Hey folks,
Elliot here.
When it comes to AI performance, there are a few questions you’re probably hearing all the time from leadership:
- Are we showing up on ChatGPT?
- How visible are we across LLMs?
- Is that visibility turning into traffic, pipeline, or revenue?
The aim of this newsletter is to give you a primer on how you can effectively report on LLM performance: looking at what metrics matter, and what just isn’t that important.
Let’s get into it!
Why LLM traffic isn’t the best metric to track
If we can see lots of traffic from an LLM source, surely that means we’re appearing a lot in LLMs?
The answer is…kind of.
Think about how you use LLMs: how often do you click on the links? Most likely, you’ll do a bit of research, then type the brand name into the Google search bar and go from there. That’s how most people use LLMs.
In reality, LLM traffic doesn’t even come close to capturing the true impact of LLM search. Semrush data shows that 86% of consumers verify AI brand recommendations before buying, with Google being the most common place they go to validate them.
So even if someone discovers your brand through ChatGPT, the visit may still show up in analytics as organic search, direct traffic, or referral traffic: not LLM traffic.
Referral tracking is also inconsistent across LLMs. Some clicks appear as sources like chatgpt.com, chat.openai.com, perplexity.ai, or gemini.google.com. Others lose the referrer completely, especially when someone clicks from an app, private browser, copied link, or in-app browser. In those cases, the visit is recorded as direct or unknown.
A real example: 0.88% of all traffic came from LLMs…but LLMs brought in ~30% of inbounds
Here’s a real example to demonstrate why LLM traffic isn’t a good metric for measuring success in LLMs.
For this client, we track their LLM traffic through GA4. In April, they had the lowest LLM traffic this year: see the trend line below.
And all in all, LLM traffic was a measly 0.88% of all traffic to the website.
Despite this, in the same month, around 30% of all inbounds were influenced by LLMs. That is a huge number and shows that AI search is a major pipeline channel for this business.
This example illustrates how misleading the traffic number alone can be.
If you reported that 0.88% of all traffic came from LLMs to leadership, there would be a bit of an awkward silence…but tell them that 30% of all leads came from LLMs, and it would be a different story. This is why it’s so important to track your LLM inbounds and not to rely on traffic alone.
A caveat: LLM traffic may increase due to a recent ChatGPT update
Worth noting, there has been a significant change here recently. On May 7, ChatGPT started embedding clickable brand homepage links inline in answers roughly 5x more often than before. Referral traffic from OpenAI to monitored brand sites nearly doubled overnight.
I’ve linked a great article from Josh at Profound that details these changes at the bottom of this newsletter.
Despite these changes, I still think traffic is just not a strong enough metric to answer the question you’ll likely get from leadership: “How are we doing in AI search?"
The two most important LLM metrics worth tracking
1. Self-attributed LLM leads through ‘How did you hear about us?’ fields
This is the most important metric to track, and you can literally set it up in minutes.
The easiest way to track LLM leads is to add a mandatory "How did you hear about us?" field to your lead gen forms, with ‘AI search tools’ or similar as an option. It’s that simple.
When our clients set this up, we found that many of their inbound leads were already coming from AI tools. Yet without this field, we would have no idea the impact LLMs were having. One client was getting 3 - 4 leads via LLMs (tracked via UTM links). Once we added this form, we saw they were getting 10 - 15 leads, and that actually LLMs were a much more important channel than they realized!
This gives you a useful baseline figure and lets you track how your LLM visibility strategy is working.
If your visibility increases, you should also see an increase in LLM leads. Worth noting this isn’t 100% accurate, as we’re relying on the user to accurately self-attribute. But it’s the best we can do.
Having this data available enables you to show leadership how LLMs are impacting commercials, and gives you the data you need to justify investment into improving LLM visibility. Instead of treating AI search as an abstract channel, you can connect it to real commercial results, with an actual dollar value attached to each lead.
2. Brand visibility for relevant prompts
Of course, if your brand isn’t appearing for relevant prompts, you’re unlikely to get any self-attributed LLM leads in the first place.
So how do you tell if you are appearing for the right prompts? In simple terms: choose the prompts that matter, track whether you show up, and measure whether that visibility improves over time.
Here’s our process:
- Make a list of BOFU prompts. Start by making a list of prompts your ideal customers might ask when researching a solution like yours. Focus on prompts where it makes sense for LLMs to mention your brand: otherwise you will end up training the models but not being mentioned.
For example, don’t go after “what is open banking?”. Instead, try “open banking providers for wealth management platforms”: the former probably won’t ever cite a brand. We always work closely with our client and interview lots of sales people and SMEs to find out the best prompts to go after.
- Track your brand visibility with a third party tool. Use a tool like Peec.ai or Profound to track these prompts. You’ll be able to see how often your brand appears for your tracked prompts, and also see a 'visibility’ percentage, like so:
You can also see which specific content is being cited.
In this example, we can see that the GPT article (more on what this is later) we created for Fiska is being cited by LLMs, increasing our visibility.
Of course there are millions of ways you can write these prompts, but we've found that if you go after one, you tend to reliably appear for similar prompts.
Finally, with these tools, you can track your total visibility over time:
This gives you a really clear and compelling visual to reassure leadership that your brand is growing in LLM visibility.
Another metric people track is citations: how often an LLM uses your website as a source in its answers. This is useful, but brand visibility matters more.
For example, ChatGPT might cite your blog post in an answer to “what is open banking?” without ever mentioning your brand by name. This means your content may be helping the model answer the question, but your brand isn’t actually being surfaced to the user.
How do you improve LLM visibility and LLM leads?
So now you know what to track, what do you do about improving your brand’s performance in LLMs?
To get more LLM leads, you need to improve your LLM visibility for relevant prompts.
To do that, you should write what we call GPT articles.
This is the framework that’s helped us significantly grow LLM visibility and leads for many of our clients, including increasing one client’s visibility from 3.6% to 58.8% across 50 key prompts (full case study coming soon).
At its core, GPT articles are short, super-specific pieces of content that target specific prompts you know your ICP will be searching.
For instance, if you’re an embedded payment solution provider, there’s a good chance your customer may search something along the lines of ‘Which embedded payment solutions are best for [use case]?’
The goal here is to create quality content that answers a specific question and that LLMs then pick up. Crucially, this content should still be of good quality and contain unique insights: AI slop won’t cut it here.
Here’s an example of what they look like:
These articles work because LLMs aren’t just looking for broad, generic content. They’re looking for clear answers to very specific questions. A traditional SEO article might cover a topic in depth, but the exact answer an LLM needs could be buried halfway down the page.
GPT articles answer one high-intent prompt clearly, quickly and directly. That makes it much easier for LLMs to understand when your brand is relevant, cite the page, and mention you in the answer.
You can learn more about GPT articles here, and see some of the results we’ve been able to get for our clients: Want LLMs to Recommend Your Brand? Here's What We've Found Works: GPT Articles
Mint Studios Recommended Reads
- Is Zero Click marketing dead? The branded link update
- Website not ready for SEO? Prioritise LLM visibility instead
- Does increased LLM visibility lead to more customers? Here's what we found via 3 case studies
Thanks for reading,
Elliot & the Mint team 🎉
We help financial services and fintech companies acquire customers and position themselves as experts with content marketing. Learn more about what we do.
