An SEO roadmap for AI-built websites.
You generated a site, deployed it, and nothing happened. That is normal, and it is also fixable — but only in order. Working on stage five while stage two is broken is the most common way to waste a month.
Seven stages, in order
Each stage has an exit condition. Until it is met, the stages after it cannot be honestly evaluated — so they stay NOT EVALUATED rather than being scored on assumptions.
Site live
A real public URL returns a real HTML page. This sounds trivial and is where a surprising number of AI-built sites fail: the deployment succeeded, but the URL a visitor would use returns a 404, a redirect loop, or an empty shell.
What can be checked automatically
- Protocol, final HTTP status, and the full redirect chain
- Whether the response is HTML, and what the final URL actually is
- Whether the www and non-www versions both serve a full page
What still needs outside evidence
- Whether the CDN or the hosting platform is serving the content you think it is from another region
Exit condition
One public URL returns 200 HTML, and the alternate host does not serve a second full copy.
Crawlable & indexable
Search engines are allowed in, and the site declares one preferred address. This is where most "I published it and nothing happened" cases actually live.
What can be checked automatically
- robots.txt: reachable, and whether a site-wide
Disallow: /is blocking everything meta robotsand theX-Robots-Tagresponse header, including accidentalnoindex- The canonical URL: present, absolute, HTTPS, and whether it agrees with the live URL
- The canonical target: does it actually return 200
- The sitemap: reachable, valid XML, index or urlset, and which host it lists
- Whether the sitemap, the canonical and the live page agree on one host
What still needs outside evidence
- Whether Google has actually indexed the page. That needs Search Console.
Exit condition
Crawling is allowed, nothing is marked noindex, and the canonical and sitemap both point at the same preferred URL.
Page structure
The served HTML carries the metadata a search engine reads: a title, a description, one H1, a language, a viewport, share tags and structured data.
What can be checked automatically
- Title presence and length, meta description presence and length
- H1 count and heading structure,
lang, viewport - Open Graph tags and JSON-LD presence
- Internal link count, malformed
hrefvalues, and imagealtcoverage
What still needs outside evidence
- Which page should own which query, and whether two pages are competing for the same intent
- Whether the structured data is accurate rather than decorative
Exit condition
Every page you want indexed has a unique title, a description, one H1, and valid structural signals.
Search demand
Someone actually searches for this, and your page matches what they wanted. No amount of technical correctness fixes a page that answers a question nobody asks.
What can be checked automatically
- Nothing, honestly. This needs a keyword data source and an actual SERP.
What still needs outside evidence
- Search volume and how competitive the query is
- What kind of page currently wins: a tool, a landing page, a comparison, or an article
- Whether sites your size rank for it at all
Exit condition
One named query, with real demand, owned by one page whose format matches the current results.
Content readiness
The content exists in the HTML the server returns, and it is worth reading. Client-rendered pages that ship an empty <div id="root"> depend on everything going right further down the pipeline.
What can be checked automatically
- How much readable text the server-returned HTML actually contains
- Whether the page relies on scripts to show its own content
- Heading structure and media accessibility
What still needs outside evidence
- Whether the content is more useful than what already ranks
- Whether the claims are true, sourced, and written by someone who knows the subject
Exit condition
Disable JavaScript and the page still shows its content, its headings and its links.
Publish & verify
This is the stage almost nobody does. Your agent says it fixed the canonical. The build passed. The deploy succeeded. None of that proves the live URL changed.
What can be checked automatically
- Whether the specific problem reported before the fix is still present in the live response
- Whether the fix introduced a new problem somewhere earlier in the sequence
- Whether the declared state — canonical, sitemap, robots — still agrees with itself
What still needs outside evidence
- Nothing. This stage is fully deterministic.
Exit condition
The previously failing check passes when the production URL is read again, not when someone says it is fixed.
Measure & improve
Once the page is published correctly, the question changes from "is this broken" to "is this working". That needs Search Console: impressions, clicks, queries, and which pages earn them.
What can be checked automatically
- Nothing yet. TDKSEO does not connect to Search Console in this version.
What still needs outside evidence
- Clicks, impressions, CTR and average position
- Which queries map to which page, and where you rank 5–15 or 11–20
- Pages with impressions but almost no clicks
Exit condition
You can name the queries the page is appearing for and decide what to improve next based on real numbers.
What to do with this
Do not read the seven stages as a checklist to complete in one afternoon. Read them as a filter. When you are unsure what to work on, run the live check on the page you care about and work on the earliest failing stage. Everything after it is a guess until that is resolved.
TDKSEO checks stages one, two and three automatically, reports what it can prove, and deliberately leaves search demand, content quality, rendering fidelity and Search Console data as UNKNOWN until you bring real evidence. That is the whole point: a smaller set of true statements beats a large set of confident ones.