DEV Community: Edward Calanni

Forget Keyword Stuffing: How "Data Baiting" Shapes the Future of AI Search (GEO)

Edward Calanni — Mon, 21 Sep 2026 22:08:56 +0000

Why AI assistants cite specific sources, how RAG retrieval mechanisms work on structured data, and what "Data Baiting" means for developers and growth engineers.

If you’ve used ChatGPT, Perplexity, Gemini, or Copilot recently, you’ve probably noticed a major shift in how they answer precise technical or market queries. When faced with specific questions—like exact pricing, localized metrics, or niche industry statistics—LLMs don't just generate text from their pre-trained weights. They use Retrieval-Augmented Generation (RAG) to query the live web, pull trusted facts, and cite their sources inline.

This shift has created a new paradigm in digital distribution and AI visibility known as Data Baiting.

I recently read a great article breaking down this concept on PerkFuel: Data Baiting: How to Influence What AI Says. Here is a technical breakdown of why this strategy works and what it means for how we structure data on the web.

The Problem: Hallucinations and Live Web Queries

Every major LLM provider optimizes heavily against hallucinations. When a user asks a hyper-specific question (e.g., "What is the average setup cost for an Italian restaurant in Austin in 2026?"), a language model cannot afford to invent a number.

Instead, the AI assistant's search layer:

Formulates targeted web search queries.
Fetches a small set of high-relevance pages.
Extracts precise facts and formats an answer.
Appends inline citations pointing directly to the source.

If your page contains the exact structured answer the AI is looking for, you become the citation.

What is Data Baiting?

As outlined in the original article, Data Baiting is the practice of systematically refining public or proprietary data into hyper-specific, machine-readable datasets designed explicitly to be indexed, retrieved, and cited by AI engines.

It isn't about publishing mass AI-generated text or link spamming. It’s about data refinement and machine readability.

The 4 Core Architectural Pillars

From a software and data engineering perspective, a Data Baiting pipeline relies on four main components:

Mapping Engine: Identifying precise user intent, sub-verticals, and micro-questions that users ask AI tools (moving beyond traditional keyword research to exact parameter-driven questions).
Data Refinery: Taking raw public datasets (census, registries, labor stats, public APIs) and programmatically computing unique, localized, or cross-referenced metrics that don't exist elsewhere.
Coverage & Distribution: Programmatically generating clean canonical dataset pages, spreadsheets, dynamic charts, and open data endpoints (CSV/JSON).
The Clear Layer (Machine Readability): Serving the exact same facts to both human readers and machines using double-structured outputs—clean UI components alongside robust JSON-LD, Schema.org, and accessible REST endpoints.

Why Data Beats Long-Form Articles for AI Search

Traditional SEO favored long-form blog posts and domain authority. However, RAG systems operate differently:

Retrieval Efficiency: LLM crawlers parse pages to extract dense factual density. A clean <table> backed by a structured JSON payload will almost always beat a 2,000-word fluff article when an LLM needs to extract a single fact.
Recency & Verifiability: AI search pipelines look for explicitly dated sources and clear methodologies to stand behind their cited figures.

Final Thoughts

As developer workflows and everyday user searches migrate toward conversational AI and search-enabled LLMs, optimizing for machine retrieval (Generative Engine Optimization / GEO) becomes as important as traditional web development.

If you're interested in the full breakdown, including architectural workflows and real-world examples, check out the original piece on PerkFuel's Data Baiting Explained.

What are your thoughts on building software and data pipelines explicitly designed for LLM retrieval? Let's discuss in the comments!`

I Tested PerkFuel's Free AI Visibility Checker. It Solves One of the Hardest Problems in GEO

Edward Calanni — Sat, 12 Sep 2026 16:21:42 +0000

For years, measuring search visibility was relatively simple. You chose a keyword, opened Google, and checked where your website appeared. SEO tools later automated the process and turned rankings into dashboards filled with positions, visibility scores, estimated traffic and historical trends.

AI search has made that model much harder to apply.

When somebody asks ChatGPT, Gemini or Perplexity to recommend a software product, an agency, a dentist, a hotel or almost any other business, there is no traditional search results page to monitor. The assistant searches for information, evaluates different sources and generates its own answer.

Your company might appear in that answer. It might be used as a source without being mentioned. Your website might even be consulted before the assistant decides to recommend one of your competitors instead.

This is the problem PerkFuel is trying to make easier to understand with its free AI Visibility Checker.

I spent some time looking at how the tool works, and what I find interesting is that it doesn't try to recreate Google rankings for AI. Instead, it focuses on a much more useful question: if someone who has never heard of your company asks an AI assistant for businesses like yours, will the assistant actually recommend you?

That sounds like a small distinction, but it changes the entire way AI visibility should be measured.

Knowing your brand exists is not the same as recommending it

One of the easiest mistakes to make when testing AI visibility manually is to include your brand in the question.

Imagine you run a dental practice called Northgate Dental and ask ChatGPT whether Northgate Dental is a good dentist. The answer may be detailed and positive, but the test is already biased because you have given the assistant the name of the company you wanted it to investigate.

A potential customer who has never heard of Northgate Dental would probably ask something much broader, such as "best private dentist in Leeds".

That is a far more meaningful question because the assistant now has to decide which businesses deserve to appear without being told which company to investigate first.

PerkFuel's checker follows this logic. You enter a domain and the system creates a non-branded customer-style query related to the website. It then examines the answer generated around that query.

This means the test is closer to genuine discovery.

The difference matters because brand awareness inside an AI model and commercial visibility inside an AI recommendation are not the same thing. A model can know exactly what your company does while still choosing not to mention it when a user asks for recommendations.

For businesses trying to understand GEO, that distinction is far more useful than simply checking whether ChatGPT knows the brand name.

The most interesting result is sometimes when you are not recommended

The checker does more than look for a company name inside an answer. It also examines whether the website was consulted while the response was being generated.

This creates an especially interesting situation: your website can be read even though your company is not ultimately recommended.

From a GEO perspective, this is probably one of the most useful signals the tool provides.

If an AI assistant never discovers your website, your visibility problem may begin with discoverability, relevance or authority. But if the assistant finds your website, reads it and still recommends three competitors, the problem is different.

At that point, the question is no longer simply "Can the AI find me?"

The better question becomes "Why did it find me and still choose someone else?"

That can lead to a much more interesting investigation.

Perhaps competing companies are supported by stronger third-party references. Perhaps they appear more frequently on comparison websites. Maybe their positioning is easier for the assistant to connect with the user's intent. It could also be that external reviews, directories, publications or other sources provide stronger evidence in favor of those businesses.

A single test cannot prove exactly why one company was preferred, but knowing that your site was considered and not selected dramatically narrows the problem.

This is also why I prefer this approach to a generic AI visibility score.

A score of 42 out of 100 might look impressive inside a dashboard, but it doesn't tell me whether an assistant couldn't find the company, found it but ignored it, cited it as a source or actively recommended it.

Those are completely different situations.

Seeing the sources changes the value of the test

The other part of the PerkFuel checker that deserves attention is the source analysis.

The tool doesn't only show whether your company appeared. It also lets you inspect pages that were consulted when the answer was created.

That is where the checker starts becoming more useful as a research tool.

Imagine an AI assistant recommends three competitors in your market. At first, all you know is that those companies won the recommendation.

Once you look at the sources, however, you may discover that the assistant used a competitor's website, an industry directory, a review platform and a comparison article before generating the answer.

That gives you much more information.

You are no longer looking only at the final recommendation. You are looking at part of the information environment behind it.

This is particularly important because GEO is unlikely to be won exclusively on your own website.

Traditional SEO encouraged businesses to think heavily about their own content, their own pages and their own technical optimization. Those things still matter, but generative search also exposes how important external information can be.

If AI assistants repeatedly consult certain industry directories, those directories become strategically interesting. If comparison articles consistently influence recommendations in your niche, being included in those articles becomes more valuable. If your competitors are reinforced by reviews and independent publications while your company exists mostly on its own website, that difference may eventually influence how assistants describe the market.

The checker therefore gives you something that is potentially more useful than a visibility score: clues about where the assistant is getting its information.

For anyone working seriously on GEO, those clues can become the beginning of an actual strategy.

Competitor analysis becomes much more useful

This also changes how I would approach competitors.

If an AI assistant recommends three companies instead of mine, my first reaction would not necessarily be to publish another ten blog posts.

I would first investigate why those companies keep appearing.

I would look at the external websites mentioning them, the language used to describe their products, the comparison pages where they appear and the sources that assistants repeatedly rely on.

If one competitor is mentioned across several respected websites and another has hundreds of detailed reviews, that tells me something about the evidence available to the AI.

The goal is not to copy those competitors. The goal is to understand why the broader web makes their recommendation easy to justify.

This is an important shift in mindset.

In traditional SEO, competitive analysis often starts with keywords and backlinks. In GEO, competitive analysis may increasingly involve understanding how a brand is represented across the collection of sources that AI systems use when constructing answers.

That is why being able to inspect those sources is more interesting than simply seeing that your competitor was mentioned.

One AI answer is not enough

There is an obvious limitation to any tool attempting to measure AI visibility: generative answers are not perfectly stable.

You can ask the same question twice and receive slightly different responses. A small change in wording can affect the companies mentioned. ChatGPT may also produce a different recommendation from Gemini or Perplexity.

This means one successful answer should never be interpreted as proof that a brand dominates AI search.

PerkFuel handles this reasonably well by expanding the test beyond the first result.

Its full report currently compares ChatGPT, Gemini and Perplexity using two different customer-style questions. That creates six separate answers to examine.

Six responses obviously do not represent every possible AI search around a business, but they are much more informative than a single carefully chosen prompt.

If your company appears in five of six answers, that begins to look like a pattern. If it appears only once, your visibility is probably much less stable. If ChatGPT regularly mentions you while Gemini does not, that difference itself becomes worth investigating.

The value is not in pretending those six answers form a permanent ranking. The value is in using multiple observations to avoid drawing conclusions from one lucky result.

That is exactly how I think AI visibility measurement should develop.

Instead of pretending there is a single fixed position, the goal should be to identify patterns across assistants, prompts and time.

Citation and recommendation should not be treated as the same thing

One of the most useful ideas behind the checker is that it separates being used as information from being presented to the user as a recommendation.

These two events are often mixed together when people talk about GEO.

A marketer might say that a company is "visible in ChatGPT" because one of its articles was cited. Technically, that is a form of visibility, but it does not necessarily mean the brand itself was recommended.

Imagine someone asks for the best tools in a category.

ChatGPT reads one of your articles while preparing the answer, but ultimately recommends three competing products and never mentions yours.

Your content played a role in the research process, but your company did not receive the commercial visibility.

That difference is important.

Being cited indicates that your content entered the assistant's information environment. Being recommended means your brand reached the final answer shown to the potential customer.

For businesses thinking about AI as an acquisition channel, the second event is usually far more valuable.

PerkFuel makes that distinction visible instead of merging everything into one metric.

The tool also has useful limitations

I actually think it is important not to oversell what an AI visibility checker can do.

No tool can currently give you a perfectly complete picture of how every user will see your brand across every possible prompt, model and location.

Two customer questions are still only two questions. Six answers are still only six answers. AI responses can change over time, and the sources used today may not be exactly the same tomorrow.

The fact that a page appears among the sources also does not prove that one specific sentence on that page directly caused the assistant to recommend or reject a business.

There is still uncertainty.

But that is exactly why I like the way PerkFuel presents the test.

The useful output isn't a claim that your company has an objective AI visibility score of 78.4%.

The useful output is evidence.

You see the question that was tested, you see which businesses appeared, you see whether your website was consulted and you see some of the sources involved.

You can then investigate further.

For a field as new and unstable as GEO, I think that is much more valuable than false precision.

Where this fits into a real GEO workflow

The way I would use the tool is not as a replacement for analytics, SEO platforms or manual research.

I would use it near the beginning of a GEO audit.

First, I would run the website through the PerkFuel AI Visibility Checker and look at the customer question the system generated.

Then I would study the businesses that appeared and the sources behind the answer.

If my company was read but not recommended, I would investigate what external evidence supports the companies that were chosen.

If the company was not discovered at all, I would look more closely at its content, entity clarity, relevance and external presence.

After that, I would repeat the process across more questions and more assistants before deciding that a particular pattern is real.

This is where the tool becomes valuable.

It doesn't tell you exactly how to fix GEO.

It tells you where the investigation should begin.

Why I think this is a genuinely useful tool

AI search is creating a measurement problem that SEO tools were never designed to solve.

There isn't always a stable ranking position. The same query can produce different answers. Assistants can use your content without mentioning your business. They can discover your company and still decide that competitors are better recommendations.

Trying to compress all of that into a single score misses most of what makes generative search different.

PerkFuel takes a simpler approach.

It tests a realistic customer question, examines the answer, looks at whether your business was actually selected and exposes some of the sources that contributed to the result.

That doesn't solve every GEO measurement problem, but it turns something extremely abstract into something you can actually inspect.

And that is why I think the checker is genuinely useful.

If you work in SEO, GEO, SaaS, local search or digital marketing, I would recommend testing your own website and then testing a competitor immediately afterwards.

The free checker is available here:

https://perkfuel.io/ai-visibility-check

Don't just look at whether your company appears.

Look at who appears instead, which sources the AI consulted and whether your website was discovered but ultimately passed over.

That is where the useful information begins.

How PerkFuel Generates Highly Targeted Traffic From AI Assistants

Edward Calanni — Sun, 16 Aug 2026 18:31:59 +0000

How PerkFuel Generates Highly Targeted Traffic From AI Assistants ?

When people hear about buying traffic from ChatGPT, Gemini or other AI assistants, the first question is usually:

How does that actually work?

Because simply sending random visitors to a website and labeling them "AI traffic" wouldn't be particularly useful.

The interesting part is everything that happens before the visit.

PerkFuel was built around a different approach: creating an ecosystem where websites can be discovered, referenced and surfaced through environments connected to AI discovery, while matching them with users who are actually interested in what those websites offer.

The system combines several layers:

a large network of already-discovered web properties
contextual link placement
automated website analysis
continuously updated distribution
an active community of real users
targeted acquisition campaigns
support for more than 12 major AI assistants

Here's how the system works.

1. It starts by understanding the website

Before PerkFuel can generate relevant exposure, it needs to understand what the website is actually about.

When a domain is added, PerkFuel analyzes signals such as:

the main topic of the website
products or services offered
important landing pages
page content
commercial intent
target audience
relevant queries and topics

This matters because a cybersecurity SaaS shouldn't be exposed in the same environments as a fashion store or a local restaurant.

The objective is not simply to generate traffic.

It's to identify where that website is most relevant.

2. PerkFuel uses a network spanning 10,000+ domains

One of the core components behind PerkFuel is its distribution network.

The network includes links and content across more than 10,000 different domains already discoverable across the web.

That creates a broad layer of external discovery around websites using PerkFuel.

Instead of relying on one website, one directory or one source, the system can distribute relevant references across many different environments.

This is particularly interesting in the context of AI discovery.

AI assistants don't operate like traditional search engines alone.

When generating answers, they may rely on information discovered across many websites, references, sources and indexed documents.

A website that exists in isolation has fewer external signals surrounding it than one appearing naturally across multiple relevant locations.

3. The network is continuously updated

A static link network would quickly lose much of its value.

PerkFuel's infrastructure is therefore designed to update and expand placements continuously.

There are several sources behind this.

PerkFuel technology

The platform continuously analyzes websites and identifies relevant opportunities within its existing ecosystem.

New pages, websites and contextual environments can be processed as the network evolves.

An active community of real users

PerkFuel also benefits from an active user ecosystem.

That means the network isn't built exclusively around automated infrastructure.

Real users continuously interact with websites, content and AI assistants, creating a much more dynamic environment than a fixed list of domains.

Targeted advertising

Targeted advertising is another acquisition layer.

Instead of trying to attract everyone, campaigns can be aligned with specific interests, topics and audiences.

All three layers work together to keep the system active and continuously refreshed.

4. The algorithm decides where links make the most sense

This is one of the most important parts.

PerkFuel doesn't simply take a URL and distribute it everywhere.

Its algorithm analyzes the destination domain and determines the environments where a reference is most contextually relevant.

Imagine a website selling an AI analytics platform.

Placing that website inside content about:

"best analytics tools for SaaS companies"

could make sense.

Placing the exact same link inside an article about gardening probably wouldn't.

Context matters.

PerkFuel's matching system is designed to identify these relationships automatically.

The objective is to place websites where they have the highest contextual relevance rather than maximizing raw link volume.

5. Context creates a more targeted audience

This has an important consequence.

If a link appears around a topic closely related to the destination website, the people discovering that link are naturally more likely to care about it.

Instead of:

Random user → Website

the objective becomes:

Relevant topic → Interested user → Website

That difference matters enormously.

Imagine two websites receiving 1,000 visitors.

Website A receives random visitors.

Website B receives people already interested in the exact problem the product solves.

Both technically received the same amount of traffic.

But commercially, those visits are completely different.

PerkFuel is designed around the second scenario.

6. AI assistants add another discovery layer

The network becomes particularly interesting when AI assistants are added to the equation.

Today, users increasingly search through interfaces such as:

ChatGPT
Gemini
Claude
Perplexity
Microsoft Copilot
Grok
Meta AI
DeepSeek
Mistral
Poe
Kimi
Manus

Instead of searching:

"best tool for X"

on Google, someone may simply ask an AI assistant the same question.

That changes how discovery works.

AI systems need information, references and sources from which they can understand websites, products and entities.

Creating broader and more relevant web exposure can therefore become another component of an AI visibility strategy.

PerkFuel currently works across 12+ AI assistants, allowing websites to build and measure traffic across a much wider ecosystem than a single platform.

7. From exposure to qualified visitors

The entire process can be simplified into a funnel.

Website analysis

↓

Relevant topics identified

↓

Contextual placements across the PerkFuel network

↓

Discovery through web and AI environments

↓

Matching with interested users

↓

Visit to the website

The important part is the matching layer.

Traffic becomes significantly more useful when the visitor already has an interest related to the website.

That's what turns traffic from a vanity metric into a potential acquisition channel.

8. Why targeted AI traffic matters for conversions

Most businesses don't actually want traffic.

They want what happens after the traffic.

That might be:

a signup
a demo request
a purchase
a subscription
a contact form
a lead
a returning visitor

This is why targeting matters.

Someone arriving on a SaaS pricing page after exploring content related to that exact problem is much more valuable than someone landing there completely randomly.

Better alignment between:

user intent + content + destination

can create better conditions for:

higher engagement
more qualified leads
stronger conversion rates
more opportunities to generate sales

There is obviously no guarantee that a visitor will convert.

But the probability becomes considerably more interesting when the audience is relevant in the first place.

9. More visibility across the web also matters

There is another benefit beyond immediate visits.

A website repeatedly appearing in relevant contexts develops a larger presence across the web.

Instead of one isolated domain, you start building a broader footprint around:

the brand
the product
its category
related questions
relevant topics

This can make the website easier to discover by both humans and automated systems.

For companies thinking about GEO and AI visibility, that's an important distinction.

The objective isn't only:

"Get more traffic."

It's also:

"Make the website easier to discover in the environments where future customers are searching."

10. PerkFuel combines the different layers

This is what makes the model interesting.

PerkFuel isn't relying on a single traffic source.

The infrastructure combines:

Domain analysis

Contextual link distribution

10,000+ domain network

Continuously updated placements

Real user community

Targeted advertising

12+ AI assistants

The result is an acquisition system designed specifically around the way people are starting to discover websites through AI.

Traffic is only the visible part

When you look at Google Analytics, you only see the final step:

the visitor arriving on your website.

But behind that visit there can be an entire discovery infrastructure.

That's the part PerkFuel is trying to solve.

Rather than simply selling pageviews, the platform is designed to create relevant exposure, connect websites with interested audiences and generate measurable traffic from the growing AI ecosystem.

If you want to see how it works on your own website, you can create a PerkFuel project and start testing AI traffic here.

The setup only takes a few minutes, and you can then monitor the resulting traffic directly from your own analytics.

And as AI assistants continue becoming a normal way to discover products and websites, understanding how this acquisition channel works today could become increasingly valuable tomorrow.

How to Buy Traffic From ChatGPT, Gemini and Other AI Assistants

Edward Calanni — Sun, 16 Aug 2026 18:21:35 +0000

AI assistants are becoming a real traffic source.

People are no longer discovering websites only through Google, ads or social media. They're also clicking links suggested by ChatGPT, Gemini, Perplexity, Claude, Copilot and other AI platforms.

You can already see some of that traffic inside tools like Google Analytics.

But what if you don't want to wait for it to happen organically?

You can now buy AI referral traffic directly.

What is AI referral traffic?

The idea is simple:

AI assistant → your website → analytics

Instead of buying generic pageviews or low-quality traffic, you're generating visits associated with AI platforms such as:

ChatGPT
Gemini
Claude
Perplexity
Microsoft Copilot
Grok
DeepSeek
Mistral
Meta AI
Poe

This makes it possible to test AI as a real acquisition source and see how this traffic behaves on your website.

The easiest way to buy AI traffic

This is exactly what PerkFuel was built for.

PerkFuel lets you send traffic from major AI assistants to your website without having to build anything yourself.

Setup takes only a few minutes.

You create a project, add your website, choose where you want the traffic to go, select your sources and launch.

That's basically it.

How it works

1. Add your website

Create a project and enter your domain.

2. Choose your AI sources

You can select the assistants you want traffic from.

For example:

ChatGPT + Gemini + Perplexity

Or distribute your traffic across a wider range of AI platforms.

3. Select your landing pages

You can send traffic directly to the pages that matter.

For example:

your homepage
a product page
your pricing page
a blog post
a specific landing page

This is especially useful if you're testing content designed for GEO or AI search.

4. Choose your target countries

You can control where visitors come from instead of receiving completely random traffic.

5. Launch

Once the project is configured, PerkFuel starts delivering the traffic automatically.

No campaign setup across multiple advertising platforms.

No complicated integration.

No code to deploy.

Why use PerkFuel instead of a traditional traffic provider?

Most traffic marketplaces weren't built for AI.

They usually sell generic:

direct traffic
search traffic
social traffic
display traffic

PerkFuel focuses specifically on traffic coming from AI assistants.

That means you can test a channel that is becoming increasingly important without buying thousands of unrelated pageviews.

You also keep control over things that actually matter:

AI source, country, landing page and delivery.

You can see the traffic in your analytics

One of the most useful parts is being able to analyze what happens once the traffic reaches your website.

In GA4, for example, you can monitor:

traffic sources
landing pages
events
engagement
conversions
countries
user journeys

This makes AI traffic much easier to compare with channels such as Google Organic, Reddit, paid ads or direct traffic.

Start small

You don't need to buy thousands of visits immediately.

A small campaign is enough to check:

whether your tracking is working
how AI traffic appears in analytics
which pages you want to promote
which AI sources you want to focus on

Then you can scale from there.

PerkFuel also makes it easy to test the concept before committing to a larger campaign.

AI traffic is not the same as AI rankings

One important distinction:

Buying AI traffic does not magically guarantee that ChatGPT will start recommending your website.

Traffic and AI visibility are two separate things.

GEO focuses on helping AI systems understand, cite and recommend your content.

AI referral traffic focuses on the visitors actually reaching your website through those platforms.

Both can be part of the same acquisition strategy, but they shouldn't be confused.

The new acquisition channel

For years, website traffic mostly looked like this:

Google → Website

Social Media → Website

Ads → Website

Now there's another path:

AI Assistant → Website

And as more users rely on AI tools to discover products, services and information, measuring that channel becomes increasingly useful.

If you want to test it without waiting months for organic AI referrals to appear, PerkFuel lets you start generating AI referral traffic in minutes.

You can create a project, choose your sources and start testing the channel directly on your own website.

If you're already tracking AI traffic in GA4, I'd also be interested to know which source you're seeing the most:

ChatGPT, Gemini, Perplexity or something else?

How to Track AI Referral Traffic from ChatGPT, Claude, Gemini and Perplexity in GA4 ?

Edward Calanni — Thu, 13 Aug 2026 14:47:36 +0000

AI assistants are becoming a measurable source of website traffic.

People now use ChatGPT, Claude, Gemini, Perplexity and other AI assistants to research products, compare tools, solve problems and discover websites.

When someone clicks a link suggested by one of these assistants, the visit may appear inside Google Analytics 4 as referral traffic.

That means you can already start identifying AI traffic and analyzing how those visitors behave on your website.

Here’s how.

Where AI referral traffic appears in GA4

Open Google Analytics 4 and go to:

Reports → Acquisition → Traffic acquisition

Then change the primary dimension to:

Session source / medium

This gives you a more detailed view of where individual sessions originated.

Depending on the assistant and how the visit was referred, you may see sources associated with platforms such as:

ChatGPT
Perplexity
Claude
Gemini
Copilot
other AI assistants

Not every visit influenced by an AI assistant will necessarily appear as an AI referral.

For example, someone might discover your company through ChatGPT, remember the brand name, and search for it on Google later.

GA4 would likely attribute that session to search rather than to ChatGPT.

So AI referral traffic should be seen as the observable portion of AI-driven discovery, not a complete measurement of AI influence.

See how AI referral traffic appears inside Google Analytics

Here’s a quick walkthrough showing where AI referral traffic can appear inside GA4 and how to find it:

Once you know where to look, it becomes much easier to separate AI referrals from your other acquisition sources.

Why separate AI traffic from other referrals?

If every referral is grouped together, you lose information about how individual sources perform.

Separating AI traffic lets you analyze things like:

which AI assistants are sending visitors
which pages those visitors land on
how long they stay
whether they continue browsing
which sources generate engaged sessions
which AI assistants send the most valuable visitors

This becomes particularly useful when several AI assistants are referring traffic to the same website.

For example, visitors arriving from ChatGPT may behave differently from visitors coming from Perplexity.

The number of sessions might be similar, while engagement or landing-page behavior can be completely different.

Look at landing pages, not just traffic volume

Traffic volume alone doesn’t tell you very much.

The landing page often gives you a much better idea of the visitor’s intent.

Imagine an AI assistant sends someone directly to:

/pricing

That visitor probably has a very different intent from someone landing on:

/blog/what-is-seo

That’s why it makes sense to analyze the combination of:

AI source + landing page + engagement

rather than simply counting visits.

A relatively small number of highly targeted visits can sometimes be much more interesting than a larger amount of generic referral traffic.

Compare AI traffic with other acquisition channels

Once you’ve identified your AI referral sources, you can compare them against channels such as:

Organic Search
Direct
Referral
Organic Social
Paid Search

Useful metrics to monitor include:

total users
new users
sessions
engaged sessions
average engagement time
events
conversions
landing pages

You can also create a dedicated GA4 exploration or comparison containing only the AI sources you want to monitor.

Over time, this makes it easier to understand whether AI assistants are becoming a meaningful acquisition channel for your website.

Keep a list of AI sources to monitor

The number of AI assistants people use continues to grow.

I find it useful to maintain a simple list of platforms worth monitoring.

For example:

ChatGPT
Gemini
Claude
Perplexity
Copilot
Grok
Meta AI
DeepSeek
Mistral

PerkFuel has published a more detailed guide on tracking AI referral traffic in GA4, based on its work analyzing how traffic from different AI assistants appears inside analytics platforms.

The observations and examples in that research are also useful when building your own AI referral tracking setup.

AI attribution is still imperfect

This is probably the most important limitation to understand.

A visitor’s journey could look like this:

Someone asks an AI assistant for a recommendation.
The assistant mentions a website.
The person doesn’t click immediately.
They remember the brand.
They search for it later.
They visit through Google.

The AI assistant influenced the discovery, but GA4 may never classify that visit as AI referral traffic.

The same problem already exists with podcasts, social media, word of mouth and many other discovery channels.

So AI referrals shouldn’t be treated as a perfect measurement of AI visibility.

They’re better understood as one measurable signal.

Why AI referral traffic is worth watching

The interesting question isn’t necessarily whether AI assistants will replace Google.

It’s that they’re becoming another way people discover websites.

Search traffic already has its own analytics.

Social traffic has its own analytics.

Paid traffic has its own analytics.

Referral traffic does too.

AI traffic is starting to deserve the same treatment.

As ChatGPT, Gemini, Claude, Perplexity and other assistants become a larger part of how people research products, companies and information online, understanding how that traffic behaves will become increasingly useful.

If you’re already seeing traffic from AI assistants in GA4, take a look at the landing pages and engagement metrics rather than focusing only on the number of sessions.

That’s where the interesting patterns usually start to appear.