never publish with tags intact.

Related articles

AI prompts for blog posts 2026 — full chain templates

RTF framework — section prompt skeleton

Positive framing — trust rules as positive constraints

Lost in the middle — keep step inputs focused; bookend long pastes

Few-shot vs zero-shot — voice in step 6

Structured output — if publishing FAQ schema from JSON

Prompt engineering best practices 2026 — model pick per step

Claude vs ChatGPT vs Gemini — dialect per formula step

Bottom line

The perfect SEO prompt formula is seven steps: lock intent, map keywords, set trust rules, find gaps, outline for snippets, draft by section, optimize then verify.

AI handles structure and speed. You handle SERP judgment and facts. That split is what makes SEO content worth publishing in 2026.