Start with one useful page rather than an audit of your entire website. Your goal is to help a visitor complete a specific task and remove concrete obstacles to finding or understanding the page.
The output is a short review with evidence, proposed changes, and a verification result. There is no proprietary SEO score, invented search volume, or ranking guarantee.
Prepare your inputs
Choose one published public URL you control, the audience it serves, and the question it should answer. Have permission to edit before making changes. You need a browser and access to your normal editor or deployment workflow; no paid SEO API is required.
Use this fictional example to understand the exercise: a page explaining how a first-time gardener starts basil seeds indoors. It is an example, not a validated keyword opportunity.
Write one sentence before opening the page: “A beginner should leave knowing what to prepare, what to do, and how to recognize a problem.” Substitute your actual reader and outcome.
1. Check the prerequisite: the right page is reachable
Open the URL while signed out. Confirm the final address, readable main content, and absence of an unexpected error or access gate. Use the crawlability walkthrough if anything is unclear. Record “not checked” for indexing if you have no account evidence; public access alone does not establish index status.
Do not remove a login from intentionally private material. Select a public page for this exercise instead.
2. Read the page as a first-time visitor
Read the opening paragraph without consulting your product knowledge. Can you tell who the page is for and what it helps you do? Then try to carry out the instructions or locate the promised answer.
For the fictional gardening page, a useful review might notice that “prepare your tray” never explains what container is suitable. Write the missing question in your review. Do not pad the article to reach a target word count.
Check whether examples are genuinely yours, factual statements have appropriate support, and dated instructions still match the product or process described. Google's SEO Starter Guide emphasizes useful, original, well-organized content. Use headings to expose the order of the answer rather than repeating a phrase in every heading.
3. Inspect the title and description
Look at the browser tab, then View Page Source and find <title>. Is it specific to this page and consistent with the visible main heading? A fictional title such as “Start Basil Seeds Indoors: A Beginner's Setup” is more informative than “Home” or a chain of repeated keywords. The promise must match the actual content.
Find <meta name="description"> in the head. If present, read it as a short invitation to the correct reader. If missing, draft one concise, page-specific summary for your editor. For the fictional page: “Prepare a small indoor seed tray, follow the sowing steps, and learn what to check when basil seedlings struggle.”
A description does not force the snippet shown in search. Google may choose page text and vary snippets by query; it does not prescribe a universal character count that guarantees an untruncated result. See Google's snippet guidance.
Record the existing and proposed text separately. Do not describe your proposal as deployed until you have checked the live page.
4. Check the canonical address
In the HTML head, look for a link whose rel is canonical. Compare its target to the intended public page. If it points at staging, another article, or an unexpected domain, stop and investigate the template before editing broadly.
For a unique page, a self-referencing canonical is a sensible implementation. If multiple URLs show equivalent content, choose a preferred version and align internal links and the sitemap with it. Canonical signals express a preference; they do not prove which URL Google selected. See Google's canonical guidance.
5. Follow links and review images
From the homepage or a relevant category page, try reaching your article through visible links. On the article, click the main internal links and any sources supporting important claims. Note broken or misleading destinations.
Use meaningful link text: “seedling watering guide” is clearer than a row of identical “click here” links. Standard HTML links with an href are the dependable crawlable pattern; avoid making navigation depend solely on click handlers. Google's link guidance explains the pattern.
For each informative image, ask what a person needs to understand if they cannot see it. Write concise alt text that conveys that purpose, without stuffing search terms. Purely decorative images generally need empty alt text. Keep captions and surrounding prose relevant. See Google's image guidance.
Finally, narrow the browser to a phone-sized view. Can you read the instructions, operate navigation, and dismiss overlays? Record actual usability problems rather than treating a resized window as proof of performance on every device.
6. Make one small change and verify it
Choose the highest-impact observable problem: an access failure before wording polish, a wrong canonical before extra metadata, or a missing answer before decorative additions. Save the original and apply only changes you are authorized to make.
After publication, reload the live page. Check the visible content and the served HTML, not just your editor's preview. Retest affected links and the phone-sized layout. If available, an authorized owner can run Google or Bing URL Inspection; account login and site permissions are required. A Skill has no automatic access to those reports.
Page and reader task:
Review time:
Finding and evidence:
Why it affects this reader:
Proposed change:
Published change and time, or not published:
Live verification and remaining issue:
Indexing / traffic observations, or not checked:
Worked record: conflicting canonical tags
Synthetic example. Not a live result. The canonical-conflict teaching example provides invented HTML for https://example.com/trail-guide/ plus a matching sitemap exercise. This example focuses on one defect; it is not a complete SEO audit or a claim about a real search index.
Page and reader task: https://example.com/trail-guide/; read one public trail guide
Review time: Not live-tested; synthetic teaching record
Finding: Two canonical tags disagree
Evidence: One points to /trail-guide/ and one to /trails/ on example.com
Why it matters: The page sends contradictory representative-URL preferences
Proposed change: Keep only https://example.com/trail-guide/ after owner confirmation;
keep the sitemap aligned with that intended address
Published change and time: Not published; corrected teaching fixture only
Live verification: Not checked
Remaining issue: Owner must verify the actual target, deployment, and rendered HTML
Indexing / traffic observations: Not checked / Not checked
Interpret every field:
- Page and reader task define what is in scope. Other pages have not been audited.
- Review time avoids pretending that an invented record is a time-stamped production test.
- Finding states a specific defect, not an unsupported score or diagnosis of lost traffic.
- Evidence identifies the conflicting destinations. Inspect the complete absolute URLs in the fixture rather than relying on the abbreviated record alone.
- Why it matters explains the signal inconsistency without claiming which representative a search engine chose.
- Proposed change is conditional on the owner's intended preferred URL; a tool should not choose a different destination silently.
- Published change and time distinguish the corrected example file from a live release.
- Live verification remains unknown until someone checks the deployed page.
- Remaining issue supplies a clear handoff and the evidence needed to close it.
- Indexing / traffic observations remain unchecked. Removing the duplicate tag alone cannot establish either outcome.
If the page is inaccessible, begin with the crawlability walkthrough. For inventory consistency, use the sitemap walkthrough.
If your result looks wrong
- Live title unchanged: Check deployment completion, caching, and whether you edited the template used by this route.
- Two conflicting tags: Identify the CMS, plugin, or template producing each before removing anything.
- Google shows different text: Check its crawl date and remember that titles and snippets can be generated from multiple page signals.
- A green checklist produces no traffic: Technical and editorial improvements do not establish demand, indexing, or competitiveness. Keep the verified improvement; do not invent a growth claim.
You are finished when the selected change is demonstrably live, or the review names a specific unresolved dependency. A small, reproducible improvement is a more useful result than an unsupported “100/100” score.
Optional matching Skills
You can complete this guide manually. If you already use a compatible assistant, inspect SEO Skill: Local CLI before using it. Source reviewed CLI run · synthetic · Reviewed 2026-10-10.