How We Optimized BlogBurst for AI Search Engines (GEO in Practice)
We implemented GEO (Generative Engine Optimization) for BlogBurst's website — the practice of optimizing content so AI search engines (ChatGPT, Perplexity, Claude, Google AI) cite your website. Here is exactly what we did and the early results.
What We Implemented
1. llms.txt and llms-full.txt
These are plain-text files at your website root that AI crawlers read to understand your site. Think of them as robots.txt for AI.
- llms.txt: Concise summary — what your product does, key features, pricing, links
- llms-full.txt: Comprehensive version with detailed feature descriptions, use cases, and technical details
We serve these at blogburst.ai/llms.txt and blogburst.ai/llms-full.txt. Both return 200 OK. Whether AI engines actually read them is a different question, and we checked our logs rather than assuming — see the numbers below.
2. Comprehensive JSON-LD Structured Data
Every page has schema.org markup:
- Organization: Company info, social profiles, contact
- SoftwareApplication: Product features, pricing tiers, platform support
- BlogPosting: Article metadata for each blog post
- FAQPage: Question-answer pairs AI can directly reference
- BreadcrumbList: Site navigation structure
3. Answer-Optimized Blog Content
Each blog post is structured to directly answer specific questions. If someone asks ChatGPT "How do I automate Bluesky posting?", our content has a clear, direct answer near the top — not buried under 500 words of introduction.
4. AI Crawler Access
Our robots.txt explicitly allows all AI bots:
- GPTBot (ChatGPT)
- ClaudeBot (Claude/Anthropic)
- PerplexityBot (Perplexity)
- Google-Extended (Google AI)
Many sites block these by default, which means AI cannot cite them.
5. Specific Data Over Vague Claims
AI models favor specific, verifiable claims. "BlogBurst saves time" is weak. "BlogBurst users report saving 1.5-3 hours per day across 50+ accounts" is citable.
What Our Logs Actually Show
Every number here comes from our own nginx access logs for blogburst.ai, covering 12–26 July 2026: 14 days, 270,860 total requests. We give the denominator because a crawl count without one can be made to mean anything.
AI crawlers visit constantly
| Crawler | Requests in 14 days |
|---|---|
| Bytespider (ByteDance) | 12,478 |
| ClaudeBot (Anthropic) | 1,212 |
| Amazonbot | 1,045 |
| GPTBot (OpenAI) | 527 |
| OAI-SearchBot (OpenAI) | 432 |
| PerplexityBot | 239 |
| Google-Extended | 9 |
| Total | 15,942 — 5.9% of all traffic |
Turn this topic into a buyer-intent map for your product.
Paste your URL and see which search, community, and proof gaps the engine would attack first.
Not one of them requested llms.txt
Zero. Of those 15,942 AI-crawler requests, the number that asked for /llms.txt or /llms-full.txt is zero.
Those two files did get 65 requests in the same window — 39 from AgenstryBot, 13 from NotHumanSearch, 12 from an unlabelled Mozilla user-agent, and 1 from NetRanksPolicyAudit. Those are GEO audit tools scanning us. The main consumers of llms.txt today are the products that check whether you have an llms.txt.
We are keeping ours. It costs nothing to serve and the standard may yet be adopted. But we are not going to tell you it drives citations, because our own logs say it is not being read.
What the crawlers did fetch: ordinary HTML
All 15,942 requests went to normal pages — blog posts, landing pages, docs. That is the practical lesson: the thing AI engines read is the same HTML a browser reads. If your page renders its content with JavaScript and ships a near-empty document, a crawler sees the empty document. No side-car file compensates for that.
The caveat that matters
A crawl is not a citation. These logs prove AI companies fetched our pages; they prove nothing about whether an AI answer ever named us. Those are separate measurements, and the only honest way to make the second one is to ask the engines your customers' questions and record what comes back, answer text included, so the claim can be checked. Crawl logs tell you that you are reachable. They do not tell you that you are recommended.
Housekeeping we also did
We fixed robots.txt and our sitemap, after which Google indexed 176 URLs, and we pushed 732 URLs to IndexNow (which feeds Bing and Copilot, not Google). Both were worth doing, and neither is a ranking or citation win — that is hygiene. Being crawlable is the entry requirement, not the achievement.
GEO vs Traditional SEO
| SEO | GEO |
|---|---|
| Optimize for ranking in blue links | Optimize for being cited in AI answers |
| Keywords and backlinks | Content quality, structured data, specificity |
| Meta descriptions for click-through | Direct answers for AI extraction |
| Compete on page 1 | Be the source AI models trust |
GEO is still early — massive first-mover advantage for sites that implement it now while competitors focus on traditional SEO only.
Start BlogBurst growth run — built with GEO-optimized content and structured data from day one.
Related Reading
- Why generic AI growth content fails without tracked diagnostics.
- GEO: The Complete Guide to Generative Engine Optimization
- AISO is the New SEO
- Day 6: Google Can't See 172 of My Pages
Build the first growth map for your product.
BlogBurst starts with buyer intent, product proof, and attribution. The first run tells you what to ship and why.
Start with my URLFirst growth run · No credit card required