SEO Prompts for AI: Research, Briefs, and Content Checks

Effective SEO prompts give an assistant real search data, a specific reader question, your existing page inventory, and evidence for the answer. Use AI to organize those inputs, propose a page outline, and flag gaps before a person reviews the result. A prompt cannot establish search demand, rankings, or product facts that you have not supplied or verified.

fenbs keeps reusable prompts and approved content rules with the project, so connected assistants can follow the same instructions across writing sessions. Save the method that worked, the decisions behind it, and the checks still outstanding rather than relying on an old chat.

What inputs make an SEO prompt useful?

Start with a reader and a real task. A hypothetical repair shop might need a page explaining how customers prepare for an estimate. Its evidence could include the actual appointment process, service area, and questions staff receive. The assistant should work from those facts rather than inventing customer demand or filling space with general advice.

• Query data: a labeled export, its source, date range, filters, and metric definitions.
• Page inventory: existing URLs, titles, intended reader questions, and any pages planned next.
• Evidence: approved product facts, original examples, and relevant primary sources.
• Boundaries: intended audience, location, action, editorial rules, and topics owned by other pages.

Keep a list of missing inputs. If you have no query data, ask for possible questions to investigate, with that label attached. A model’s plausible keyword list is a brainstorming result. It is not a measurement of how many people search or a reason to create every suggested page.

How can you group queries without duplicating existing pages?

Ask for groups based on the answer a reader needs, then compare each group with your inventory. Different wording can describe the same question. Conversely, similar wording can hide a different task, such as choosing a service versus troubleshooting an existing purchase. A person should review those distinctions before assigning pages.

Analyze this query export: [paste labeled data].
Definitions, date range, and filters: [paste].
Existing and planned page inventory: [paste].
Group queries by the question a reader wants answered.
For each group, show supplied metrics without inventing missing values, likely intent labeled as an inference, existing owner URL, and unresolved questions.
Recommend updating an existing page when it already answers the same question. Do not recommend a new page from wording alone.

Inspect ambiguous groups manually. One page may cover several expressions of the same intent, while a genuinely different audience may require a separate explanation. Record the chosen owner URL so the next assistant does not repeat the exercise and recommend a competing article.

This is where an AI brain helps preserve editorial decisions for the next session. Keep the template focused on page ownership rather than treating it as a full SEO strategy.

How do you produce an outline with something original to say?

Make the assistant identify the answer before suggesting headings. Supply the experience, examples, or source material your team can contribute. Ask which questions cannot yet be answered. An outline that looks complete can conceal an absence of evidence, especially when every section is drawn from a generic topic list.

Plan one page answering [specific reader question].
Reader and desired action: [paste].
Approved facts and original material: [source-labeled packet].
Neighbor pages and their questions: [paste].
Return a two-sentence answer, proposed title, question-based headings, source needed for each factual section, and missing evidence.
Link to neighboring questions without covering them in depth.
Do not copy competitor headings, invent experience, or pad the outline to a word target.

Google’s people-first content guidance↗ (opens in a new tab) asks whether content offers useful original information and a satisfying answer. Use that as a review question: what can your reader do or understand after this page that an empty outline would not provide? If the answer is unclear, improve the material before drafting.

Store the outline prompt in fenbs Skills and prompts, and record a person’s page-ownership choice in Decisions and rules. That preserves both the reusable instruction and the specific editorial judgment behind this assignment.

Which prompts help with titles and descriptions?

Give the assistant the finished page, not just a keyword. Ask for several accurate ways to describe its actual answer. A catchy title that promises a calculator, comparison, or checklist the page does not contain creates a mismatch before the reader even arrives.

Write three title and description pairs for this completed page: [paste].
Main reader question: [question]. Editorial length rules: [rules].
Each pair must describe the actual answer and distinguish this page from these neighbors: [list].
Use natural language. Do not add unsupported superlatives, guarantees, dates, or features.
Show character counts and identify the passage supporting each promise.
Treat length rules as our editorial constraints, not a guarantee of search display.

Google’s title-link documentation↗ (opens in a new tab) explains that title links are generated from several sources and recommends descriptive, concise title text. Your submitted wording does not guarantee the exact display. Check the page title and visible heading together, and remove repeated keywords that do not help someone recognize the answer.

How can an assistant review a draft without inventing fixes?

Ask for findings tied to exact passages, evidence, and proposed actions. Keep factual support separate from stylistic preference. If a product detail lacks a source, the appropriate action may be to remove or verify it, not to rewrite it more confidently.

Review this draft: [paste].
Intended question: [question]. Approved sources: [packet]. Neighbor page boundaries: [list].
Return findings with exact passage, issue, supporting evidence, and recommended action.
Check: unanswered reader questions, unsupported claims, missing qualifications, repetitive sections, misleading headings, and overlap with neighbors.
Separate factual errors from editorial suggestions.
Do not invent search metrics, citations, product behavior, or promises about ranking. Do not publish changes.

Google’s generative-content guidance↗ (opens in a new tab) emphasizes accuracy, quality, and relevance, including review of metadata. Its spam policies↗ (opens in a new tab) address scaled content made primarily to manipulate rankings. A useful prompt should help identify missing value; producing more words does not resolve that problem.

Related guides cover SEO workflows, copywriting prompts, research prompts, teaching prompts, and marketing prompts for broader processes or different assignments. System prompt examples explain how to keep recurring review instructions clear without folding every task into one enormous request.

What changes when you use ChatGPT or Claude?

Use the same inputs and review requirements in either app. For ChatGPT SEO prompts, explicitly say whether the job is analysis of a supplied export or research that needs page access. For Claude prompts for SEO, preserve the same distinction. Confirm available capabilities rather than assuming a model can open a URL or inspect your website.

When you provide a spreadsheet excerpt, describe each metric and filter. When you supply a page, say whether it is a live page, a draft, or an older export. If the assistant cannot access something, ask it to stop that portion of the task and state what material you need to provide. A different model does not turn missing evidence into an established fact.

How can you evaluate changes after publication?

Choose the question you will evaluate before interpreting a chart. Did the page begin appearing for relevant queries? Are people finding the intended answer? Google’s Search Console performance documentation↗ (opens in a new tab) describes clicks, impressions, click-through rate, and average position, with dimensions such as queries and pages. Preserve the report scope when asking AI to summarize it.

Summarize this labeled report: [paste].
Page changes and publication dates: [paste]. Comparison periods and filters: [paste].
Separate observed changes from possible explanations. Show calculations from the supplied values.
Flag small samples, changed filters, missing data, and competing explanations.
Do not attribute a ranking or traffic change to our edit without supporting evidence.
Return follow-up questions and checks, not a promise that one change will improve performance.

Save the observation and its limits. A change after an edit is a reason to investigate, not automatic proof that the edit caused it. Keep your prompt library focused on checks you can repeat; explore the AI prompts collection for other assignments.

How to do this with fenbs

1. Create a free account, then add a project for one website or content program.
2. Save a tested outline prompt and review prompt in Skills and prompts, with their required source inputs and output checks.
3. Record the rules people approved, including page ownership and claim-checking requirements, in Decisions and rules.
4. Connect a supported assistant using its setup guide and ask it to read the project context before using the appropriate prompt.
5. Save progress with completed reviews, unresolved claims, and the next measurement check. Keep source exports and publishing work in their normal approved systems.

fenbs is free for one person, with no credit card. It helps preserve the instructions and decisions behind recurring SEO work. It is not a keyword database, rank tracker, or website crawler. Start with one page and a review method you can verify before expanding the library.