← Videos

How We Built Our Own AI Visibility Stack With Bright Data

How Airfleet built its own AI visibility and GEO stack, why we run it on Bright Data, and what our B2B clients gain from it.

Elad Hefetz Elad Hefetz
22 August 2026
GEO

Quick Jump

0:00
Why off-the-shelf AI visibility tools couldn't scale to a full category
0:45
Why an API answer differs from what users see in the browser
0:58
The browser-automation problem that kept breaking
1:12
Choosing Bright Data for scale and stability
1:42
One API across ChatGPT, Perplexity, Gemini, and Claude
1:56
Turning it into a browser layer for AI agents
2:20
How the content pipeline researches and builds client content
3:10
What large-scale research reveals about AI visibility
4:08
Why more of your traffic is bots than humans

Most B2B marketers check whether their brand shows up in ChatGPT, Claude, or Perplexity the same way: run a handful of prompts, read the answers, move on. That’s fine for a spot check. It falls short the moment you need to know how a whole category shows up across every major AI model, at the scale of thousands of prompts a day.

So we built our own tools for it. And to run those tools, we needed something underneath them that wouldn’t fall over every time an AI model changed its interface.

Here’s what we built, why Bright Data powers it, and what our clients get out of it.

TL;DR

  • We built our own AI visibility platform and AI content pipeline because off-the-shelf GEO tools couldn’t scale to analyze a full category. AI models answer differently through an API than they do in a live browser, and it was important for us to see the answer humans see.
  • Running real browsers at scale was really hard. Captchas, breakpoints, and constant interface changes kept breaking our scrapers. Bright Data took that problem off our hands, and it keeps running when  ChatGPT, Perplexity, Gemini, or Claude models change.
  • The same infrastructure doubles as a browser layer for our AI agents, so our content pipeline can research subreddits, competitor sites, and search results the way a person would.
  • Client content is built with real-time research, we understand what actually moves AI answers across the funnel, and can analyze at a scale you can’t reach by hand.
  • The bigger point: your site already gets more bot traffic than human traffic, and the brands that design for that audience now are the ones AI keeps surfacing later.

We couldn’t get the GEO analysis we needed, so we built it

When we set out to track how B2B brands surface across AI models, we started with the AI visibility tools already on the market. For watching a short list of prompts, they did the job. To understand an entire category, we needed to analyze thousands of prompts across many companies, subtopics, and regions. That’s when these tools hit their limit or charged per prompt until the idea stopped making sense.

So we built an AI visibility platform we could push as wide as a category required without the meter running on every prompt.

Building it exposed a problem most marketers never run into. What an AI model returns through its official API is not what a real person sees when they ask the same thing in the app.

When we started using the APIs, we quickly realized that querying ChatGPT through an API gives a very different answer than prompting it on the web. There’s a layer on top of the model that changes how it answers.

Elad Hefetz
CEO @ Airfleet

An API response can read as authoritative and still be the wrong thing to optimize against, because it isn’t the version your buyer will ever read. The only way to get the real answer is to collect it the way a person would, in a live browser. That is the part that got hard.

It was very hard to automate that with real browsers. Captchas, breakpoints, new models. It kept breaking.

Elad Hefetz
CEO @ Airfleet

Running and repairing browser automation across every model, through every interface change, is a standing engineering job. We’re an agency. We wanted the data those tools produced, not a maintenance queue to babysit.

Why we run it all on Bright Data

Our shortlist came down to two questions: can it scale, and will it stay online. Everything else was a detail.

We’re not a product company. We build products for the agency so we can deliver services. We don’t have time to tinker or figure out why something isn’t working. We need something that just works.

Elad Hefetz
CEO @ Airfleet

Bright Data lets us keep our focus on client work instead of maintaining products. It was easy to implement, scaled quickly, and gave us consistent visibility across ChatGPT, Perplexity, Gemini, and Claude. It’s also reliable. When a model reworks its interface, our monitoring keeps collecting data instead of breaking, which means we’re not focused on troubleshooting.

The surprise was how far that reliable browser layer could stretch beyond monitoring. Most AI can’t freely read the open web on its own.

People assume any AI is connected to the internet. It isn’t. If you want an agent to do real research on the web, you have to build a browser layer for it. That’s hard to build, and Bright Data just provides it. It basically opens up the internet for the AI.

Elad Hefetz
CEO @ Airfleet

How our clients benefit from these tools

That covers why Bright Data sits inside our visibility platform. The bigger story for the people we work with is everything else we’ve been able to point it at.

The same technology now plays a big part in our Content Pipeline tool. When it builds a client’s content, it reads what’s ranking right now, follows what a category is arguing about on subreddits, and checks what competitors have just changed. An AI agent working on its own would be shut out of most of those pages. With Bright Data handling access, the drafts our clients get are grounded in what’s true this week, not in whatever the model happened to absorb during training.

It also drives the research that tells us where a client’s effort is worth spending. We ran a study across hundreds of thousands of AI responses in the cybersecurity category to learn what actually influences whether a brand shows up in an AI answer, from broad awareness questions down to decision-stage comparisons.

If you run one prompt in ChatGPT, you see the sources and think that’s what to optimize for. But at scale, across thousands of prompts and dozens of categories, you learn what really makes the difference, which is often not what you expected.

Elad Hefetz
CEO @ Airfleet

What we think should correlate to influence is typically wrong. Awareness-stage questions rarely name any brand or link back to a branded website, so focusing on awareness-level content is a waste of time. What actually influences brand mentions in LLMs at the consideration and decision stages are the peer review and comparison sources buyers already trust.

You don’t get to that realization by skimming a few prompts. It only surfaces when you can run a full category at volume, which is what this stack makes possible.

The partnership matters as much as the product

Dependable infrastructure is the baseline. What moved Bright Data from vendor to partner was the people behind it. Early on, someone there reached out, gave us space to actually put the product through its paces, and asked what we were trying to build rather than focusing on a standard pitch. That’s where the working relationship really started, around our use case and what our clients needed.

A brand is a lot more than one person or one experience.

Elad Hefetz
CEO @ Airfleet

We bring it up because it maps onto our own work. The tools matter, but the judgment about where to aim them matters more.

Your Next Audience Isn’t Human

Here’s the shift every B2B marketer should be thinking about. The traffic reaching your site is already more machine than human, and that split is only going to widen.

The internet already has more bots than humans visiting websites. A lot of your future traffic will be agents, non-human visits. If you only think about the human experience, you’re missing a big part of your traffic. So think about the new persona: the AI agents, even the bot itself, and how you cater to their needs, not just your human visitors.

Elad Hefetz
CEO @ Airfleet

That’s the real reason we built this. Not to run more prompts, but to understand and serve an audience most sites still act like isn’t there. The brands that start designing for it now are the ones AI will keep recommending later. We built the stack, and partnered with Bright Data to power it, so our clients are in that group.

Register to our Newsletter

Stories and trends of tech b2b marketing websites, zero fluff, tons of fun stuff

Thank you for submitting the form.

Related posts

Why B2B AEO Needs a Different Playbook for Every Channel
Why-B2B-AEO-Needs-a-Different-Playbook-for-Every-Channel-2

Why B2B AEO Needs a Different Playbook for Every Channel

The Trick to GEO: Remember Your Audience Is Human
GEO for B2B

The Trick to GEO: Remember Your Audience Is Human

How Airfleet’s Approach to GEO for B2B Tech Differs From Other Agencies
Airfleet's approach to GEO and AEO

How Airfleet’s Approach to GEO for B2B Tech Differs From Other Agencies

Expecting more from your website?

So do we.Β Let’sΒ chat about ways to improve your buyer journey.