Lead Generation with ChatGPT: The Power of AI Referrals

Over the past several months, we’ve noticed an increase in referral traffic coming from ChatGPT. It’s a tiny segment of our web traffic but a valuable one. Using data from sixteen of the busiest Google Analytics accounts we manage, we segmented ChatGPT referral traffic against traffic from all sources. The performance difference is stark:

Users who arrive via ChatGPT spend more than twice as long browsing the website on average. The number of engaged sessions per user is 50% higher for ChatGPT traffic. These users generate 25% more events per visit and convert 33% more often.

Essentially, ChatGPT referral traffic is the gold standard for engagement in the B2B sector, outperforming any other single traffic source across most metrics. So, how do we get more of it? SEO has proven long-term ROI. Can the same tactics be applied to improve visibility on ChatGPT? Is AI Citation Optimization the next big thing for digital marketers?

This article will explore possible tactics to improve website visibility in ChatGPT and other AI applications. We’ll explore some top-performing pages across websites we manage, identifying possible reasons why they’re appearing in AI—and why others aren’t. Then, we’ll offer our top tips for marketers looking to scale their website’s AI visibility. Let’s get started.

Chatbot referrals: what are they, and where do they come from?

When AI chat tools like ChatGPT first launched to the public in 2022, they were unable to connect to the internet. Relying on prebuilt models, they were unable to provide up-to-date information or fact-check their own statements. This could lead to severe ‘hallucinations,’ particularly when queried on current events.

Within months, developers had patched in the ability to search the internet for context. This allows ChatGPT and similar tools to provide more accurate responses on timely topics and—at least in theory—provide credible sources for its claims. These sources appear as in-line citations—hyperlinks in the chat response linking out to relevant web pages.

This system is far from perfect. In many cases, cited web pages are only broadly applicable to the topic at hand. The feature has steadily improved over the past few months, correlating with a sharp increase in the amount of referral traffic observed from the platform.

Chatbot citations present an interesting marketing opportunity. Where SEO is a crowded, highly competitive field dominated by the world’s biggest brands, AI citations seem to be a more level playing field for now. How to rank for them isn’t fully understood, but it seems that small and mid-sized websites and brands have more opportunity to capture citation links than to rank first overall in Google.

Search engines vs. answer engines?

Some marketers are beginning to refer to AI chat tools as “answer engines” and the process of optimizing websites to appear on them as “answer engine optimization” or AEO.

It points to a key difference in the functionality of these so-called answer engines versus conventional search engines.

A search engine’s job is to find websites relevant to your search query. Users search simply and act quickly, often relying on volume (clicking several results to find the right answer) over specificity.

An answer engine’s job is to actually provide the answer, potentially supporting it with supplementary information and citations. While a search engine is part of a user journey, an answer engine may itself be the destination. For many queries, users can get their answers directly from ChatGPT without ever leaving the chat window.

Different user intent on ChatGPT vs. Google

The average Google search term is four words long. The average Google search session lasts for less than fifteen seconds before the user clicks a result. Nearly all Google searches result in a click on one of the first ten results. This means a Google search is unspecific and conducted quickly.

The average ChatGPT session lasts over six minutes—24X longer than the average Google session! A chat format is conducive to deeper, more thoughtful searching. Users on ChatGPT don’t input four words and choose from one of ten results. A user who clicks on a ChatGPT citation may have been discussing the topic for five minutes beforehand, digging deep and searching for a highly specific answer.

All this is to say that traffic from AI platforms is warmer and more engaged than search traffic. Both are high-intent inbound platforms, but ChatGPT is on a different level of engagement.

Generative AI may have gone public about two years ago, but it’s still an emerging technology. A much smaller portion of internet users are on ChatGPT than on Google. This smaller audience is of higher quality than the general population and is more likely to consist of working professionals and tech early adopters.

Best practices to increase AI referrals

A note on data limitations

Before we dive into our recommendations, let’s talk about the data we have available to us—or, perhaps more importantly, the data we don’t have.

• SEO benefits from a really high level of visibility from search platforms and website analytics. We have data and attribution from end to end. In many cases, we know which search terms resulted in traffic to which pages, along with all kinds of demographic and engagement data along the way.
• AI chat applications don’t give us this end-to-end visibility. When a user clicks an AI citation link, we do not receive any data regarding the nature of the conversation that led to the click. All we get is the referrer value—where the click came from in a general sense.
• There are also many AI models, all of which work somewhat differently. None have released specific guidance on how to optimize web content for citations. In short, we cannot say that improving this metric or that will directly increase your ChatGPT referral traffic. We can only observe and report on the factors and conditions that top ChatGPT referral pages share. These attributes are like best practices by proxy: it’s not always clear that they directly contribute to visibility in AI platforms, but it’s likely that by improving the following metrics, your pages will be more attractive to AI chat applications.

Internal linking: higher PageRank correlates to more AI referrals

Our analysis shows that pages with strong internal PageRank are considerably more likely to get cited by AI. PageRank is a well-understood component of SEO. It’s a score generated by Google based on the number and quality of links to a given web page. If a page is considered highly authoritative, other websites will likely reference it, providing backlinks to the content. If those other websites are themselves authoritative and high-quality, they will pass on higher PageRank scores with their links. It’s fundamental to how Google and other search engines rank websites.

Internal PageRank is just what it sounds like. Rather than measuring external backlinks, it measures internal link structures on your website. The same principles apply: High-quality, high-traffic pages confer more PageRank than lower-quality pages.

Improving internal PageRank must be done mindfully. Trying to boost a specific page by having every single page on your website link back to it will not be very effective. Internal linking should be logical and focused on user intent: does adding a link help users access your most important content? Are interlinked pages relevant to each other? These are powerful internal links that may boost your visibility on AI platforms. Less relevant links may confer little PageRank.

Our recommendation is to audit your internal link structures, looking for gaps and inefficiencies. What information on your website is crucial for customers to understand? Are you sure that users will encounter that information in a typical visit, or is it possible that they will miss it?

Content length: word count matters, but not as much as the topic

We found a moderate correlation between word count and AI chat visibility. Though some pages that received referral traffic had as few as one hundred words, the average word count for ChatGPT referral pages was 1128. A typical website’s average word count per page is far shorter than that, often between 200 and 500 words.

In most cases, shorter pages that received referral traffic were homepages. The longer pages tended to be blog posts, service pages, news articles, and case studies. Top performers are niche informational articles, often with a focus on educating readers. This is intuitive; ChatGPT and other AIs excel at answering highly specific questions and will often rely on citations to back up their assertions.

Home pages get top billing on ChatGPT.

Of the sixteen websites we analyzed for this article, eleven saw their homepage receive the most referral traffic from ChatGPT. For five other websites, their homepage was the second-most visited referral page. Only one website did not receive clicks to the home page from ChatGPT.

This may reflect a particular and seemingly common type of query on AI chat programs. Many users report using ChatGPT to generate lists of products or services that suit their needs. For instance, users might ask ChatGPT something like:

“I’m looking for the best suppliers of eco-friendly building materials for kitchen renovations. They need to ship to Arizona. Give me five options to choose from.”

This type of query will inevitably result in a list of links, and those links will be to the homepage of each identified company.

There’s little to be done to increase the visibility of your homepage beyond typical SEO best practices. Instead, consider how your homepage might be adjusted to better address the needs of this type of AI user. We know that your homepage is the most likely page on most websites to appear in a chat citation, so the goal is more about optimizing the conversion rate of visitors from AI. Perhaps more social proof elements above the fold would put this user’s mind at ease? Maybe your homepage could do a better of job of communicating your unique selling propositions to new users?

If ChatGPT is an “answer engine,” ensure your content has the answers!

In the age of the answer engine, marketers may want to rethink their approach to content production. We find that direct, instructional content—including case studies and Q&As—is more likely to be cited by ChatGPT. Consider what AI chat users love about AI: quick, easy-to-read answers on almost any subject. ChatGPT wants to reference things you know about. Do you have that type of content on your website? Consider your existing content paragraph by paragraph. Ask yourself how well each section answers a user’s possible questions. Clarify and expand upon thin content. Remove or replace unclear, vague, or low-effort content.

The Citation Ceiling: AI links are unlikely to deliver traffic volume

While we’re big believers in the potential of ChatGPT as a referral channel, it’s unlikely to become a top referrer by traffic volume for your website.

Not every chat response provides citations. Not every citation in a chat response gets clicked. As of writing, a typical response with citations may have anywhere from two to five links—not the effectively infinite choices offered by a Google search. In short, even if ChatGPT completely replaced Google Search tomorrow, you would see far fewer users from ChatGPT than you do from Google today.

AI referrals are and will likely always be a quality-over-quantity play. While AI should be prioritized in your marketing strategy, don’t expect to grow your website user count through AI.

AI citations are a level playing field…for now.

Anyone who’s used AI for long enough knows that it can generate some amusing or even disturbing hallucinations. From telling people to eat rocks or put glue on pizza to making eerie claims about sapience, AI gets things wrong.

If a user requests a list of the top five industrial marketing agencies in the world, AI doesn’t actually know anything about that. It’s going to perform an internet search on your behalf, using criteria we don’t directly understand. The list it delivers isn’t arbitrary, but it’s also not necessarily true.

It’s somewhat reminiscent of earlier days of SEO. Over the course of decades, the internet and its search engines matured together. While many early winners of the SEO battle don’t even exist anymore, others are entrenched. Websites like Forbes have a kind of SEO critical mass that’s unshakeable. They rank on the first page for millions of queries spanning every topic imaginable. It can be difficult, if not impossible, to take a top ten search ranking from these entrenched websites.

At least for now, we control our narrative in ChatGPT. It uses—some would say steals—our content to generate its responses. Whatever ranking factors it considers aren’t as biased toward top websites. What we discuss on our websites and the tone in which we discuss it materially affects how AI will describe us.

Be aspirational, bold, and upbeat about your company and its offerings. Write authentically but energetically. Present yourself as you’d like to one day be seen. Right now, there is a golden opportunity for businesses to establish themselves in the answer engine economy. It probably won’t last forever.