ChatGPT SEO prompts are the structured queries that turn an LLM into a useful research and drafting partner for SEO work — without letting the model decide what your content argues. The prompts below are what PNP uses internally for B2B SaaS clients in 2026. Each one comes with a goal, the literal prompt text, the expected output shape, and when to use it.
This is a cluster support under the MOFU pillar on how to rank in AI Overviews. For the strategic context, see the pillar; for the broader workflow, see our piece on using ChatGPT for SEO.
The shape of a good SEO prompt
Three properties define a good SEO prompt. First, specific output format — table, list, JSON, paragraph with constraints. Vague format produces vague output. Second, named constraints — word counts, banned phrasing, required keywords. Constraints force the model to make trade-offs visible. Third, an evaluation criterion — tell the model what "good" looks like for this task. Without it, the model defaults to plausible-sounding output that may not be useful.
Prompt 1 — Cluster keywords by buyer-journey stage
Goal: turn a flat list of keywords into a structured topic map.
Prompt: "Cluster these 200 keywords by buyer-journey stage (TOFU, MOFU, BOFU). Output as a table with columns: keyword, primary intent (Informational, Commercial, Transactional), stage, one-sentence rationale. If a keyword is ambiguous, pick the most likely stage and flag it with an asterisk."
Expected output: a structured table you can drop into a spreadsheet. When to use: at the start of every cluster planning sprint, after you pull keywords from Ahrefs but before briefs get written.
Prompt 2 — SERP gap analysis on a target keyword
Goal: identify the angle missing from the existing SERP.
Prompt: "I will paste the top 10 SERP results for the keyword '{X}.' For each result, summarize the article's primary angle in one sentence. Then identify two angles that no article in the top 10 takes but that a buyer searching this keyword would value. Cite the source URL for each angle you summarize."
Expected output: 10 summaries plus 2 gaps. When to use: before writing the brief for any TOFU or MOFU pillar.
Prompt 3 — Buyer-language extraction
Goal: surface the exact phrases customers use to describe the problem your product solves.
Prompt: "I will paste 30 sales call transcripts. Extract the top 20 phrases customers use to describe {specific pain point}. Cluster the phrases by similarity. For each cluster, output the most common phrasing and the frequency count. Ignore phrases the salesperson said."
Expected output: a phrase corpus organized by frequency. When to use: before rewriting page headers, FAQs, or opening paragraphs anywhere your customer language matters.
Prompt 4 — Draft-versus-brief evaluation
Goal: identify gaps between what the brief promised and what the draft delivered.
Prompt: "I will paste a brief and then a draft. Read both. List every claim, framework, or proprietary stat the brief promised that the draft does not include or weakens. For each gap, suggest a one-sentence fix. Do not rewrite the draft — just identify the gaps."
Expected output: a punch list. When to use: as a final-pass review before the senior editor sees the draft.
Prompt 5 — Tight paragraph rewriting
Goal: compress a paragraph without losing meaning.
Prompt: "Rewrite this paragraph to keep the same meaning but compress to {N} words. Use the verb '{X}' as the main action. Avoid the words 'leverage,' 'utilize,' and 'in order to.' Output only the rewritten paragraph, no commentary."
Expected output: a single rewritten paragraph. When to use: when an opening or transition paragraph is too long but the meaning must be preserved exactly.
Prompt 6 — H2 declarative-test
Goal: convert weak section headings into extractable claims.
Prompt: "For each of these H2 headings, evaluate whether it would make sense as a standalone claim if extracted from the article. If yes, mark it 'OK.' If no, rewrite it as a full declarative sentence or precise noun phrase that contains the section's primary claim. Output as a table: original, evaluation, rewrite if needed."
Expected output: a table evaluating each heading. When to use: after the outline is set but before drafting begins. The H2s decide which sections get cited; getting them right at outline stage saves a rewrite later.
Prompt 7 — FAQ generation from People Also Ask
Goal: build the article's FAQ section from real query data.
Prompt: "Here are 15 People Also Ask questions Google shows for the keyword '{X}.' Pick the 5 most relevant for a {audience} reader, rewrite each as a clean question, and write a 60-to-80-word answer for each. Each answer must contain the keyword '{X}' once and end with a definitive statement."
Expected output: 5 ready-to-use FAQ items. When to use: after the draft is complete but before the FAQ section is built.
Prompt 8 — The information-gain check
Goal: pressure-test whether a draft has anything new to say.
Prompt: "Here is a draft article on {topic}. Compare it against the consensus framing of this topic on the open web. List every claim in the draft that goes beyond consensus. List every claim that simply restates consensus. If the count of beyond-consensus claims is less than three, recommend specific sections to strengthen."
Expected output: a punch list of original-versus-consensus claims. When to use: before any draft ships. If the article does not pass the information-gain check, it should not be published.
Prompt 9 — Internal-link suggestion
Goal: identify cluster supports that should be linked from a new article.
Prompt: "Here is a draft article on {topic}. Here is a list of 20 articles already published in this cluster (with titles and URLs). Identify which of the 20 should be linked from the draft, where in the draft to add each link (cite the paragraph), and what anchor text to use. The anchor text must contain the target article's primary keyword."
Expected output: a table of recommended links with paragraph and anchor text. When to use: after the draft is complete, before publish. Internal-linking is the single most under-done AISEO step in most workflows, and ChatGPT handles it well when given the existing cluster as input.
Prompt 10 — Schema markup generator
Goal: produce schema.org JSON-LD for a finished article.
Prompt: "Here is the article title, slug, author byline, and FAQ section. Generate complete schema.org JSON-LD that includes Article, FAQPage, and Author types. The Author block must include sameAs links to {LinkedIn URL}, {GitHub or Twitter}, and {one external talk URL}. Output valid JSON-LD only."
Expected output: production-ready JSON-LD. When to use: at the schema-injection step before publish. Saves the 20 minutes per article a developer would otherwise spend hand-rolling the schema and produces consistent output across the program.
How to turn these into a team prompt library
Save the working prompts in a shared document with three pieces of metadata: the use case, the prompt text, and an example of good output. The example matters most — it lets new team members calibrate quickly. Without an example, prompts get applied to the wrong tasks and produce bad output that the team blames on the prompt rather than on the placement.
Refresh the library quarterly. ChatGPT model updates change what works. A prompt that was reliable in Q1 may produce drift by Q3. Re-run a small validation set — 10 prompts with 5 outputs each — at the start of every quarter and update phrasing where the model has gotten sloppier or sharper. The validation pass takes an hour and saves a quarter's worth of degraded output.
Where this fits in the cluster
This piece is one of four cluster supports under the MOFU pillar. The neighbors:
- SEO for ChatGPT — citation mechanics on the chat engine
- How to use ChatGPT for SEO — the operator workflow
- ChatGPT SEO content — where the tool fits inside your content workflow
Prompts are not magic. They are interfaces. The interface only works when the operator on the other side knows what good output looks like for the task.
