ChatGPT SEO Audit Prompts: Copyable HTML Audit Templates
ChatGPT SEO audit prompts work best as focused instructions applied to evidence you supply. These five templates cover metadata, headings, structured data, internal links and images. Copy a prompt, replace its placeholders, paste your source HTML, then verify each finding against the source before implementing it.
- Start with the page URL, intended search task and actual HTML, rather than assuming the model fetched the page.
- Require an evidence excerpt, a practical fix and a separate list of checks the input cannot support.
- Use the worked example below to distinguish observable defects from unsupported claims about rankings, indexing or speed.
Before You Run a Prompt
- Copy the relevant source HTML, including the head and main content. Remove personal data, session tokens, API keys and private URLs.
- Provide the public page URL, target reader and primary task. For link analysis, also supply the destination-page inventory.
- If the page relies on JavaScript, supply rendered HTML separately and label it. Source and rendered documents may differ.
- Save the model output beside the input and review the evidence. Do not deploy generated facts, URLs or JSON-LD without checking them.
Five Copyable SEO Audit Prompts
Prompt 1: Metadata and Title Review
Review only the HTML and context below.
Page URL: [PUBLIC PAGE URL]
Search task: [WHAT THE READER NEEDS TO DO]
Primary keyword: [KEYWORD]
Extract the title, meta description, canonical and Open Graph fields.
Return a table: Finding | Source excerpt | Current value and character count | Suggested fix | Verification needed.
Flag missing or empty values, contradictory URLs and unclear wording.
Suggest wording that reflects the actual content.
Character counts are editorial guidance, not Google's fixed display limits. Do not claim a title will definitely truncate.
Do not infer HTTP status, Google-selected canonical, indexability or rankings from this HTML alone.
If evidence is missing, write "Not checked".
[PASTE HTML]A title's character count is useful for editing, but Google does not publish a fixed character limit that guarantees display. See Google's title-link guidance; final titles can also be rewritten.
Prompt 2: Heading Structure and Task Completion
Extract H1-H6 from the supplied HTML in document order.
Reader and task: [READER AND TASK]
Return the heading tree and a table: Section | Evidence | Missing answer or structural concern | Proposed edit.
Review whether the main heading identifies the page's purpose and whether section headings form a usable hierarchy.
Treat skipped levels or multiple H1s as review items, not automatic ranking penalties.
Use word counts to describe coverage only; do not invent a minimum word count or require padding short sections.
Identify questions the supplied content leaves unanswered. Separate observation from inference.
[PASTE HTML]Prompt 3: Structured Data Review
Inspect structured data in the supplied HTML: JSON-LD, Microdata and RDFa.
Return: Type | Source evidence | Syntax or content mismatch | Proposed fix | Validation step.
Check that names, URLs, dates and other facts match the visible content.
Check that @id references are consistent and distinct entities do not accidentally share an identifier.
Do not invent reviews, ratings, prices, authors or business details.
If recommending a Google Search feature, cite its current official documentation; if you cannot verify eligibility, mark it "Needs documentation check".
Do not promise a rich result from valid markup.
[PASTE HTML]Validate any proposed JSON-LD independently. Our complete @id graph example explains references; sameAs examples explain identity URLs.
Prompt 4: Internal Link and Anchor Review
Page URL: [PUBLIC PAGE URL]
Destination inventory: [URL AND TOPIC FOR EACH KNOWN INTERNAL PAGE]
Extract relative and same-origin links from this HTML.
Return: Anchor text | Resolved destination | Source excerpt | Concern | Suggested anchor.
Flag generic text such as "click here" where a descriptive replacement helps the reader.
Use the destination inventory; do not invent a destination topic from the URL alone.
Do not label pages orphaned or destinations broken without a site graph or HTTP evidence.
Keep good contextual links and recommend relevant additions only to known pages.
[PASTE HTML]Prompt 5: Image Markup Review
Review img elements in the supplied HTML.
Return: Image URL | Source excerpt | Observable concern | Suggested fix | Evidence still needed.
Check alt presence, whether context indicates decorative or informative use, and width/height attributes.
An empty alt can be correct for a decorative image. Do not invent visual details if the image was not supplied.
Treat missing dimensions as a layout-stability review item, not proof of measured CLS.
Do not infer above/below-fold position from source order alone or recommend lazy-loading every image.
Do not claim a measured LCP, CLS, INP, compression ratio or file size without measurements.
[PASTE HTML AND OPTIONAL IMAGES]Worked Example: Input and a Human-Checked Review
This synthetic page intentionally contains an empty title, a heading-level jump and a generic anchor. It is a teaching fixture, not a real client case or a claim about a particular model run. Page URL: https://example.com/services. The destination inventory says /delivery explains bakery delivery in Portland.
<!doctype html>
<html lang="en">
<head>
<title></title>
<meta name="description" content="Order bakery delivery in Portland.">
<link rel="canonical" href="https://example.com/services">
</head>
<body>
<h1>Bakery delivery in Portland</h1>
<h3>Delivery areas</h3>
<p>See our service area and delivery times.</p>
<a href="/delivery">Click here</a>
<img src="/delivery.jpg" alt="Delivery van" width="1200" height="800">
</body>
</html>| Review output | Human check | Action |
|---|---|---|
| Empty title; 0 characters | The title element contains no text | Use a descriptive title, such as Bakery Delivery in Portland | Sunrise Bakery |
| H1 is followed by H3 | The supplied tree skips H2 | Review hierarchy; H2 is appropriate here for Delivery areas |
| Generic anchor: Click here | Its href is /delivery and the supplied inventory identifies the topic | Use Bakery delivery areas and times |
| Image dimensions are present | Width 1200 and height 800 are visible in the source | Do not flag missing dimensions; inspect the actual image to assess alt accuracy |
| Indexing, page speed, broken links and rankings | No fetch, measurement, site graph or Search Console data was supplied | Not checked; collect the relevant evidence |
The first three findings can be reproduced from the fixture. A response claiming that Google cannot index this page, its LCP is slow, or its destination is broken goes beyond the evidence. This is the test to apply to your own model output.
Chain the Prompts into a Reviewable Audit
- Run the heading review to establish the page's task and structure.
- Review metadata, links and images against that same task.
- Review structured data against visible facts and the intended Search feature.
- Combine duplicate findings and assign an owner, implementation step and verification method.
Combine the verified findings below into an implementation checklist.
Columns: Finding | Source evidence | Proposed change | Owner | Verification method.
Deduplicate overlapping findings. Preserve "Not checked" items separately.
Do not promote suggestions or missing evidence into confirmed defects.
[PASTE THE INPUT CONTEXT AND VERIFIED FINDINGS]For a client-facing deliverable, follow our SEO audit report template. Keep supplied-page observations separate from crawl evidence and AI readiness checks.
Checks That Need Additional Evidence
A single HTML document does not establish redirect chains, TLS health, measured Core Web Vitals, Google index status, cross-page duplicates or orphan pages. Those need HTTP responses, browser measurements, Search Console, logs or a site crawl. A tool-enabled workflow can collect additional evidence, but the report must name what was actually inspected.
See what ChatGPT can contribute to an SEO audit, then view the illustrative MendMySEO demo for a report format. The demo does not establish production availability; check the current release status.
Frequently Asked Questions
Can I give ChatGPT just a URL?
Supply the source HTML when you need a reproducible on-page review. If the workflow fetches the URL, record what it fetched and which tools were used. Do not assume a link alone provided the complete head, rendered content, HTTP headers or all related pages.
Which prompt should I run first?
Start with the heading and task review, or the metadata prompt for a specific title issue. A focused input makes the result easier to verify than an unrestricted request to find every SEO issue.
Can these prompts replace a crawler?
They can assist a supplied-HTML review. A crawler adds multi-page fetching and relationship evidence. Use each result for the question its evidence can support, then verify proposed changes.
How do I check whether the output is reliable?
Find the quoted evidence in your input, reproduce the count or relationship, and confirm that the proposed change preserves the page's meaning. Check structured data with a validator and use live measurements for claims about speed, redirects or indexing.
Do I need a particular ChatGPT model?
The templates specify the input and output rather than a model version. Context limits and responses vary, so split large documents into clearly labeled sections when needed and verify every finding. The worked fixture provides a small check before applying the workflow to your site.