How Airfleet Scaled AI Visibility Monitoring With Bright Data

The infrastructure behind our GEO practice. Thousands of prompts a day, across every major LLM, without babysitting a single scraper.

TL;DR

• Airfleet built its own AI visibility monitoring to see how brands show up across ChatGPT, Claude, Gemini, and Perplexity
• Off-the-shelf tools couldn’t reach the scale or detail we wanted, and our own browser-based scrapers kept breaking on captchas, model swaps, and interface changes.
• We run the whole operation on Bright Data now. Thousands of prompts a day, one consistent API across every LLM, and it holds up even when the AI platforms change their interfaces.
• The fit was technical and human. The product scaled fast, and a Bright Data marketer’s genuine interest in our use case turned our curiosity into a real partnership.
• It’s since become core infrastructure across our SERP research, content pipeline, and agent workflows, including a cybersecurity study that ran close to a million prompts.

We run our AI visibility work on Bright Data because it works

Airfleet is a B2B website development, design, and services agency. We build and optimize websites that generate pipeline for B2B technology companies, and a growing part of that work is understanding how our clients show up inside AI answers and LLM tools, not just search.

To do that well, we run our own AI visibility monitoring across the major LLMs, including ChatGPT, Claude, Gemini, and Perplexity. We watch how brands get cited, which sources move those answers, and what actually changes an outcome. The engine underneath all of it is Bright Data.

We’ll get to how we ended up building our own tool. But the punch line is that Bright Data has helped us scale and minimized the annoying gaps in data we had when we used other GEO monitoring tools on the market. We run thousands of prompts a day through it, and we don’t spend our time keeping the tool running.

“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 to figure out why something isn’t working. We need something that just works.” – Elad Hefetz, CEO of Airfleet

What we set out to build, and where it first broke

We started with a straightforward goal. We wanted to see how brands appear across AI answers at the scale of a whole category, not a tracked list of thirty or a hundred prompts.

We looked at the existing AI visibility tools first. Most of them were solid. The problem was the ceiling. If you want to understand an entire industry, you need thousands of prompts across many companies, many subtopics, and many regions, and the off-the-shelf tools either couldn’t go that wide or charged per prompt to get there. So we decided to build our own.

That’s when we hit the real problem. When we started pulling answers through the official APIs, the answers didn’t match what a real user sees.

“We quickly realized that querying ChatGPT through an API gives you a very different answer than prompting it on the web. There’s a layer on top of the model that changes how it answers. Same for Gemini, Claude, and the rest.” – Elad Hefetz, CEO of Airfleet

If the API answer isn’t the answer your buyer sees, it’s the wrong data. So we did the obvious thing and started driving real browsers, acting like a human to capture the real response. And that is where it broke.

Captchas. Breakpoints. New model releases. Interface changes. Any one of them could take a scraper down, and keeping a fleet of browser automations alive across every LLM turned into a full-time job we didn’t want. We tried a cheaper alternative to carry that load first. It didn’t hold up the way we needed.

Why we chose Bright Data

Two things mattered in the selection, and they were the same two things that had been breaking us: scale and stability.

Bright Data solved both quickly. The implementation was simple, and the API was consistent across LLMs, so one call pattern covered ChatGPT, Claude, Gemini, and Perplexity instead of a separate integration for each. Once it was working, it stayed working.

“Even when ChatGPT changes its interface, it keeps working. We don’t have to build fallback mechanisms for when a scraper breaks. We just use it.” – Elad Hefetz, CEO of Airfleet

There was an upside we didn’t plan for. Alongside the answers, we started getting metadata we didn’t think we could pull from a browser session at all.

“We get the final search queries and the sources ChatGPT used to build its answer. That’s the information we actually optimize against.” – Elad Hefetz, CEO of Airfleet

The fit wasn’t only technical. Elad came in through the free tier to test whether the approach would even work, curious more than committed, the way most early evaluations start. Then a marketer at Bright Data reached out. They’d noticed he had started and then paused, sent over credits to try again, walked him through how other companies were using the product, and asked, genuinely, what he was trying to do.

“He showed real interest in what we were building, and that completely changed how I saw Bright Data as a company.” – Elad Hefetz, CEO of Airfleet

That conversation is where the partnership actually started, built around our use case rather than a contract.

Bright Data became core infrastructure, not just a scraper

We brought Bright Data in to automate one thing: browser-based prompting for ChatGPT. It didn’t stay in that lane.

Once we had a reliable browser layer for AI, we started using it everywhere that layer was missing. We use it to read Google search results pages inside our SEO work. We use it as the browser behind our AI agents, so the agents can reach the open web instead of getting blocked.

“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.” – Elad Hefetz, CEO of Airfleet

That layer now feeds our content pipeline directly. When the pipeline researches a topic, checks what’s ranking, monitors subreddits for what a category is actually talking about, pulls signals from public LinkedIn activity, or scrapes a competitor’s site for what’s changed, Bright Data is doing the fetching underneath. If our AI tried to hit those sources on its own, it would get blocked. Bright Data sits between the AI and the internet and opens the web back up for it.

A cybersecurity study we couldn’t have run any other way

The clearest example is a research project in the cybersecurity category. We wanted to understand what actually influences AI answers across the funnel, from broad top-of-funnel questions down to bottom-of-funnel, buying-stage prompts. That takes volume no off-the-shelf tool could give us.

We used Bright Data to run the monitoring at the volume the study required: 371,000-plus AI responses, across 21 cybersecurity companies and more than 5,000 third-party sources, looped over time and across ChatGPT, Gemini, and Perplexity. Then we analyzed what the AI cited at each stage. A few findings stood out:

• Review platforms are the biggest lever. G2, Peerspot, Capterra, Software Advice, and TrustRadius showed the highest lift on brand mentions at the consideration and decision stages, often tens of percentage points above the stage baseline. When AI cites one of them, it usually names a specific brand alongside it.
• Reddit is the most-cited source at every funnel stage. AI leans on real user discussion more than almost anything else when it decides which brands to surface.
• Wikipedia earns its citations for definitions, not recommendations. It shows up for educational queries, but its influence on brand mentions turns negative as prompts get more commercial.
• Top of the funnel is thin ground. With a 7.9% baseline mention rate, broad awareness prompts rarely surface a specific brand at all, which makes them the wrong place to measure whether you’re winning.

None of those conclusions come from eyeballing a handful of prompts. They come from running the category at scale, which is exactly what Bright Data made possible.

Where we’re taking Bright Data next

AI visibility monitoring is now a standing part of how we work, not a one-off study, and Bright Data is the layer we build the rest on. As more of the traffic on the web shifts from humans to agents and crawlers, the ability to see the web the way those systems see it, and at their volume, only gets more central to what we do. We plan to keep widening what we monitor and automate on top of it.