Seo magic

SEO Magic

Repackage an existing article so it reads like the answer the searcher came for.

A deeper article can still be the worse search result when its verdict is buried, its title promises the wrong format, or its first screen answers a different question. SEO Magic fixes that mismatch by reordering and reframing useful material before adding anything new.

This is editing surgery, not a blank-page writing workflow and not a technical SEO audit.

Capability and privacy boundary

Read references/privacy-and-tooling.md before using this skill.

At runtime, the only permitted external capabilities are:

• web search
• reading publicly accessible webpages

Everything else is out of scope. Do not use analytics accounts, APIs, plugins, connectors, MCP servers, command-line tools, local files, browser cookies, logins, forms, content-management systems, publishing endpoints, or stored credentials.

Do not publish or modify anything. Return proposed, paste-ready edits for the user to review and apply manually.

Never put personal data into a search query, URL, prompt sent to a third party, or output. If supplied content contains names, email addresses, phone numbers, customer records, private URLs, account identifiers, or unpublished analytics, redact those details before analysis. If redaction would remove evidence needed for the task, ask for an anonymized substitute.

Load contract

Read these files before editing:

1. references/privacy-and-tooling.md
2. references/repackaging-playbook.md
3. references/serp-and-geo-patterns.md

If a target project provides editorial, brand, or formatting rules in the conversation, follow those rules. Do not open local project files.

Use this skill for

• an existing article that ranks for the wrong variation of a query
• a review that delays its buy-or-skip verdict
• a comparison that does not name the winner early
• a listicle with no top pick or quick comparison
• a tutorial that hides prerequisites and expected outcomes
• an informational page that takes too long to define the subject
• a title, snippet, introduction, or heading structure that does not match the live search results
• a useful page that became padded during repeated updates

Do not use this skill for

• writing a new article from scratch
• technical SEO, crawling, indexing, Core Web Vitals, robots, or backlinks
• keyword-volume estimates, rank tracking, or analytics analysis
• site administration or publishing
• logged-in, private, paywalled, or personally identifying content
• fabricated first-person experience, tests, testimonials, or metrics

If less than roughly 40% of the existing article is worth preserving, explain that it needs a fresh draft rather than forcing a surgical repackage.

Required input

Collect:

1. Target query: the one query the page should answer.
2. Existing asset: a public URL or pasted, redacted content.
3. Audience: optional, but useful when the query can serve different readers.
4. Constraints: required sections, claims that must remain, house style, and output format.

If the user provides a private URL, do not attempt to log in. Ask for a redacted excerpt instead.

If the target query is missing, infer one only when the public title, H1, and page topic make the intent unambiguous. Otherwise ask.

Workflow

1. Read the existing article

For a public URL, browse the page and record:

• title tag and H1
• first 150 words
• current meta description if visible in the page source or search result
• H2 and H3 sequence
• where the direct answer or verdict first appears
• original evidence, examples, tables, and screenshots described in the article
• honest limitations or disqualifiers
• sections that serve adjacent rather than core intent

For pasted content, work only with the supplied redacted text.

Preserve useful evidence and distinctive reasoning. Do not invent facts to fill missing fields.

2. Read the live search results

Search the exact target query with normal web search. Browse a small, representative sample of public results, usually three:

• one top-ranking result that clearly matches the dominant format
• one result with a strong title or snippet
• one credible alternative when the results show mixed intent

Record:

• dominant result type: review, comparison, ranked list, tutorial, or informational answer
• repeated title and heading patterns
• what the snippets answer
• how quickly each page gives its main answer
• whether tables, ordered steps, pros and cons, or FAQs are common
• questions and objections repeated across results

Search results are evidence of the format currently rewarded, not permission to copy competitors. Do not reproduce distinctive wording.

3. Classify the query intent

Intent	Common query shape	Answer expected near the top
Review	{product} review, is {product} worth it	verdict, best-fit reader, and one reason to skip
Comparison	{A} vs {B}	winner, deciding factor, and the case where the other option wins
Best-X	best {category}, best {category} for {audience}	top pick, ranking logic, and best-by-use-case summary
How-to	how to {outcome}, {tool} setup	expected outcome, prerequisites, time or difficulty, then ordered steps
Informational	what is {topic}, why {topic}	direct definition or explanation in sentence one

Then classify what the first 30% of the existing article actually delivers. The gap between these two classifications is the main diagnosis.

4. Run the repackaging diagnostic

Check for:

1. promotional framing where the query expects evaluation
2. a snippet that names the topic but does not answer the doubt
3. content formatted for the wrong intent
4. adjacent sections placed before the core answer
5. original evidence buried too deep
6. an exaggerated or vague title
7. missing trade-offs or "not for you" guidance
8. padding that increases time-to-answer

Report only triggered signals. Do not produce a generic scorecard.

5. Apply fixes in leverage order

Use this order:

1. title tag
2. H1
3. first 100 to 150 words
4. snippet-target paragraph
5. proof or methodology summary, using only evidence already present
6. trade-offs and disqualifier
7. quick comparison, ordered steps, or key-facts table when the intent calls for one
8. heading sequence
9. meta description
10. FAQ questions supported by the article and public research

Reorder before rewriting. Rewrite before adding. Add a new section only when public research exposes a meaningful question the page does not answer.

6. Make passages extractable

• Open major sections with a direct answer.
• Use explicit entities instead of ambiguous pronouns.
• Keep one main idea per paragraph.
• Use semantic tables for real comparisons.
• Use ordered lists for procedures.
• Use bullets for parallel attributes, pros, cons, or requirements.
• Phrase headings as clear questions when that matches how people search.
• Keep every claim understandable when read outside the full article.

Do not force a fixed number of entities, keywords, or words into each section. Clarity and factual support come first.

7. Protect evidence integrity

• Never turn researched information into first-person experience.
• Never claim a product was tested unless the supplied article shows who tested it, how, when, and under what conditions.
• Never invent scores, prices, dates, statistics, screenshots, testimonials, or results.
• Mark time-sensitive facts for manual verification.
• Cite the public source beside any newly introduced factual claim.
• When sources disagree, state the disagreement instead of choosing silently.

8. Stop before publishing

This public skill never writes to a website or account.

Return:

• proposed title and H1
• proposed meta description
• rewritten introduction or verdict block
• snippet-target paragraph
• proposed heading sequence
• paste-ready tables, lists, or FAQ text where useful
• a short manual verification checklist
• public sources consulted

Every change must be labeled [proposed]. Never label anything [published], [pushed], or [verified live].

Intent skeletons

Review

Verdict → who it is for → who should skip → evidence or methodology → quick pros and cons → pricing and limitations → alternatives → FAQ

Comparison

Winner and deciding factor → quick comparison table → dimension-by-dimension sections → where A wins → where B wins → final recommendation → FAQ

Best-X

Top pick → best-by-use-case summary → selection criteria → compact comparison table → individual entries with one differentiator and one honest limitation → FAQ

How-to

Outcome → prerequisites → ordered steps → common failure points → troubleshooting → verification checklist

Informational

Direct definition → why it matters → how it works → examples → limits or exceptions → related questions

Output format

Diagnosis

Three to six lines:

• target query and intent
• current page format
• biggest mismatch
• highest-leverage changes

Proposed surfaces

• [proposed] Title tag
• [proposed] H1
• [proposed] Meta description
• [proposed] First 100 to 150 words
• [proposed] Snippet-target paragraph
• [proposed] Heading sequence
• [proposed] Table, list, or FAQ text when useful

Preservation notes

• evidence and sections to keep
• sections to move
• sections to cut or merge
• claims requiring manual verification

Sources consulted

List only public webpages actually browsed. Do not include search queries containing private information.

Manual next steps

• review facts, links, and house style
• apply approved edits in the publishing system
• verify the rendered page
• monitor performance through tools the user chooses outside this skill

Final gate

• Does the first screen answer the target query?
• Does the page use the format the live results currently reward?
• Is the title accurate rather than exaggerated?
• Does the snippet resolve the searcher's doubt?
• Is original evidence visible early?
• Are trade-offs honest and specific?
• Did the edit reduce time-to-answer?
• Did every new factual claim receive a public source?
• Is all personal data removed?
• Were only browsing capabilities used?
• Is every change still labeled [proposed]?

If any answer is no, keep editing.