How is ChatGPT Sending Traffic to My Website (And From Where)?

AI has been changing the world of SEO a lot, and not just because LLMs can be used to create content for websites.

One of the claims made by OpenAI is that ChatGPT can “replace search engines” as an alternative to Google. Instead of asking Google a question and getting a bunch of website results you need to click through, read, evaluate, and think about, you can just ask ChatGPT the question and get the answer.

“What if the AI is wrong?” OpenAI’s claim is that ChatGPT can use live web results to pull data and get accurate answers. It uses Bing (because of their partnership with Microsoft) and pulls the top results to summarize for the answers it generates.

There are some clickbait headlines circulating about ChatGPT challenging Google’s throne as the dominant search engine, though for the time being, the numbers say that’s still a long, long way off.

But, if you’ve noticed ChatGPT (or Perplexity or other AIs) in your Analytics under traffic referrers, you might wonder where that traffic is coming from and if it’s valuable to you. Let’s talk about the kinds of traffic ChatGPT and the other AIs send your way and if it’s something you should pay attention to or not.

Where is ChatGPT Traffic Coming From?

First, let’s talk about the ways ChatGPT may be sending traffic in your direction.

The major source everyone is talking about these days is from ChatGPT Search, the version of ChatGPT that can use Bing to search the web to provide answers rather than relying solely on training data.

Functionally, it’s a lot like Google’s AI Overviews, which are likely built out as a direct competitor. It’s also a reflection or iteration of the rich results and knowledge graphs. It’s all just a way to summarize and present information to a searcher and answer their question without them having to visit the websites the information is pulled from.

With ChatGPT Search, the answer is generated as normal, but two banks of links appear on the side. One is “citation” links, which are the results that ChatGPT is pulling most directly from. In fact, in many cases, you can find the content of the ChatGPT answer is almost word-for-word copied from the cites being cited. The second bank of links is Search Results, and gives a broader review of the answers.

Basically, you can figure that ChatGPT is doing a Bing Search for whatever your query is, copy-and-pasting info from the top 1-2 results, and handing you a list of the rest of the results if you wanted to skim through them. It’s sort of like the user behavior you would do when searching, except with less of a critical eye to making sure the top results are relevant before clicking on them.

This brings up a significant question. If ChatGPT is presenting you with the answer, would you still click through to the cited site? This is a question Google has been grappling with for a long time, and it’s why a lot of their knowledge graph and rich results are limited in scope and heavily feature the link to the source. ChatGPT doesn’t have that kind of ethical concern and really only includes the links as a response to people bringing up AI hallucinations as a problem.

Well, the answer is fuzzy. Some people certainly don’t click through. Some certainly do. Many websites can check their analytics and find referral traffic sent from ChatGPT.

One huge caveat here is that the referral data is largely coming from UTM tracking automatically added by ChatGPT. These UTM parameters are only reliably added to citation links – not to search result links – and they also muddy the waters from people who copy and paste ChatGPT links in other sources with the tracking data intact. All of this means:

• You likely have some traffic coming in from ChatGPT.
• You likely have more traffic coming in from ChatGPT that isn’t being tagged properly.
• You likely have traffic tagged as ChatGPT that isn’t actually directly from ChatGPT but rather from people who used it as a resource and pasted the links elsewhere.

Different kinds of search intent likely have different levels of traffic coming in from ChatGPT, as well. Informational queries probably aren’t getting all that many clicks beyond faith-check clicks, while queries with more of a transactional intent – where the user wants to ask the AI for a recommendation and go take that recommendation – likely refer more. That said, a huge proportion of the traffic being referred from ChatGPT is to news sites like the BBC and Fox News, so really, it’s all up in the air.

Darker Traffic from ChatGPT

There’s another source of traffic hitting your site from ChatGPT, and it’s definitely one you want to know about, though how you view it will depend on your perspective.

When a user performs a query in ChatGPT powered by search, they get links. What if they aren’t using that feature? ChatGPT can still provide answers. More to the point, it can still provide your answers, just without the citation there.

This is because OpenAI has one of the largest fleets of scraper bots this side of Google, actively crawling the web and hoovering up everything it possibly can. They aren’t subtle about it either, and the crushing weight of the ChatGPT scrapers can absolutely blast a site. In one example, it was akin to a DDoS attack.

This traffic is bot traffic, not human traffic. It uses a whole laundry list of IP addresses and other mechanisms to avoid easy blocks and detection while slurping up every bit of server resources it can in order to scrape data.

There are three problems with this.

The first is that, obviously, it can crush underpowered servers or small sites without a blink. Any web host that has data caps and charges extra after the bandwidth is exceeded – rare but still extant in some places – would immediately kill a site hammered by these bots.

The second is the technical considerations. Yes, other bot fleets, from Google to Ahrefs, scrape the web. But they have rate limits, they have mechanisms to be lighter-weight in their scraping, like not loading scripts or images unless necessary, and they don’t repeatedly hammer the same site over and over.

Note: You can take action to block this scraping from ChatGPT and others. This post is a good rundown of the options you have available to you. There are also services springing up to help as well.

The third is ethical. There’s no way to block the hundreds of rotating IPs, no user agents to easily flag, and no easy way to tell OpenAI you don’t want them scraping your site. Because, make no mistake, they’re scraping your site to feed your data into their LLMs for training. Some people don’t care about this, while others find it a flagrant violation of copyright and usage rights.

How much of the traffic that you get from ChatGPT comes from these sources and not from real people? It can be hard to say, and it depends on how well your analytics identifies and filters bots from your reports.

As site owners and marketers, we generally want our analytics to reflect real people, not bots. It’s currently unclear how much of the traffic being referred by ChatGPT is real people, at least in the reports I’ve seen thus far. There’s too much inconsistency to be certain.

Is ChatGPT Traffic Something to Encourage?

If ChatGPT is sending traffic your way, is that a good thing or not?

Obviously, scraper traffic isn’t valuable to you. Setting ethical arguments about LLM training aside, the GPT bot is never going to click on your ads (or if it does, it’s clear bot traffic and doesn’t earn you anything), it’s definitely never going to make a purchase, and it doesn’t help your business out at all.

What about traffic from ChatGPT answers?

Whether or not this traffic is something you want to encourage comes down to whether or not it’s actually valuable to you. There’s no clean answer here; you have to track it specifically and measure if those people are taking actions like subscribing or purchasing. If they are, then sure, encourage more traffic. If not, then you can feel free to ignore it.

I’ve seen a few posts indicating surprisingly high conversion rates from ChatGPT traffic. This one makes the argument that ChatGPT is doing the work of convincing users to buy from you before they arrive on your site, so their conversion rate is higher by default – because the people who aren’t convinced aren’t clicking through.

It’s not like you can neglect your on-site sales funnel, though; without that data, the LLM has nothing to scrape to recommend you, so it’s really just another channel in the funnel, not something external or entirely separate.

Posts like these also tend to compare conversion rates from ChatGPT to conversion rates from organic Google search, but as we all know, organic search is fairly low in converting because users want to do their consideration first. Where that consideration is done – social media, repeat direct traffic, offline research, ChatGPT – doesn’t necessarily matter as much as some people think.

The heart of the point I want to make here is that all of this is interconnected. People often pit Google against other channels, but the truth is that it’s all part of a greater ecosystem, and to ignore or neglect any one part of it is likely to make a mistake.

Of course, certain kinds of sites stand to lose out a lot from ChatGPT. Sites that don’t have products to recommend, sites where the value is the information itself, and sites where they rely on the ad revenue to survive are more likely to get nothing of value from the ChatGPT traffic.

Think about a food blog that relies on ad revenue, affiliate links for tools, and incidental cookbook sales to survive. When ChatGPT scrapes and presents the recipe and leaves out all the rest, what happens to that site?

One way or another, we’re in for a shakeup of the way the internet works, and I don’t think anyone really knows the way it’s all going to settle.

How to Optimize for AI Traffic

I know a lot of you are here for this question, and you might wonder why I’ve relegated it to the end.

The truth is, while a lot of people are talking a lot about AIO (Artificial Intelligence Optimization), especially as a replacement for SEO, it’s all a smokescreen. If you were paying attention above, you already know why.

ChatGPT, when it has links to give, is getting those links from Bing. It is, almost universally, just the links you would get if you typed your question into Bing directly.

That means “AIO” is (in many ways) just Bing SEO. Bing, which takes after Google in so many ways, functions so closely to how Google functions that the SEO is more or less identical.

The same holds true for other AI tools, as well. Pretty much all of them except Google’s are using either Google’s search results or Bing’s search results, whether they tell you and provide links or not.

So, if you think ChatGPT traffic is worth it to your business, all you need to do is work on SEO with an emphasis on Bing SEO.

Leave a Comment