SEO Without an Agency in 2026: 12 Prompts That Build a Semantic Core, Intent Clusters, and a Content Plan

SEO Without an Agency in 2026: 12 Prompts That Build a Semantic Core, Intent Clusters, and a Content Plan

In 2026 the bottleneck in SEO is no longer tooling — Semrush, Ahrefs, and Google Keyword Planner are all still there. The bottleneck is the manual labor between the tools: cleaning thousands of keyword rows, grouping them by intent, writing briefs, mapping internal links. That work used to justify a retainer. Now a single specialist with a well-structured prompt library can run the same pipeline in an afternoon.

Below are 12 prompts I use for the full cycle: harvesting semantics, clustering intents, building a content plan, generating metadata, wiring internal links, and auditing existing pages. Every prompt follows the same shape — task → prompt → example → metric. They are model-agnostic: ChatGPT, Claude, and Gemini all handle them; where reasoning quality matters (clustering, cannibalization), prefer a model with a long context window.

A note on trustworthy data: never ask an LLM to invent search volumes. Feed it a CSV export from a real source (Google Keyword Planner, Ahrefs, Semrush) and let the model do the reasoning, not the data generation. That single rule prevents most hallucinated SEO advice.

Stage 1: Harvesting the Semantic Core

Prompt 1 — Seed expansion. Task: turn one seed keyword into a broad candidate list before you touch a paid tool. Prompt: "Act as an SEO strategist. For the seed '[topic]', produce 60 candidate keywords grouped by modifier type: informational (how, what, why), commercial (buy, price, cost), comparison (vs, best, alternative), and local (near me, in [city]). Output as a markdown table with columns: keyword, modifier type, estimated intent, language." Example input: topic = CRM for freelancers. Output includes rows like `crm for freelancers pricing

| commercial | transactional`. Metric: keyword coverage — this is your research backlog, not final data.

Prompt 2 — Cleaning a raw export. Task: deduplicate and filter a Keyword Planner dump. Prompt: "Here is a raw keyword export. Remove duplicates, drop branded terms matching [brand list], drop keywords with fewer than [N] monthly searches, and normalize plural/singular variants to the higher-volume form. Return a clean CSV with columns: keyword, volume, competition, normalized_form." Example: a 4,000-row export usually shrinks to 800–1,200 usable rows. Metric: research efficiency — fewer wasted briefs.

Prompt 3 — Question mining. Task: surface People Also Ask-style queries. Prompt: "Generate 40 natural-language questions a user might type before buying [product] or solving [problem]. Group by funnel stage: awareness, consideration, decision. Mark which ones are answerable in a featured snippet." Metric: traffic — question pages win snippets and long-tail impressions.

Stage 2: Clustering by Intent, Not by String

Prompt 4 — Intent clustering. Task: group keywords so one page can rank for a cluster. Prompt: "Cluster the following keywords by search intent, not by lexical similarity. Two keywords belong to the same cluster only if a single page could satisfy both. Return clusters with: cluster name, primary keyword (highest volume), supporting keywords, suggested URL slug, page type (blog/product/landing)." Example: buy crm software and crm software pricing often merge; what is crm stays separate. Metric: positions — one strong page beats three cannibalizing ones.

Prompt 5 — Cannibalization check. Task: detect pages competing for the same query. Prompt: "Given this list of existing URLs and their target keywords, flag any cluster where two or more URLs target overlapping intent. For each conflict, recommend: merge, redirect, or re-target, with reasoning." Metric: rankings — cannibalization is a silent rank killer.

Prompt 6 — Topic authority map. Task: build topical clusters around a pillar. Prompt: "Build a hub-and-spoke map: one pillar page on '[head topic]' and 8–12 supporting articles that link back to it. For each spoke, give the target intent, primary keyword, and the anchor text the pillar should use to link to it." Metric: domain authority signals — topical completeness.

Stage 3: The Content Plan

Prompt 7 — Prioritized roadmap. Task: order clusters by effort vs impact. Prompt: "Score each cluster on: business value (1–5), ranking difficulty (1–5, using the difficulty column provided), and content effort (1–5). Compute a priority score and output a 90-day publishing calendar with one cluster per slot." Metric: traffic — sequencing beats volume.

Prompt 8 — Brief generation. Task: turn a cluster into a writer-ready brief. Prompt: "Write an SEO brief for the cluster '[name]'. Include: target keyword, secondary keywords, search intent, recommended H2/H3 outline, questions to answer, internal links to include, and a 150-character meta description." Metric: CTR and time-on-page.

Stage 4: On-Page and Internal Linking

Prompt 9 — Title and meta tags. Task: generate variants under length limits. Prompt: "For URL [url] targeting '[keyword]', write 5 title tags (≤60 characters) and 3 meta descriptions (≤155 characters). Front-load the keyword, include a benefit, avoid clickbait." Metric: CTR.

Prompt 10 — Schema markup. Task: generate valid structured data. Prompt: "Produce JSON-LD for a [Article/FAQPage/Product] page with the following fields. Validate against schema.org specifications and flag any required property I omitted." Example output uses @context: https://schema.org and @type: FAQPage. Metric: rich results, CTR.

Prompt 11 — Internal link suggestions. Task: find link opportunities across a site. Prompt: "Given these 30 page titles and their target keywords, suggest internal links: source page → target page, with anchor text. Prioritize links that pass relevance from high-authority pages to new ones." Metric: crawl depth and rankings for new pages.

Stage 5: Auditing What You Already Have

Prompt 12 — Content decay audit. Task: identify pages losing traffic. Prompt: "Here is a list of URLs with their traffic for the last 6 months. Flag pages with a declining trend, classify the likely cause (outdated data, thin content, lost backlinks, intent shift), and recommend refresh, merge, or delete." Metric: recovered traffic.

What Actually Moves the Needle

Prompts don't replace judgment. The three rules that keep this workflow honest: feed real data, never let the model invent volumes, and validate every cluster against an actual SERP before publishing. A prompt library is leverage — the strategist is still you.

Start with Prompts 1, 4, and 7 this week. That triad alone covers the gap most solo SEOs feel: turning a messy keyword dump into a publishable plan. Build the rest as you go.