GEO on a brand-new domain: my landing page was invisible to AI search

The short version: if you want AI search engines to cite you, every page needs its full content on first render. To AI crawlers, a JavaScript-rendered single-page app is close to a blank page.

Context

I'm building Verbatim, a search engine over real podcast conversations. The domain was brand new: zero backlinks, no history, no authority.

I set myself a goal: go through the whole SEO and GEO (Generative Engine Optimization, getting your content cited by AI search like ChatGPT and Perplexity) cold start from zero, and treat every step as an experiment: hypothesis → intervention → result → advice for the next person.

These are the results of the first few foundational experiments.

Experiment 1: How long does SEO setup take?

Hypothesis: a new domain with standard SEO setup gets fully indexed by Google within two weeks.

What I did:

• Deployed robots.txt and sitemap.xml, and submitted to Google Search Console
• Added WebSite structured data, meta descriptions, and canonical tags
• One domain, no backlinks

Result: all 10 submitted URLs were indexed within two weeks, and Google found one more page on its own.

Conclusion: the bar for modern SEO setup is low. A new domain is not an obstacle.

Advice: work through the checklist and move on. It's not worth more than a day.

Experiment 2: Does structured data help?

Hypothesis: adding FAQPage and Quotation structured data will trigger rich results in Google.

What I did: each topic page got three Quotation blocks, one FAQPage block, and one BreadcrumbList block in JSON-LD.

Result: Google's Rich Results Test passed on everything; all three types were valid.

But: "detected" is not the same as "shown as a rich card in search." Actually showing up takes time and some domain authority.

Advice:

• Structured data is a must-do, not a magic fix. On a new domain, add it and check back in three to six months.
• One trap I hit: don't use straight ASCII double quotes inside JSON strings on Chinese pages. Use full-width 「」 instead, or you'll break the JSON and the whole block stops working.

Experiment 3: Can AI crawlers see JavaScript-rendered pages?

This was the most important experiment of the set.

Hypothesis: a JavaScript-rendered landing page is unfriendly to AI crawlers.

The setup:

• The homepage was a bundled single-page app (SPA) generated by a design tool, with all content inside JavaScript.
• The topic pages were plain static HTML.

Result:

• Homepage: Perplexity picked up no content at all.
• Topic pages: Perplexity found them within three days. Even a thin fallback page made the top 10 sources for one query (position 8).

Conclusion: strongly confirmed. Static HTML is a prerequisite for GEO. A JavaScript-rendered SPA is close to invisible to AI crawlers.

Advice:

• If GEO is one of your goals, don't build content pages as an SPA. Every page must have its full content on first render.
• Many AI site builders and design tools export SPAs by default. They look great, and to AI search they're empty. Before launch, check the page source: if you can't see your text there, AI crawlers probably can't either.

Experiment 4: Do the AI search engines treat new domains the same way?

Hypothesis: different engines have different tolerance for new domains.

What I did: ran the same queries on Perplexity and ChatGPT (Claude isn't finished yet).

Result:

• Perplexity: discovered the new domain fastest, in three days. Chinese-language queries pull from Chinese-language sites first.
• ChatGPT: a very strong bias toward authority. It often answers from training data without searching at all, so older domains dominate.

Conclusion: Perplexity is the best entry point for a GEO cold start. ChatGPT is the hardest.

Advice:

• In the cold-start phase, watch Perplexity's data first. ChatGPT's feedback may lag by six to twelve months.
• Don't skip Bing: ChatGPT's web search relies on Bing's index, so submitting to Bing Webmaster Tools is a must.

A checklist you can take

1. Do SEO setup in a day: robots.txt, sitemap, Search Console, Bing Webmaster Tools, meta tags, canonical tags, WebSite structured data.
2. Build content pages as static HTML or server-rendered pages, not an SPA. Check the page source before launch.
3. Add structured data, but don't expect rich cards on a new domain right away.
4. On Chinese pages, use 「」 instead of ASCII double quotes inside JSON-LD.
5. In the cold start, watch Perplexity first, be patient with ChatGPT, and always submit to Bing.

The next post covers the other question: which metrics should you actually use to judge whether any of this worked?