The ultimate guide to AEO: How to get ChatGPT to recommend your product | Ethan Smith (Graphite)

Answer engine optimization is different. You can be cited tomorrow and immediately start appearing in results. For instance, a Reddit thread, a YouTube video, or a blog mention can quickly elevate a new company. If a brand new YC company launches and everyone discusses them, they could appear in an answer tomorrow. Therefore, early-stage companies can win rapidly by securing as many citations as possible.

It's significantly more valuable. Webflow observed a six times higher conversion rate from LM Traffic compared to Google search traffic.

When you ask, "What's the best tool for X?" the most frequently cited answer will appear first. This differs significantly from Google's approach. For Webflow, we focused on YouTube and Vimeo videos, along with mentions on Reddit and other blogs, and affiliate partnerships. We experimented with various strategies. The most effective were, first, straightforward SEO; second, YouTube videos; and third, Reddit optimization.

Another strategy leverages Reddit's core purpose: to host useful, high-quality, authentic comments from real people. At Webflow, we have a few individuals who actively comment. They identify themselves by name and workplace, then provide valuable information. The strategy involves finding a relevant thread where you want your citation to appear, stating your identity and affiliation, and then offering a useful piece of information. This approach has proven very effective.

First, I would identify the questions I want to rank for. To do this, I would analyze my search data, perhaps even my paid search data, to determine my top-performing keywords. I'd also examine my competitors' money terms. For example, if I'm "Rippling," what keywords is "deal.com" bidding on for their paid search? Then, I would transform these keywords into questions. You can simply feed these keywords to ChatGPT and instruct it to "Make these into questions"; it performs quite well. So, take your competitors' paid search data, or your own, input it into ChatGPT, and generate relevant questions. That's the initial step.

I would take 200 questions. For 100 of them, I wouldn't intervene; this would be my control group. For the other 100, I would intervene. We're already seeing a fair amount of variance in answers without any intervention, so a control group is essential. As more people use LLMs and LLM traffic increases, a control group becomes even more crucial, especially for answering engine optimization.

What happens in an infinite loop of derivatives? You essentially converge on one opinion. For example, if you ask what the best flavor of ice cream is, it will eventually say "vanilla," and only vanilla, with no other flavors. While this is a simple example, feeding derivatives of derivatives into the model shrinks the wisdom of the crowd, leading to a single opinion on everything. This outcome is highly undesirable. This is what occurs if 100% unassisted AI content is used.

People in chat are asking follow-up questions, looking for tools, and inquiring about specific features or integrations. These questions can often be answered in help centers. Typically, an SEO team wouldn't prioritize a help center. However, given the frequent questions in chat about specific use cases, a help center becomes an excellent resource. Therefore, optimizing the help center is crucial.