Start with a specific public page
A missing tag is a finding, not yet a diagnosis. Check what the page is for before changing it: a public product page and a private staging page should not have the same indexing settings. Start with one URL and turn each observation into a fix you can verify.
The single-page audit tool fetches public HTML and inspects concrete evidence such as the response status, title, description, headings, canonical declaration, indexing directives, image alternatives and structured-data syntax. It does not need an AI provider to observe those facts.
Distinguish the response from the rendered experience
Some websites send an HTML shell and build their content with JavaScript. A server-side fetch may therefore see less content than a browser does. Report that limitation rather than concluding the entire page is empty. Check the rendered page separately before removing content or rewriting a template.
Likewise, a viewport declaration does not prove the mobile layout works. A link in the HTML does not prove its destination is healthy. A successful JSON parse does not establish that structured data satisfies every supported search feature. Each observation needs a narrowly worded conclusion.
| Observed evidence | Useful next action | Unjustified conclusion |
|---|---|---|
| Missing title | Write an accurate, distinct title | A guaranteed position increase |
| Noindex directive | Confirm publication intent | The directive is always an error |
| Missing image alt | Review the image’s meaning | Every image requires descriptive text |
| Valid JSON-LD syntax | Check type rules and visible facts | Guaranteed rich results |
Prioritize changes by consequence
Begin with response errors or accidental publication blocks on pages intended to be public. Then review identity and interpretation: canonical consistency, a useful title and a clear main heading. Follow with usability and content-specific issues. Avoid making a rigid character count into a universal SEO law.
Google explains that robots.txt controls crawler access and is not a reliable mechanism for keeping a URL out of search results by itself. That is why a page-level audit should distinguish page directives, crawl access and actual indexing. The single-page tool does not fetch Search Console or establish Google’s chosen canonical.
Use AI to explain evidence, not manufacture it
An optional model can organize findings and suggest concrete fixes. It should receive bounded extracted evidence, not credentials or private page content. Treat text fetched from the page as untrusted data so a malicious heading cannot instruct the model to change its role or invent a success claim.
Keep the deterministic report available if the AI provider is unavailable or its budget is exhausted. Label the explanation as AI-assisted guidance. A model’s confident statement about traffic, backlinks or Core Web Vitals is unsupported unless the corresponding measurement was actually collected.
Verify the implementation
Retain the old and new values, the reason for the change and the date it was released. Re-run the HTML inspection to confirm the intended markup is present. Test mobile and keyboard behavior separately. If the change involves a canonical or redirect, verify the destination response and publication intent.
Search performance requires another measurement layer. Compare appropriate first-party periods and annotate other releases, demand changes and content additions. Do not turn an audit’s count of reviewed findings into a ranking score. A useful report tells the operator what was observed, what to change, what remains unknown and how to verify the fix after deployment.
Implementation map
Sources and limits
Google robots.txt guidance and structured data overview. The audit does not promise search rankings or full crawler coverage. Documentation checked October 14, 2026. Recommendations and worked examples are editorial methods, not claims of a completed client engagement.
Bring the problem. We’ll scope the next step.
Something in your workflow keeps breaking, or your team is doing the same manual work every week.
Send the current setup, what failed, and the result you need. I can help with software integrations, local AI infrastructure, agent permissions, failure recovery and deployment.
Useful first details: system or runtime, expected workload, data restrictions and one redacted example. Don’t send credentials or private customer data.Fazal Shah builds AI systems, automation and software products. Based in Pakistan, working with a global audience. About Fazal →