Going from 3% to 17% traffic from ChatGPT - here's what worked and didn't work for me

Introduction

6 months ago I saw the first 2 visitors from ChatGPT to my website and one thing shocked me. Both users signed up. The data was obviously not statistically significant, but my intuition told me that users who come from ChatGPT would convert at a higher %, because they are being told what to do, by an AI that they see as a trustworthy authoritative source.

Since then, I've been running hundreds of tests (I have a portfolio of 6 side-projects that I've built over the years) across multiple LLM models, to see what's working and what's not working in driving traffic to my website. Here are the top 5 things that drove results for me:

(1) Add llms.txt to your website

Just as robots.txt guides search engine crawlers, llms.txt is emerging as a standard to help LLMs (ChatGPT, Gemini, Claude, Perplexity etc.) understand and process your website's content more effectively.

llms.txt is a plain text file, typically written in markdown, placed at the root of your website (e.g. https://yourdomain.com/llms.txt). It provides a structured summary of your site's key content, making it easier for LLMs to access and interpret the information without the clutter of HTML, JavaScript etc.

In case it helps, here's a basic template I've created based on my own tests as well as research and standards proposed by teams at Anthropic and OpenAI:

# Your Project Name (https://yourdomain.com)

> Brief summary of your project or website.

## Website

- https://yourdomain.com: Main site and access point.

## About Us

Provide a detailed description of your company, product, or service.

## What We Do (Features)

- [Feature 1](https://yourdomain.com/feature1): Description of Feature 1.
- [Feature 2](https://yourdomain.com/feature2): Description of Feature 2.
- [Feature 3](https://yourdomain.com/feature3): Description of Feature 3.

## Topics We Cover

- Topic 1
- Topic 2
- Topic 3

## Key Blog Posts & Resources

- [Blog Post 1](https://yourdomain.com/blog/post1)
- [Blog Post 2](https://yourdomain.com/blog/post2)
- [Blog Post 3](https://yourdomain.com/blog/post3)

## Supported AI Platforms (optional)

List any AI platforms your content is optimized for.

## Preferred Summary Style for AI

Provide guidance on how AI should summarize your content.

## Target Audience

- Audience Segment 1
- Audience Segment 2
- Audience Segment 3

## Contact

For inquiries, visit: https://yourdomain.com/contact

## Additional Guidance for AI

- Specific instructions for AI models when referencing your content.

Place the llms.txt file in your website's root directory so it's accessible at https://yourdomain.com/llms.txt

There are 2 things that are making me believe llms.txt is here to stay (compared to other standards such as agents.md):

• Anthropic implemented llms.txt to guide LLMs through their documentation. You can view their implementation here:https://docs.anthropic.com/llms.txt
• Google included a llms.txt file in their Agents to Agents (A2A) protocol:https://github.com/google-a2a/A2A/blob/main/llms.txt

(2) Add FAQ sections to every page

I've noticed that integrating well-structured FAQ sections on each (marketing) page significantly boosts visibility in LLMs. These models tend to favor content that directly addresses user queries in a clear and concise manner.

I believe that one thing I did well, was to use customer feedback sessions, call transcripts etc. to guide the questions and answers written in the FAQs so I can copy the style of user prompting as much as possible. I think my logic makes sense here but I don't have any clear data to back it up apart from the fact that the blog posts that have an FAQ tend to get more traffic from ChatGPT (and a bit from Perplexity too)

To further optimize my FAQs for search engines and AI models, I implement the FAQPage schema markup. This structured data helps search engines understand the content, potentially leading to rich results in search listings.

Here's a basic example using JSON-LD:

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [{
    "@type": "Question",
    "name": "How does post bridge help with social media scheduling?",
    "acceptedAnswer": {
      "@type": "Answer",
      "text": "post bridge allows users to schedule and cross-post content across multiple social media platforms from a single dashboard."
    }
  }]
}

I recommend readingGoogle's documentation on FAQPage structured data.

(3) Create quality content with original statistics

Content featuring original statistics and unique data points significantly increases the likelihood of being cited by LLMs. They prioritize evidence-based information to support their responses, making original research a very valuable asset for making your startup more visible (imagine my face when one user mentioned being referenced to my website because of one research paper I wrote about health wearables and AI from a few years ago).

I recommend reading this research paper on GEO:https://arxiv.org/pdf/2311.09735. It covers on many aspects of GEO and includes lots of insights into different experiments that can be used to get more traffic from LLMs. Here are 2 key insights I pulled out:

Luckily, one of my cofounders has accumulated 20 years of data in his niche (government contracting) so we could run some tests on this that gave us great results.

(4) Create dedicated feature & use-case pages

Instead of relying only on your homepage, build separate pages for:

• Features (e.g., "Automated Proposal Generator")
• Case studies or use cases (e.g., "For Staffing Small Businesses" or "For Manufacturing Companies")

I ran ~140 tests designed specifically around pain points that users ask for, based on various companies and their ICP, and I've noticed that LLMs love citing feature/case study pages more than other pages that might contain the needed information. For instance, take the following prompt:

The responses referenced HubSpot as the top option, but what shocked me is that the citation did not use HubSpot's landing page, but instead used https://www.hubspot.com/feature/crm, so a page dedicated to the CRM feature.

I also found that specifically ChatGPT loves to reference these deeper pages over homepages when answering specific prompts (e.g. "best crm for freelancers").

Do not forget to add an FAQs to those pages 😇

(5) Publish on Medium, LinkedIn & YouTube

I had a few feedback sessions and demos with people who came from ChatGPT and they mentioned seeing my cofounder's YouTube videos and LinkedIn articles be referenced in answers, hence why they came across our platform. I think that this increased visibility can be attributed to the structured and authoritative nature of content on LinkedIn articles (maybe also because Microsoft has a large chunk in OpenAI 😉), but I can't really understand why our YouTube videos are referenced.

The only thing I can tell you is that my cofounder has about 1.5k videos on his YouTube channel:https://www.youtube.com/@neilmcdonnell

I did a few tests on questions that I know are answered in his videos; here's an example for the following prompt:

Towards the end of the response, his video was referenced.

Also, when carrying out tests on other companies, I've noticed Medium was cited quite a few times (especially by ChatGPT). I assume that same goes for Medium as it does for LinkedIn articles and YouTube videos, the more quality content you post the better. Just one thing, make sure your articles are not for premium members only, so LLMs can freely reference it (this is only my assumption, backed by a few tests).

What I'm testing next

Conclusion

GEO is changing super fast (and so is SEO). I'm trying to learn as fast as possible and the best way to do so is by helping other founders.

Feel free to ping me (alex@seoforge.ai) and I'd be more than happy to share any knowledge I have to help you drive more traffic from LLMs. I do those sessions for free just so I can understand the space better and hopefully build a solution that is genuinely valuable.

Really appreciate you!