Generative engine optimization · Methodology
ChatGPT picks different sources in the API and the browser
We put three identical owner questions to ChatGPT twice: one time via the API, the next through the live browser interface. We got zero matching domains, and it wasn't even a near miss.
That gap is a bigger deal than you might think. Every tool that claims to show you "what ChatGPT says about your business," including our own free check, uses the API. But what an actual customer sees when they go to chatgpt.com and type in that same question is a totally different set of sources.
What we tested
We used three real questions from a business owner: why is my business not showing up on ChatGPT, my competitor shows up but I don't, and how do I check if my business appears in AI search results. We already had citation data for these from our own API sweeps. Then we ran the exact same three questions through the browser using a different tool made for that, and we put the source lists next to each other.
Here's what we found
For "why is my business not showing up," the API cited fonzy.ai, aeoaction.com, planwebbies.ca, workmatix.com, and reddit.com. The browser pointed to help.openai.com, sandbox.co, and shiftflow.app.
For "my competitor shows up but I don't," the API surfaced eight sources, including reddit.com and six niche marketing sites. The browser surfaced just one: help.openai.com.
For "how do I check if my business appears," the API pulled twelve sources, mostly GEO-tool sites like scanmygeo.com, getcitely.com, and searchscore.io. The browser pulled two: help.openai.com and support.google.com.
The API produced twenty-five citations, the browser produced six, and not a single domain appeared on both lists.
The pattern we saw
The browser leaned on official platform documentation every single time. help.openai.com was in all three browser results. The API version never cited it once. Instead, the API pulled a scattered mix of marketing blogs, Reddit threads, and small GEO-tool startups, the exact kind of sites that rank for those search terms in regular Google.
This suggests something about how the two systems work. The browser interface seems to prioritize official sources for questions that sound like support or how-to queries. The API, which runs a live web search without that specific context, acts more like a basic search engine, just pulling whatever ranks high.
What this means if you're measuring AI visibility
If a tool tells you which domains ChatGPT cites for your industry, you need to ask which layer it measured. API-based tools, ours included, are inexpensive and consistent, which is great for tracking trends over time. But they are not a preview of what your customer actually sees when they type the question in themselves.
The honest fix isn't to choose one over the other. It's to label which one produced the data you're seeing, and to treat API data as a directional signal instead of a screenshot of reality.
We're building both paths into our own tracking from now on, and every report we put out will state which layer the citation came from.