Claude SEO: Prompts for Keywords, Briefs, Audits, Links and Schema

The common mistake with Claude and SEO is asking it for keyword volumes. It will answer, and the numbers will look plausible, and they will be invented. Give Claude real data and keep it on the jobs where reading and judgment matter. That covers most of the work.

What Claude does well and what it needs

Keyword clustering

Export a list from your keyword tool with volume, difficulty and the current ranking URL if you have one. Then:

Here are 400 keywords with monthly volume, difficulty and our current
ranking URL (blank if we do not rank).

1. Cluster them by search intent: keywords that one page should target together.
2. For each cluster: a primary keyword, total volume, the intent
   (informational, commercial, transactional, navigational), and the page
   type that wins it (guide, comparison, product, category, tool).
3. Mark clusters where two of our URLs already rank (cannibalization risk).
4. Sort clusters by total volume divided by average difficulty.
Use only the numbers in the file. Do not estimate missing volumes.

The last line matters. Without it, Claude may fill blanks with numbers that look like data.

A worked example of three clusters from a project management software list:

For a repeatable structure, the topic cluster builder turns clusters into a hub-and-spoke plan.

Content briefs

With web search on, Claude can read what ranks now. Ask for a brief that covers what those pages miss, not a summary of them:

Target query: "[query]". Search the web and read the top 5 ranking pages.

Write a content brief:
- Search intent in one sentence, and the format that ranks (list, guide, tool).
- The questions every top page answers (must-cover).
- Questions or data none of them cover (our angle).
- Recommended H2 and H3 outline.
- Target length based on the ranking pages, not a guess.
- 5 internal link targets from our site list below, with anchor text.
- Title tag (under 60 characters) and meta description (under 155).

Here is our existing article on [topic] and the brief above.
List what to add, what to cut, and what to rewrite to beat the current
top result. Keep every section that already earns traffic (data below).

The second prompt is for refreshes, which usually move faster than new posts. Pair it with query data from Search Console, covered in Claude and Search Console.

On-page audits

Audit this page for the query "[query]". Page HTML is below.
Check: title tag and length, meta description and length, one H1 that
matches intent, heading structure, whether the first 100 words answer the
query, thin sections, missing subtopics competitors cover, image alt text,
internal links in and out, and schema present.
Return a table: issue, where it is, severity (high/medium/low), exact fix.

Claude reads copy and structure well. It cannot measure load speed or render JavaScript from pasted HTML, so treat technical findings from a paste as hints. A crawler or an audit tool has to confirm them. The technical SEO audit template lists the checks that need a tool.

Internal links

Here is our sitemap as a list of URLs with page titles, and a new
article (pasted below).
1. Find 5 to 8 places in the new article to link to existing pages.
   Quote the sentence, the anchor text and the target URL.
2. Find 5 existing pages that should link to the new article, with the
   anchor text and the paragraph where it fits.
Anchor text must be descriptive and must not repeat the exact same phrase
on more than two pages.

From this URL list, identify orphan-risk pages: pages whose topic
no other page in the list would naturally link to. Suggest a hub page
for each group of orphans.

Schema markup

Claude writes valid JSON-LD reliably. Give it the page content and the type:

Write JSON-LD for this page. Type: FAQPage plus Article.
Use only questions and answers that appear on the page, word for word.
Include headline, author, datePublished and dateModified from the page.
Return the JSON only, then list any required property you could not fill.

Always validate the output with Google's Rich Results Test before shipping. And match the markup to visible content: schema that describes text the reader cannot see is a common reason rich results get dropped.

Titles, metas and AI search

Here are 30 pages with their title tag, meta description, target query,
impressions and CTR from Search Console. Rewrite titles and descriptions
for the 10 pages with the most impressions and the lowest CTR. Keep the
target query near the start of each title. Under 60 and 155 characters.

For the query "[query]", write a 40 to 60 word answer paragraph that an
AI answer engine could quote directly: the answer first, one concrete
number or example, no hedging. Then the H2 it should sit under.

The second prompt helps pages get quoted in ChatGPT, Perplexity and AI Overviews. The longer playbook is in LLM SEO.

Where Claude stops

• No data of its own. Volumes, difficulty, rankings and backlinks must come from a tool or Search Console.
• Web search is not a crawler. It reads individual pages; it does not crawl your site or report status codes across it.
• It does not publish. Drafts end in the chat until something puts them in your CMS.

Analytics questions about organic traffic are covered in Claude and Google Analytics, and the full prompt library is in Claude marketing prompts.

Where Sprites fits

Sprites is an official MCP connector that gives Claude the SEO data and the publishing step it lacks. Through it, Claude gets keyword research with Keyword Planner volume checked against clickstream volume, difficulty, intent, SERP features including AI Overviews, and an opportunity ranking. It reads what a competitor ranks for that you do not, who ranks for a query, Search Console queries and URL inspection, and runs a 52-check page audit with measured Core Web Vitals or a site-wide crawl. Posts are written into WordPress, Webflow, Shopify or Framer as drafts with metadata and internal links set, and page edits arrive as a diff. Publishing is always a separate approval.

Setup in Claude is on the SEO connector setup page. Plans start at $99 a month, with a 7-day trial for $1.

Claude SEO questions

Is Claude good for SEO?

Claude is good at the reading and writing parts of SEO: keyword clustering, content briefs, on-page audits, internal link plans and schema markup. It has no search volumes, rankings or crawl data of its own, so pair it with exports or a connected data source for anything numeric.

Can Claude do keyword research?

Claude can cluster and prioritize keywords you give it, but numbers it produces from memory are not reliable. Export volumes and difficulty from a keyword tool, or connect a data source through an MCP connector, and tell Claude to use only the numbers provided.

What are the best Claude prompts for SEO?

The most useful are keyword clustering by intent, a content brief built from the top-ranking pages, an on-page audit against a target query, an internal link plan from your sitemap, and JSON-LD schema from the page content. Each works best with real data pasted in.

Can Claude write SEO content?

Yes. Claude writes strong long-form content when given a brief, the ranking pages to beat and your product facts. Check every statistic and claim before publishing, and give it your internal link targets so new pages connect to the rest of the site.

Can Claude write schema markup?

Yes. Claude writes valid JSON-LD for types like Article, FAQPage, Product and HowTo when given the page content. Ask it to use only content that is visible on the page, then validate the markup with Google's Rich Results Test before publishing.

Can Claude connect to Google Search Console?

Yes, through an MCP connector. Sprites connects Claude to Search Console so it can read queries, impressions, clicks, CTR and position, and inspect URLs for indexing status. Without a connector you can export the Performance report and upload it.