AI can help you write an introduction, review a page, organize research, or work through a search performance problem. The useful part is giving it enough context to do the job and knowing what to question in the answer.
This guide includes prompts for common SEO and content tasks, along with the information to provide, follow-up questions, and checks before you use the result. Adapt them to your work. You do not need to paste every instruction into every conversation.
Start with the job and the information it needs
Before writing the prompt, identify what you want to produce or decide. “Improve this page” leaves a lot open. “Help someone comparing these two products understand the differences that affect their choice” gives the task a direction.
Useful context usually includes:
The task: What you need written, reviewed, or decided.
The audience: Who the work is for and what they already know.
The source material: Page content, product information, research, data, or examples.
The constraints: Facts to preserve, changes you cannot make, tone, scope, or timing.
The output: A revised paragraph, prioritized findings, a table, or a recommendation.
You can start simply:
I’m working on [page or problem] for [audience]. They need to [understand something or make a decision]. Review the material below and help me [specific task]. Keep [important constraints]. If the information does not support a conclusion, say what is missing. Return [desired output].
Include a piece of approved writing when tone matters. A concrete example gives the model something to work from beyond “make it human.”
If you provide a URL, ask it to confirm whether it accessed the page. If it cannot, paste the content or upload a file. For data analysis, include the date range, filters, and what each metric means.
1. Write a useful page introduction
Use this when an introduction takes too long to get to the point or could belong on almost any page.
Provide: The main page content, the intended audience, the page’s purpose, and an approved writing sample if you have one.
Starting prompt
Rewrite the introduction for this page.
The reader is [audience]. They have landed here because [need or question]. The page helps them [specific purpose].
Read the full content first. Start with what the reader needs to know and make it clear what this page covers. Keep the introduction specific to the content that follows.
Use plain language. Cut generic setup, repeated points, and claims the page does not support. Work the main topic in naturally. Do not add benefits, features, or promises that are missing from the source.
Give me one recommended version and briefly explain any meaningful change. Aim for 60–100 words, but use less if that is enough.
Page content: [paste]
Tone example, if available: [paste]
Rewrite the introduction for this page.
The reader is [audience]. They have landed here because [need or question]. The page helps them [specific purpose].
Read the full content first. Start with what the reader needs to know and make it clear what this page covers. Keep the introduction specific to the content that follows.
Use plain language. Cut generic setup, repeated points, and claims the page does not support. Work the main topic in naturally. Do not add benefits, features, or promises that are missing from the source.
Give me one recommended version and briefly explain any meaningful change. Aim for 60–100 words, but use less if that is enough.
Page content: [paste]
Tone example, if available: [paste]
Rewrite the introduction for this page.
The reader is [audience]. They have landed here because [need or question]. The page helps them [specific purpose].
Read the full content first. Start with what the reader needs to know and make it clear what this page covers. Keep the introduction specific to the content that follows.
Use plain language. Cut generic setup, repeated points, and claims the page does not support. Work the main topic in naturally. Do not add benefits, features, or promises that are missing from the source.
Give me one recommended version and briefly explain any meaningful change. Aim for 60–100 words, but use less if that is enough.
Page content: [paste]
Tone example, if available: [paste]
If the result still feels generic
This could introduce almost any page on this topic. Make it more specific to what this page actually helps the reader do. Remove anything that does not help them understand the topic or decide whether to keep reading
This could introduce almost any page on this topic. Make it more specific to what this page actually helps the reader do. Remove anything that does not help them understand the topic or decide whether to keep reading
This could introduce almost any page on this topic. Make it more specific to what this page actually helps the reader do. Remove anything that does not help them understand the topic or decide whether to keep reading
Check before using it: Does the page deliver what the introduction promises? Does it explain the topic without repeating the H1? Would the intended reader understand it without additional context?
2. Summarize a page without losing the useful detail
Use this for a summary or key takeaways section when readers need the main answer quickly.
Provide: The complete source content and the question the summary should answer.
Starting prompt
Summarize this page for someone who wants to know [question].
Give the main answer first, followed by up to four useful takeaways. Preserve the conditions, exceptions, and limitations that would change the answer.
Use only the supplied content. Keep important numbers and their context accurate. Do not strengthen the conclusion or add advice that the page does not contain.
If the page contradicts itself or does not answer the question, flag that separately instead of smoothing it over.
Content: [paste]
Summarize this page for someone who wants to know [question].
Give the main answer first, followed by up to four useful takeaways. Preserve the conditions, exceptions, and limitations that would change the answer.
Use only the supplied content. Keep important numbers and their context accurate. Do not strengthen the conclusion or add advice that the page does not contain.
If the page contradicts itself or does not answer the question, flag that separately instead of smoothing it over.
Content: [paste]
Summarize this page for someone who wants to know [question].
Give the main answer first, followed by up to four useful takeaways. Preserve the conditions, exceptions, and limitations that would change the answer.
Use only the supplied content. Keep important numbers and their context accurate. Do not strengthen the conclusion or add advice that the page does not contain.
If the page contradicts itself or does not answer the question, flag that separately instead of smoothing it over.
Content: [paste]
Useful follow-up
Check the summary against the source. Which statements lost a qualification or became more certain? Correct those and remove any takeaway that repeats another
Check the summary against the source. Which statements lost a qualification or became more certain? Correct those and remove any takeaway that repeats another
Check the summary against the source. Which statements lost a qualification or became more certain? Correct those and remove any takeaway that repeats another
Check before using it: Are the main answer and its limits intact? Do figures retain their units, time periods, and comparison points?
3. Write FAQs that answer remaining questions
Use this when readers have practical questions the main content has not fully answered. Some answers may belong in the main page instead of a separate FAQ section.
Provide: Page content, verified product or service facts, and actual customer questions when available.
Starting prompt
Review this page for questions a [audience] would still have before [decision or action].
Use the customer questions below where available. Separate questions found in those sources from questions you are proposing.
For each useful question, recommend whether to answer it in the main content or an FAQ. Skip questions the page already answers clearly unless a short clarification would help.
Draft answers only where the supplied information supports them. Flag missing facts instead of inventing policies, prices, features, eligibility rules, or guarantees.
Return the question, recommended placement, draft answer, and any fact we need to confirm. There is no minimum number of FAQs.
Page: [paste]
Verified facts: [paste]
Customer questions, if available: [paste]
Review this page for questions a [audience] would still have before [decision or action].
Use the customer questions below where available. Separate questions found in those sources from questions you are proposing.
For each useful question, recommend whether to answer it in the main content or an FAQ. Skip questions the page already answers clearly unless a short clarification would help.
Draft answers only where the supplied information supports them. Flag missing facts instead of inventing policies, prices, features, eligibility rules, or guarantees.
Return the question, recommended placement, draft answer, and any fact we need to confirm. There is no minimum number of FAQs.
Page: [paste]
Verified facts: [paste]
Customer questions, if available: [paste]
Review this page for questions a [audience] would still have before [decision or action].
Use the customer questions below where available. Separate questions found in those sources from questions you are proposing.
For each useful question, recommend whether to answer it in the main content or an FAQ. Skip questions the page already answers clearly unless a short clarification would help.
Draft answers only where the supplied information supports them. Flag missing facts instead of inventing policies, prices, features, eligibility rules, or guarantees.
Return the question, recommended placement, draft answer, and any fact we need to confirm. There is no minimum number of FAQs.
Page: [paste]
Verified facts: [paste]
Customer questions, if available: [paste]
Useful follow-up
Which of these questions would actually affect someone's decision? Keep those. Remove questions that exist mainly to repeat a keyword or repackage the page headings.
Which of these questions would actually affect someone's decision? Keep those. Remove questions that exist mainly to repeat a keyword or repackage the page headings.
Which of these questions would actually affect someone's decision? Keep those. Remove questions that exist mainly to repeat a keyword or repackage the page headings.
Check before using it: Are the answers accurate and specific? Do they resolve uncertainty? Adding an FAQ is an editorial choice, not evidence that the page will gain AI citations.
4. Review titles, H1s, and meta descriptions
Use this when page labels are vague, repetitive, misleading, or disconnected from the content.
Provide: The full main content, current title/H1/description, and relevant query data if available.
Starting prompt
Review this page's title tag, H1, and meta description.
First, identify what the page actually covers and the reader need it serves. Then recommend one title, one H1, and one meta description that accurately describe it.
Use relevant language from the supplied query data when it fits the content. Do not invent search volume or force a query onto a page that does not answer it.
Make the wording specific and easy to understand. Avoid repeated phrases and unsupported claims such as “best,” “complete,” or “free.” The H1 and title can differ when that improves clarity.
Aim for a title around 50–60 characters and a description around 140–155 characters as editorial targets. Prioritize accuracy and readability. Include character counts so I can review the length.
Return the current and recommended versions, plus a short explanation. If the page itself needs a change to support the proposed wording, flag that.
URL: [URL]
Current metadata and H1: [paste]
Page content: [paste]
Relevant query data, including source and market: [paste or “not available”]
Review this page's title tag, H1, and meta description.
First, identify what the page actually covers and the reader need it serves. Then recommend one title, one H1, and one meta description that accurately describe it.
Use relevant language from the supplied query data when it fits the content. Do not invent search volume or force a query onto a page that does not answer it.
Make the wording specific and easy to understand. Avoid repeated phrases and unsupported claims such as “best,” “complete,” or “free.” The H1 and title can differ when that improves clarity.
Aim for a title around 50–60 characters and a description around 140–155 characters as editorial targets. Prioritize accuracy and readability. Include character counts so I can review the length.
Return the current and recommended versions, plus a short explanation. If the page itself needs a change to support the proposed wording, flag that.
URL: [URL]
Current metadata and H1: [paste]
Page content: [paste]
Relevant query data, including source and market: [paste or “not available”]
Review this page's title tag, H1, and meta description.
First, identify what the page actually covers and the reader need it serves. Then recommend one title, one H1, and one meta description that accurately describe it.
Use relevant language from the supplied query data when it fits the content. Do not invent search volume or force a query onto a page that does not answer it.
Make the wording specific and easy to understand. Avoid repeated phrases and unsupported claims such as “best,” “complete,” or “free.” The H1 and title can differ when that improves clarity.
Aim for a title around 50–60 characters and a description around 140–155 characters as editorial targets. Prioritize accuracy and readability. Include character counts so I can review the length.
Return the current and recommended versions, plus a short explanation. If the page itself needs a change to support the proposed wording, flag that.
URL: [URL]
Current metadata and H1: [paste]
Page content: [paste]
Relevant query data, including source and market: [paste or “not available”]
Useful follow-up
These still feel broad. Can someone tell what is on this specific page from the title alone? Tighten the wording without adding a promise the content cannot meet
These still feel broad. Can someone tell what is on this specific page from the title alone? Tighten the wording without adding a promise the content cannot meet
These still feel broad. Can someone tell what is on this specific page from the title alone? Tighten the wording without adding a promise the content cannot meet
Check before using it: Confirm character counts, page fit, and accuracy. Preview the wording in context. Treat length targets as working guidelines, not a promise about how a search result will display.
5. Build a content brief around a reader’s decision
Use this before drafting a new page or substantially changing an existing one.
Provide: The audience, reader task, business context, existing related pages, source material, and search research if available.
Starting prompt
Help me build a content brief for [topic].
The reader is [audience]. They need to [question, task, or decision]. Our business offers [relevant offering], and this page should help them [appropriate next step].
Review the supplied research and existing pages before proposing the structure. Tell me whether this needs a new page or would fit better as an update to something we already have.
Build the brief around what the reader needs to understand. Use the query group and search results as inputs, without copying competitor outlines or adding sections just because competitors have them.
Include:
- The page's purpose and main answer.
- A proposed H1 and ordered outline.
- What each section should explain.
- Evidence, examples, or expert input needed.
- Relevant existing pages to link to.
- What is outside the scope.
Separate supported findings from assumptions. Do not invent research, search volume, or product claims.
Audience and business context: [paste]
Research and query group: [paste]
Existing pages and source material: [paste]
Help me build a content brief for [topic].
The reader is [audience]. They need to [question, task, or decision]. Our business offers [relevant offering], and this page should help them [appropriate next step].
Review the supplied research and existing pages before proposing the structure. Tell me whether this needs a new page or would fit better as an update to something we already have.
Build the brief around what the reader needs to understand. Use the query group and search results as inputs, without copying competitor outlines or adding sections just because competitors have them.
Include:
- The page's purpose and main answer.
- A proposed H1 and ordered outline.
- What each section should explain.
- Evidence, examples, or expert input needed.
- Relevant existing pages to link to.
- What is outside the scope.
Separate supported findings from assumptions. Do not invent research, search volume, or product claims.
Audience and business context: [paste]
Research and query group: [paste]
Existing pages and source material: [paste]
Help me build a content brief for [topic].
The reader is [audience]. They need to [question, task, or decision]. Our business offers [relevant offering], and this page should help them [appropriate next step].
Review the supplied research and existing pages before proposing the structure. Tell me whether this needs a new page or would fit better as an update to something we already have.
Build the brief around what the reader needs to understand. Use the query group and search results as inputs, without copying competitor outlines or adding sections just because competitors have them.
Include:
- The page's purpose and main answer.
- A proposed H1 and ordered outline.
- What each section should explain.
- Evidence, examples, or expert input needed.
- Relevant existing pages to link to.
- What is outside the scope.
Separate supported findings from assumptions. Do not invent research, search volume, or product claims.
Audience and business context: [paste]
Research and query group: [paste]
Existing pages and source material: [paste]
Useful follow-up
Review this as an editor. What is repetitive, unnecessary, or too broad for the reader's task? Simplify the outline and identify what would make this page more useful than the material we supplied.
Review this as an editor. What is repetitive, unnecessary, or too broad for the reader's task? Simplify the outline and identify what would make this page more useful than the material we supplied.
Review this as an editor. What is repetitive, unnecessary, or too broad for the reader's task? Simplify the outline and identify what would make this page more useful than the material we supplied.
Check before using it: Can a writer see what to say, what evidence to gather, and what to leave out? Is a new page actually needed?
6. Review existing content for usefulness and accuracy
Use this when a page feels wordy, generic, incomplete, or difficult to act on.
Provide: Full page content, audience, page purpose, and supporting sources.
Starting prompt
Review this page for someone who needs to [reader task].
Tell me whether it answers their question clearly, provides enough detail to act, and supports its main claims.
Identify vague advice, missing explanations, repetition, unsupported claims, and anything that sounds more certain than the evidence allows. Point to the specific passage for each finding.
Prioritize the changes that would materially improve the page. Recommend what to keep as well as what to change. Do not rewrite everything by default or recommend FAQs, schema, or more content without explaining the specific problem they would solve.
Return: priority, passage or section, issue, recommended change, and any source or fact needed. Distinguish verified errors from claims that need checking.
Page purpose and audience: [paste]
Page content: [paste]
Supporting sources: [paste]
Review this page for someone who needs to [reader task].
Tell me whether it answers their question clearly, provides enough detail to act, and supports its main claims.
Identify vague advice, missing explanations, repetition, unsupported claims, and anything that sounds more certain than the evidence allows. Point to the specific passage for each finding.
Prioritize the changes that would materially improve the page. Recommend what to keep as well as what to change. Do not rewrite everything by default or recommend FAQs, schema, or more content without explaining the specific problem they would solve.
Return: priority, passage or section, issue, recommended change, and any source or fact needed. Distinguish verified errors from claims that need checking.
Page purpose and audience: [paste]
Page content: [paste]
Supporting sources: [paste]
Review this page for someone who needs to [reader task].
Tell me whether it answers their question clearly, provides enough detail to act, and supports its main claims.
Identify vague advice, missing explanations, repetition, unsupported claims, and anything that sounds more certain than the evidence allows. Point to the specific passage for each finding.
Prioritize the changes that would materially improve the page. Recommend what to keep as well as what to change. Do not rewrite everything by default or recommend FAQs, schema, or more content without explaining the specific problem they would solve.
Return: priority, passage or section, issue, recommended change, and any source or fact needed. Distinguish verified errors from claims that need checking.
Page purpose and audience: [paste]
Page content: [paste]
Supporting sources: [paste]
Useful follow-up
Rewrite the three highest-priority passages. Preserve the useful detail and my tone. Keep explanations direct enough for someone unfamiliar with the subject. Mark anything that still needs a source
Rewrite the three highest-priority passages. Preserve the useful detail and my tone. Keep explanations direct enough for someone unfamiliar with the subject. Mark anything that still needs a source
Rewrite the three highest-priority passages. Preserve the useful detail and my tone. Keep explanations direct enough for someone unfamiliar with the subject. Mark anything that still needs a source
Check before using it: Verify factual corrections against reliable sources. A convincing critique is not proof that a claim is false.
7. Find audience questions in discussions and customer research
Use this to organize relevant Reddit discussions, interviews, reviews, or support conversations.
Provide: Relevant source material with source IDs, dates, and links where available. Remove personal information that is not needed for the task.
Starting prompt
Review these discussions to understand how [audience] thinks about [problem].
Identify the questions they ask, the tradeoffs they care about, the language they use, and what leaves them dissatisfied or uncertain.
For each theme, include its source IDs and a short supporting quote. Keep quoted language separate from your interpretation. Count occurrences only within the supplied sample and explain what you counted.
Do not treat this sample as representative of the whole market. Distinguish recurring themes from isolated comments and flag conflicting views.
Then suggest how the findings could improve an existing page, a product explanation, or another useful resource. Do not turn every comment into a blog topic.
Audience and research question: [paste]
Source material with IDs, dates, and links: [paste]
Review these discussions to understand how [audience] thinks about [problem].
Identify the questions they ask, the tradeoffs they care about, the language they use, and what leaves them dissatisfied or uncertain.
For each theme, include its source IDs and a short supporting quote. Keep quoted language separate from your interpretation. Count occurrences only within the supplied sample and explain what you counted.
Do not treat this sample as representative of the whole market. Distinguish recurring themes from isolated comments and flag conflicting views.
Then suggest how the findings could improve an existing page, a product explanation, or another useful resource. Do not turn every comment into a blog topic.
Audience and research question: [paste]
Source material with IDs, dates, and links: [paste]
Review these discussions to understand how [audience] thinks about [problem].
Identify the questions they ask, the tradeoffs they care about, the language they use, and what leaves them dissatisfied or uncertain.
For each theme, include its source IDs and a short supporting quote. Keep quoted language separate from your interpretation. Count occurrences only within the supplied sample and explain what you counted.
Do not treat this sample as representative of the whole market. Distinguish recurring themes from isolated comments and flag conflicting views.
Then suggest how the findings could improve an existing page, a product explanation, or another useful resource. Do not turn every comment into a blog topic.
Audience and research question: [paste]
Source material with IDs, dates, and links: [paste]
Useful follow-up
Which findings could change how someone evaluates or chooses a solution? Prioritize those and show the source evidence. What additional research would help validate them
Which findings could change how someone evaluates or chooses a solution? Prioritize those and show the source evidence. What additional research would help validate them
Which findings could change how someone evaluates or chooses a solution? Prioritize those and show the source evidence. What additional research would help validate them
Check before using it: Open the sources. Confirm quotes and counts. Keep observations from a small sample separate from claims about market demand.
8. Analyze a sample of AI answers
Use this after collecting answers to relevant questions. Keep collection separate from analysis so the analysis instructions do not influence the answers being measured.
Provide: Exact prompts, full responses, citations, platform, collection date, and run IDs. Include model or mode details and search settings when recorded.
Starting prompt
Analyze this collected sample of AI answers for [brand] and [competitors].
Use only the supplied responses. Do not generate new answers or fill in missing runs.
For each response, record:
- Whether each brand is explicitly mentioned.
- Whether it is recommended, merely listed, or described negatively, with supporting text.
- Whether a link cites the brand's own website.
- Which third-party sources are cited, and whether the response connects them to a claim about the brand.
Keep brand mentions and website citations separate. Flag ambiguous references for review.
Summarize patterns within this sample. If you calculate rates, define the numerator and denominator and report missing or failed runs separately. Do not describe this sample as total AI visibility or infer why a source was selected from the response alone.
End with a few evidence-based questions worth investigating next.
Brand names, aliases, and domains: [paste]
Collected responses and metadata: [paste]
Analyze this collected sample of AI answers for [brand] and [competitors].
Use only the supplied responses. Do not generate new answers or fill in missing runs.
For each response, record:
- Whether each brand is explicitly mentioned.
- Whether it is recommended, merely listed, or described negatively, with supporting text.
- Whether a link cites the brand's own website.
- Which third-party sources are cited, and whether the response connects them to a claim about the brand.
Keep brand mentions and website citations separate. Flag ambiguous references for review.
Summarize patterns within this sample. If you calculate rates, define the numerator and denominator and report missing or failed runs separately. Do not describe this sample as total AI visibility or infer why a source was selected from the response alone.
End with a few evidence-based questions worth investigating next.
Brand names, aliases, and domains: [paste]
Collected responses and metadata: [paste]
Analyze this collected sample of AI answers for [brand] and [competitors].
Use only the supplied responses. Do not generate new answers or fill in missing runs.
For each response, record:
- Whether each brand is explicitly mentioned.
- Whether it is recommended, merely listed, or described negatively, with supporting text.
- Whether a link cites the brand's own website.
- Which third-party sources are cited, and whether the response connects them to a claim about the brand.
Keep brand mentions and website citations separate. Flag ambiguous references for review.
Summarize patterns within this sample. If you calculate rates, define the numerator and denominator and report missing or failed runs separately. Do not describe this sample as total AI visibility or infer why a source was selected from the response alone.
End with a few evidence-based questions worth investigating next.
Brand names, aliases, and domains: [paste]
Collected responses and metadata: [paste]
Useful follow-up
For the answers where competitors appear and we do not, what can we actually observe? Separate those observations from hypotheses about content, product fit, authority, or source selection. What evidence would help test each hypothesis
For the answers where competitors appear and we do not, what can we actually observe? Separate those observations from hypotheses about content, product fit, authority, or source selection. What evidence would help test each hypothesis
For the answers where competitors appear and we do not, what can we actually observe? Separate those observations from hypotheses about content, product fit, authority, or source selection. What evidence would help test each hypothesis
Check before using it: Manually verify classifications and calculations. An answer sample describes the observations collected under those conditions. It does not establish a universal visibility score or explain a system's source-selection process.
9. Investigate a search performance change
Use this when a page or group of pages gains or loses visibility and you need a focused investigation.
Provide: Comparable reporting periods, metric definitions, filters, affected URLs, relevant query data, and known site changes.
Starting prompt
Help me investigate this search performance change.
Start by checking whether the two periods and datasets are comparable. Flag missing fields, unequal periods, filter changes, or aggregation issues that could affect the analysis.
Describe what changed in the supplied data before suggesting causes. Separate page-level and query-level findings, and segment by device, country, or brand only when the data supports it.
Do not sum overlapping exports. Recalculate CTR from clicks and impressions where appropriate, and do not take an unweighted average of row-level positions as an overall position metric.
Rank plausible explanations by the evidence available. For each, show the observation supporting it, what we cannot conclude, and the next check that would help confirm or rule it out.
Do not assume the cause is an algorithm update, AI search, or content quality because traffic changed. End with the three most useful next actions.
Question: [what changed and why it matters]
Data, periods, definitions, and filters: [attach or paste]
Known changes and dates: [paste]
Help me investigate this search performance change.
Start by checking whether the two periods and datasets are comparable. Flag missing fields, unequal periods, filter changes, or aggregation issues that could affect the analysis.
Describe what changed in the supplied data before suggesting causes. Separate page-level and query-level findings, and segment by device, country, or brand only when the data supports it.
Do not sum overlapping exports. Recalculate CTR from clicks and impressions where appropriate, and do not take an unweighted average of row-level positions as an overall position metric.
Rank plausible explanations by the evidence available. For each, show the observation supporting it, what we cannot conclude, and the next check that would help confirm or rule it out.
Do not assume the cause is an algorithm update, AI search, or content quality because traffic changed. End with the three most useful next actions.
Question: [what changed and why it matters]
Data, periods, definitions, and filters: [attach or paste]
Known changes and dates: [paste]
Help me investigate this search performance change.
Start by checking whether the two periods and datasets are comparable. Flag missing fields, unequal periods, filter changes, or aggregation issues that could affect the analysis.
Describe what changed in the supplied data before suggesting causes. Separate page-level and query-level findings, and segment by device, country, or brand only when the data supports it.
Do not sum overlapping exports. Recalculate CTR from clicks and impressions where appropriate, and do not take an unweighted average of row-level positions as an overall position metric.
Rank plausible explanations by the evidence available. For each, show the observation supporting it, what we cannot conclude, and the next check that would help confirm or rule it out.
Do not assume the cause is an algorithm update, AI search, or content quality because traffic changed. End with the three most useful next actions.
Question: [what changed and why it matters]
Data, periods, definitions, and filters: [attach or paste]
Known changes and dates: [paste]
Useful follow-up
Which recommendation is most strongly supported by the data we have? What should we investigate before making changes? Remove recommendations that do not connect to a specific finding
Which recommendation is most strongly supported by the data we have? What should we investigate before making changes? Remove recommendations that do not connect to a specific finding
Which recommendation is most strongly supported by the data we have? What should we investigate before making changes? Remove recommendations that do not connect to a specific finding
Check before using it: Reproduce the key calculations in your reporting tool or spreadsheet. Confirm that missing rows, filters, seasonality, and changes in query mix have not distorted the comparison.
10. Plan an SEO test and interpret the evidence
Use this when you have a specific change to evaluate, such as an internal linking update or a page-template change.
Provide: The hypothesis, eligible pages, baseline data, implementation constraints, and other planned changes.
Starting prompt
Help me decide whether and how to test [specific change].
The hypothesis is [expected effect and reason]. We can change [scope], and our constraints are [traffic, page count, engineering, timing].
Review the baseline data and tell me what test design is feasible. Explain whether a comparable control group is possible and how pages should be selected or assigned. Consider differences between pages, overlapping effects, seasonality, and concurrent changes.
Define the primary outcome, guardrails, implementation checks, and conditions for interpreting the result. Identify the data needed to estimate a suitable duration or detectable effect. Do not invent a sample size, stopping date, confidence level, or expected lift.
If this can only support a before-and-after observation, say so and explain the limits. Keep the plan practical for the resources available.
Change and hypothesis: [paste]
Eligible pages and baseline data: [attach]
Constraints and other planned changes: [paste]
Help me decide whether and how to test [specific change].
The hypothesis is [expected effect and reason]. We can change [scope], and our constraints are [traffic, page count, engineering, timing].
Review the baseline data and tell me what test design is feasible. Explain whether a comparable control group is possible and how pages should be selected or assigned. Consider differences between pages, overlapping effects, seasonality, and concurrent changes.
Define the primary outcome, guardrails, implementation checks, and conditions for interpreting the result. Identify the data needed to estimate a suitable duration or detectable effect. Do not invent a sample size, stopping date, confidence level, or expected lift.
If this can only support a before-and-after observation, say so and explain the limits. Keep the plan practical for the resources available.
Change and hypothesis: [paste]
Eligible pages and baseline data: [attach]
Constraints and other planned changes: [paste]
Help me decide whether and how to test [specific change].
The hypothesis is [expected effect and reason]. We can change [scope], and our constraints are [traffic, page count, engineering, timing].
Review the baseline data and tell me what test design is feasible. Explain whether a comparable control group is possible and how pages should be selected or assigned. Consider differences between pages, overlapping effects, seasonality, and concurrent changes.
Define the primary outcome, guardrails, implementation checks, and conditions for interpreting the result. Identify the data needed to estimate a suitable duration or detectable effect. Do not invent a sample size, stopping date, confidence level, or expected lift.
If this can only support a before-and-after observation, say so and explain the limits. Keep the plan practical for the resources available.
Change and hypothesis: [paste]
Eligible pages and baseline data: [attach]
Constraints and other planned changes: [paste]
When results are available
Review this test readout against the original design.
First check implementation, data completeness, group comparability, and concurrent changes. Explain the observed result and its uncertainty using only the supplied data and analysis.
Separate a descriptive change from evidence that the treatment caused it. Do not claim statistical significance unless the relevant method and calculation support that claim.
Recommend whether to roll out, continue collecting data, revise the test, or stop. Explain what makes that recommendation reasonable and what remains unresolved.
Original test plan: [paste]
Results and analysis method: [attach]
Implementation notes and other changes: [paste]
Review this test readout against the original design.
First check implementation, data completeness, group comparability, and concurrent changes. Explain the observed result and its uncertainty using only the supplied data and analysis.
Separate a descriptive change from evidence that the treatment caused it. Do not claim statistical significance unless the relevant method and calculation support that claim.
Recommend whether to roll out, continue collecting data, revise the test, or stop. Explain what makes that recommendation reasonable and what remains unresolved.
Original test plan: [paste]
Results and analysis method: [attach]
Implementation notes and other changes: [paste]
Review this test readout against the original design.
First check implementation, data completeness, group comparability, and concurrent changes. Explain the observed result and its uncertainty using only the supplied data and analysis.
Separate a descriptive change from evidence that the treatment caused it. Do not claim statistical significance unless the relevant method and calculation support that claim.
Recommend whether to roll out, continue collecting data, revise the test, or stop. Explain what makes that recommendation reasonable and what remains unresolved.
Original test plan: [paste]
Results and analysis method: [attach]
Implementation notes and other changes: [paste]
Check before using it: Validate the calculations and analysis method. A control group helps with comparison, but its presence alone does not establish a reliable causal result.
Refine the answer with specific feedback
You do not need to restart whenever the first answer misses. Identify what is wrong and give the next instruction enough direction to fix it.
Sounds generic “Point to the details in the source that make this specific. Rewrite around those.”
Is too long “Keep the main point, supporting evidence, and next action. Cut repetition and generic setup.”
Makes unsupported claims “Which claims are supported by the supplied material? Remove or flag the rest.”
Gives too many recommendations “Choose the three changes most likely to matter. Explain the evidence and dependencies.”
Jumps straight to writing “First tell me what the reader needs and what the current page is missing.”
Sounds unlike you “Use this approved example as the tone reference. Preserve the meaning and useful detail.”
Treats a guess as a finding “Separate direct observations, interpretations, and things we need to validate.”
Misses the business problem “Connect each recommendation to this audience, offering, and decision. Remove anything that does not fit.”
Review the result before you use it
Before publishing copy or acting on an analysis, check:
Accuracy: Are facts, quotations, numbers, and product details correct?
Evidence: Can you trace important claims to the supplied material or verified sources?
Usefulness: Does the output answer the reader's question or help you make the decision?
Specificity: Is it grounded in this page, audience, and business?
Uncertainty: Does it distinguish what is known from what still needs checking?
Voice: Would you actually say it this way?
For repeated tasks, keep an approved example and a short list of checks. When the work becomes a workflow or tool, those checks still matter.
About this guide
SERPish is where I build tools, test ideas, and share what I learn about SEO, AI discovery, content, and automation. That work includes experiments run on this site.
The prompts here are starting points you can adapt. Specific test results and findings belong in Research, with the method and limitations explained.
Put it to work
Choose a task above and try it with one page or a small dataset. Review the output before applying it more broadly.
For recurring writing tasks, explore the SERPish AI writers. For experiments and findings, visit Research.
