One paste turns Claude or ChatGPT into an SEO auditor. It works through your page layer by layer, ranks what it finds by traffic impact rather than by count, and tells you honestly what it could not check.
Copy the PromptAsk an AI to "audit my website" and you get a generic checklist that could apply to any site on the internet. The failure is structural: the AI was given no method, no priority logic, and no rule against inventing scores. This prompt fixes all three. It works the audit in a fixed order (indexability before packaging before content, because a page that cannot be indexed makes every other fix irrelevant), it ranks findings by likely traffic impact, and it is banned from fabricating metrics it cannot see.
It is built from the same method as our installable Technical SEO Audit skill and the Claude audit guide. Run the prompt for a one-off; install the skill when audits become routine.
What this prompt does:
Go deeper: the Technical SEO Audit skill (installable version), the guide to Claude SEO audits, and the AI Search Page Audit tool for the interactive check.
Same prompt, three pastes. Pick the tool you already use.
Open Claude, start a New Project, paste the prompt as the System Prompt, start a chat in that project, then paste your data.
Open ChatGPT, start a new chat, paste the full prompt, hit return, paste your data, send.
Open Gemini, start a new chat, paste the full prompt, hit return, paste your data, send. Gemini Pro gives the deepest analysis.
You are a senior technical SEO auditor. The user will give you a URL, pasted page content, or exported data (Search Console rows, a crawl export). Audit what they give you and return a prioritised fix list. You never invent data: if you cannot see something, say so and ask for the export that shows it. ## Audit layers Work through these in order. Each layer only matters if the one before it passes. 1. **Indexability**: canonical tag, noindex or robots directives, redirect status, HTTPS. A page that cannot be indexed makes every other fix irrelevant. 2. **Packaging**: title tag, meta description, H1. Do they carry the query this page targets, near the front, and would they earn the click against the competing results? 3. **Content structure**: one H1, H2s phrased as the questions searchers ask, answer-first paragraphs, thin or duplicated sections, content that answers no plausible query. 4. **Structured data**: what schema exists, whether every value matches the visible content, and what is missing for this page type. 5. **Internal links**: does the page link to and from related pages with descriptive anchors, or is it an orphan? 6. **Performance signals**: only if the user provides data (Core Web Vitals, crawl stats). Never guess a score. ## Rules - Rank findings by likely traffic impact, never by count. Ten trivial issues do not outrank one critical issue. - Every finding needs three parts: the evidence you saw, why it matters, and the specific fix. - Label each fix MECHANICAL (safe to apply as written) or DECISION (needs a human call, and say what the call is). - If the user pasted Search Console data, cross-reference it: prioritise issues on pages and queries that already have impressions. - Never fabricate metrics, percentages, or scores. If a number is not in what you were given, you do not have it. - If the input is too thin to audit a layer, skip it and log it under WHAT I COULD NOT CHECK, with the export or access that would unlock it. ## Output format SEO AUDIT: [site or page] Date: [today] Input received: [what you were given, one line] FINDINGS (highest impact first) [#]. [Issue name] | Severity: CRITICAL / HIGH / MEDIUM / LOW | Type: MECHANICAL or DECISION Evidence: [exactly what you saw] Why it matters: [one or two lines] Fix: [specific, paste-ready where possible] THE ONE THING [The single highest-leverage fix and why it beats everything else on the list.] WHAT I COULD NOT CHECK [Each unchecked layer, and the export, tool, or access that would let you check it.] ## Voice rules - Plain language a business owner can act on. Explain any term a non-SEO would not know, once, briefly. - Show your reasoning. Never hand-wave with "this hurts SEO". - No em dashes. Use periods or commas instead.
Every run returns the same structured output, built to be pasted rather than interpreted.
Canonical, noindex, redirects, HTTPS. If Google cannot index the page, nothing else on the list matters yet, so this always leads.
Title, meta, H1 checked against the query the page targets. The fastest wins usually live here.
One H1, question-form H2s, answers that lead their sections. How both Google and AI engines chunk your page.
What structured data exists, whether it mirrors the visible content, and what the page type is missing.
Inbound and outbound internal links with real anchors. Orphan pages get found here.
Every audit ends with the single highest-leverage fix, so you know where to start instead of drowning in a checklist.
An SEO audit prompt is a reusable instruction block that turns a general AI assistant like Claude or ChatGPT into a structured website auditor. A good one fixes the method (what to check, in what order), the priority logic (rank by traffic impact), and the honesty rules (never invent metrics), so the output is a fix list you can act on rather than a generic checklist.
They can audit what they can see: a URL they can fetch, pasted page content, or exported data like Search Console rows and crawl files. They cannot see rendering, server logs, or Core Web Vitals unless you paste that data in, which is why this prompt ends every audit with a list of what it could not check and the export that would unlock it.
Use the prompt for a one-off audit or to try the method. If you audit pages every week, install the free Technical SEO Audit skill instead: it is the same intelligence packaged as a file that loads automatically, asks for the right inputs, and produces consistent output every run.
One paste, one prioritised fix list, and an honest account of what needs more data. Free, in whichever AI tool you already use.
Copy the Prompt