SEO audit and optimization

Output is always a prioritized fix list with concrete rewrites, not a lecture on SEO. You cannot see the site's Search Console, analytics or rendering engine; say so and focus on what the page itself and the live search results show.

Inputs

• A URL: open it with fetch_url. Also try <origin>/robots.txt and <origin>/sitemap.xml if indexing is in question (only report them if they actually loaded).
• A draft: work from the text; ask for the target URL/slug if links or canonicals matter.
• Ask once, if unknown and it matters: target country/language, primary audience, and the one action the page should drive. Otherwise state assumptions.

Workflow

1. Determine the target query and intent

• Identify 1 primary query and 3-8 secondary/related queries the page should rank for.
• Classify intent: informational (learn), commercial (compare), transactional (buy/sign up), navigational (find a brand).
• Check the live results: web_search the primary query (in the target language and, if possible, region). Note what ranks: format (guide, list, tool, product page, video), depth, freshness, and who ranks (brands, forums, publishers). The page must match the dominant format, or the intent is wrong and no tweak will fix it.

Keyword research without paid tools: see references/keyword-research.md.

2. Audit the page

Work through references/audit-checklist.md. The essentials:

3. Content quality

Ask the questions a quality rater would:

• Does it show first-hand experience or expertise (who wrote it, why trust them)?
• Is there something here that the top results lack (data, examples, a tool, a clearer answer, a more recent update)?
• Is it written for people first? Remove filler intros, repeated keyword variants and generic paragraphs.
• Is the date shown and honest? Updating a date without updating content is a bad signal.
• Would a reader leave satisfied, or go back to search?

4. Core Web Vitals notes

You usually cannot measure from here. Give notes, not scores:

• Thresholds for "good" (as of September 2026, verify with web_search if it matters): LCP <= 2.5 s, INP <= 200 ms, CLS <= 0.1, at the 75th percentile of real users.
• Point the user to PageSpeed Insights or Search Console for field data.
• From the HTML you can still flag likely issues: huge hero images without dimensions, render-blocking scripts in <head>, many third-party tags, web fonts without font-display, lazy-loading the LCP image, layout shifts from late-injected banners.

5. Prioritize and write the fix list

Rank each fix by impact x effort:

• P1 - blocks ranking or clicks: wrong intent, noindex/canonical errors, missing or duplicate title/H1, thin content vs competitors, broken key links.
• P2 - meaningful gains: title/meta rewrite, missing sections searchers need, internal links, structured data, image alt text, obvious speed issues.
• P3 - polish: slug tweaks (only on new pages), minor heading wording, extra FAQ.

Output format

**Target:** <primary query> (<intent>, <country/language>)
**What ranks now:** <1-2 lines on the dominant format and competitors>

### Fixes
| # | Priority | Issue | Fix | Effort |
|---|---|---|---|---|
| 1 | P1 | ... | ... | S/M/L |

### Rewrites
- Title: "<new title>" (<n> chars)
- Meta description: "<new description>" (<n> chars)
- H1: "<new H1>"
- Suggested outline: H2 ... / H2 ...

### Structured data (if relevant)
<JSON-LD snippet matching visible content>

### Not checked
<what needs Search Console, PageSpeed or a crawl>

Do not

• Promise rankings or traffic numbers.
• Recommend keyword stuffing, hidden text, doorway pages, buying links, or mass AI-generated pages; these violate search engine spam policies.
• Add structured data for content not visible on the page.
• Recommend changing ranking URLs without redirects.
• Invent search volumes. Without a data tool, describe demand qualitatively and say how you judged it.