Search “on-page SEO audit with Claude” and you find a wall of skills, prompt packs, and GitHub repos. Each one promises Claude will scan your page and hand back a tidy report. Paste a URL, get a score, fix the red items. It looks like a real audit and it reads like one too.
Most of them are a spell-check for your markup wearing an audit’s clothes.
Here is the part the prompt packs skip. Claude is genuinely good at the on-page work, the reading and the judgment. It can tell you your title is too long, your H1 is doubled up, your alt text is missing, your schema is broken. That is useful. But none of it answers the only question that matters, which is whether the page is good enough to rank for the query it targets. You cannot answer that by looking at the page alone. You answer it by looking at the ten pages already beating it, and at your own real numbers.
This guide is the honest version. What Claude audits well on its own, where the blind audit quietly fails, the one step that turns it into a real audit, and the exact workflow to run it. No Ahrefs subscription required, and no pretending a linter is a strategy.
- Claude can run the on-page half of an SEO audit for free. Paste a page and it checks title, headings, depth, links, alt text, and schema, then drafts the fixes.
- A blind audit is a linter, not an audit. It tells you the page is tidy, not whether it can rank.
- The real audit compares your page to what ranks. That needs live SERP data, which Claude gets through an SEO MCP server.
- Your own truth comes from Search Console. Connect GSC and Claude reads your real impressions, position, and clicks for the page.
- You do not need Ahrefs. Free tools plus a flat SEO data feed cover everything an on-page audit needs.
What an on-page SEO audit actually checks
An on-page audit looks at one page and asks a simple question. Does this page serve its target query better than the pages it is competing with?
That breaks into two layers. The first is the page itself. Title, meta, headings, content, links, images, and markup. The second is context. Who else ranks for this query, what they cover, and how your page is actually doing in search right now.
The first layer is a checklist. The second layer is the audit. Most “Claude SEO audit” tools only do the first and call it done. The 2026 on-page checklist itself is well documented across the SEO field, covering title links, headings, internal links, images, schema, and indexing signals. The checklist is not the hard part. Knowing whether your page clears the bar for its query is.
What Claude can audit on its own
Hand Claude a page, even just the raw HTML or the visible text, and it does the first layer well. This part is real, and it is free.
It reads the title and meta. It flags a title over about sixty characters, one that buries the keyword, or one that does not match the page. Worth knowing: Google rewrites title tags often. An Ahrefs study of 953,276 pages found Google rewrote the title 33.4% of the time, and long titles were far more likely to get swapped. Claude is good at writing a title Google will keep.
It checks heading structure. One H1, matched to intent, then a logical H2 and H3 outline with no skipped levels. It spots the page with three H1s or the one whose headings describe nothing.
It reads content depth and intent. Paste the page and ask whether it answers the query a searcher actually has. Claude is strong here. It will tell you the post is thin, off-intent, or padded.
It audits the mechanical bits. Missing or keyword-stuffed alt text, absent or invalid schema, a vague URL, no internal links out of the page. These are judgment calls dressed as rules, and Claude makes them quickly.
A first pass is one prompt. “Audit this page for on-page SEO. Cover title, meta, H1 and heading structure, content depth against the target query, internal links, image alt text, and schema. Give me a prioritized fix list.” You get a clean report in seconds.
So far this matches every skill on the market. Here is where they stop and the real audit starts.
Why the blind audit is not really an audit
A blind audit can give a page a perfect score and the page can still never rank.
Picture a post with a clean title, one tidy H1, good structure, valid schema, internal links, and alt text on every image. Claude scores it green across the board. The problem is the query it targets has ten results that each run three thousand words, cite original data, and sit on domains with hundreds of links. Your tidy page is a knife at a gunfight. Nothing on the checklist told you that, because the answer was never on your page.
Run it the other way too. A page that looks thin against the checklist might be the exact right depth for its query, because the top ten are all short and direct and the searcher wants a quick answer. The linter says “add more content.” The SERP says “do not.” Only one of them is right, and it is not the linter.
This is the take. An on-page audit done blind optimizes the page against an imaginary standard instead of the real competition. It makes your page tidy. Tidy does not rank. Better-than-the-top-ten ranks. To know what better means, Claude has to see the top ten and your own numbers. Give it neither and you are grading homework with no answer key.
The step that makes it a real audit
The fix is to stop letting Claude guess and give it two live data sources. They do different jobs.
The SERP, through an SEO MCP server. The Model Context Protocol is the open standard Anthropic introduced for letting Claude call external tools mid-chat. An SEO MCP server hands Claude tools for SERP results, keyword volume, and difficulty. Once connected, “who ranks for this query and what do they cover” becomes a real lookup, not a hallucination. Now the audit can compare your page to the pages actually beating it. The full explainer is in what is an SEO MCP server.
Your own numbers, through Search Console. Estimates are third-party guesses. Google Search Console is the one source of your page’s real data, your true impressions, average position, clicks, and indexing status, straight from Google. Connect it and Claude can read what the page is genuinely doing instead of guessing.
Together they change the audit completely. The SERP tells Claude what the page is up against. Search Console tells Claude where the page already stands. An audit with both is the difference between “your title could be shorter” and “you sit at position 11 for this query with 1,400 impressions and almost no clicks, the top ten all answer the comparison angle you skipped, so add that section and rewrite the title and you have a page-one shot.”
How to run an on-page SEO audit with Claude, step by step
Once Claude has data, the audit is a short conversation. Five moves.
Step one. Hand Claude the page. Give it the URL or paste the content, and name the query the page targets. Ask for the first-layer audit, the on-page checklist from earlier. This gets the mechanical fixes out of the way.
Step two. Pull the SERP for the target query. Ask Claude to fetch the top ten results, summarize each one’s angle and rough depth, and tell you what they all cover that your page does not. That gap list is the most valuable output of the whole audit. It is your content brief, drawn from the live competition rather than a guess about it.
Step three. Read your Search Console for the page. Ask “how is this page actually doing in search.” Claude pulls your real impressions, average position, and the queries you already show up for. Often you discover the page ranks for a query you were not even targeting, which changes the whole brief. The same data surfaces quick wins, pages sitting on page two that a single edit can lift.
Step four. Sanity-check the target is winnable. If the page is fighting for a brutally hard keyword, no on-page edit will save it. Ask Claude for the difficulty and read it honestly. Our guide on checking keyword difficulty without Ahrefs covers how to read those scores without overtrusting them. Sometimes the real fix is to re-aim the page at an easier query.
Step five. Get the prioritized fix list. Ask Claude to combine all of it into one ordered list, highest impact first, with the actual rewrites drafted. Title rewrite, the missing section, the internal links to add, the schema to fix. You leave the chat with edits, not a score.
The whole loop takes about ten minutes per page once the data sources are wired up. The first four steps are reading and lookups. The fifth is Claude doing the thinking you would otherwise do by hand across five browser tabs.
On-page SEO audit checklist
Here is the full checklist, and which parts Claude can clear on its own versus the parts that need live data. The split is the whole point.
| Audit item | What to check | Claude on its own? |
|---|---|---|
| Title tag | Unique, under ~60 chars, keyword near the front | Yes |
| Meta description | 120 to 155 chars, reads like a reason to click | Yes |
| H1 and headings | One H1, intent-matched, logical H2 and H3 order | Yes |
| Image alt text | Descriptive, not stuffed, present on key images | Yes |
| Schema markup | Valid, matches the page type | Yes |
| Internal links | Crawlable, relevant, enough of them | Partly, sees page only |
| Content depth | Matches what the query actually needs | Partly, needs the SERP |
| Competitive gap | What the top 10 cover that you miss | No, needs live SERP |
| Target is winnable | Difficulty is realistic for your domain | No, needs live data |
| Real performance | Impressions, position, and clicks for the page | No, needs your GSC |
Read the right-hand column top to bottom. Everything marked yes is the linter every skill already does. Everything marked no is the audit. A tool that only does the green rows is selling you the easy half.
What this audit still will not catch
An honest read of the limits, because nobody else lists them.
It is on-page, not technical. This audit looks at one page against one query. It does not crawl your whole site, check your sitemap, find redirect chains, or measure load speed. Core Web Vitals, including INP, which replaced FID as a Core Web Vital in March 2024, are a separate technical pass. Run the technical audit first, then the on-page one page by page.
Estimates are not Google’s ledger. Live SEO data is decision-useful and directionally right, not exact. Your Search Console is the exception, because that data is genuinely yours. Treat the SERP and difficulty numbers as a strong signal, not gospel.
Claude can be confidently wrong about strategy. It is excellent at pulling the data and drafting the fixes. It is not infallible about what to do with them. The fix list is a first draft. You make the calls.
Backlinks are a different job. On-page work cannot overcome a huge authority gap. If the top ten all sit on far stronger domains, the honest answer is sometimes “this query is not winnable yet,” and an audit should say so.
Running the audit with ContextBolt SEO
Full disclosure, since you are reading this on the ContextBolt blog. I build one of these.
ContextBolt SEO is a hosted SEO MCP server made for exactly this workflow. You subscribe, you get one MCP URL, and you paste it into Claude Desktop, Claude Code, Cursor, or Codex. From then on Claude can pull the SERP, keyword difficulty, and competitor data inside any chat, with no DataForSEO account to register, no deposit, and no credentials to manage. It is $35 a month for 1,000 lookups at launch pricing, which is a quarter of an Ahrefs plan and the data covers what an on-page audit needs.
The free Google Search Console connection sits right alongside it, so in one conversation Claude reads both your estimates and your real numbers. Two things also make it fit the audit loop. It remembers every lookup across sessions, so when you re-audit the page next month the answer leads with what changed, at no extra credit. And it saves each finding to a ./seo-findings/ folder in your project as markdown, so the audit trail lives where you work. Both run automatically and neither counts against your lookups.
If you would rather not pay for data at all, you still can. Read the SERP by hand, lean on free checkers, and use your Search Console, which is the only paid-grade data that is free. Our Ahrefs alternatives round-up covers the cheaper picks. The audit method is the same either way.
The real version of an on-page SEO audit with Claude is not a score out of 100. It is Claude reading your page, reading the competition, reading your own numbers, and handing you an ordered list of edits that move the page up. Wire up the data, ask good questions, and the audit does itself.