I think one of the stranger parts of GEO is that someone else can explain your company before you can.
That is fine when the information is correct. It is a problem when it is not.
My name is Giorgi Mikhelidze, and I’m a GEO / AI Search Senior Manager at Remote. The project I built is called the External Fact Checker. It helps us find third-party pages that mention Remote, flag potential inaccuracies for human review, and route anything that needs the right internal review.
Finding and fixing inaccurate info across the web
Companies all over the world are growing and hiring beyond their domestic borders. They are actively looking for global payroll, employer of record, and other products Remote offers. Increasingly, they are learning about those options through search, comparison pages, AI summaries, and their personal agents.
That means that the wider web plays a huge role in how people discover Remote. A lot of that discovery happens on pages we do not own: comparison articles, software roundups, partner pages, and search results that AI tools may pull into their answers. If those pages are outdated or missing context, people can walk away with the wrong understanding of Remote. Until now, we mostly found those issues randomly, when someone on the team happened to see a page and shared it with whoever needed to look at it.
This project came out of Remote’s internal AI training, which is now available publicly as AI for Actual Work. The course is about building AI workflows around real work, and this was a real one for me: find inaccurate external content faster, keep human review in the process, and make it easier for the right people to act on it.
How the agent works
The current version starts with a list of URLs.
For example, I can pull 20 pages that are ranking for a topic where Remote should show up. I give those URLs to the Claude agent, and it goes through them to flag potential inaccuracies about Remote for review. The agent does the first pass; a person still checks the output before anything happens.
Once the scan is done, it creates a Notion page with the results. It shows how many URLs were analyzed, how many had potential issues, what the issues were, and what kind of review may be useful.
Some issues need a human review. Some need review from the appropriate team. Some may simply need outreach to a website owner with the correct information.
After that, the agent posts into a Slack channel I created for the workflow. The Slack message gives the team the summary and, when possible, contact information for the website. So instead of someone discovering a wrong page by accident, we have a place where these issues can land and be picked up.
Why this matters for GEO
LLMs rely on the web around us, not only on Remote.com. As more companies use AI summaries and AI agents to compare vendors, the information out there about Remote becomes even more important.
If a third-party page has the wrong information about Remote and that page is visible for an important query, it can become part of the answer people get from search or AI tools. That can create confusion for prospects, customers, partners, or candidates.
The hard part is not knowing that this is an issue. The hard part is finding the pages at scale.
Manual checking takes a lot of time. You need to pull the URLs, open each page, read the content, check the claims, decide what matters, and then send it to the right team. Doing that across enough keywords and pages becomes a lot of work very quickly.
The External Fact Checker gives us a faster first pass. It can go through a batch of URLs in a couple of minutes and show where we should spend our time. The agent speeds up discovery, but humans stay in control. It surfaces possible issues; we review them and decide whether to correct, escalate, or ignore.
What I want to improve next
The biggest limitation right now is the starting point.
At the moment, I still have to download a CSV and give it to Claude. That works for a first version, but it is not the final workflow I want.
Next, I want to connect the workflow more directly to the tools we already use for search and AI visibility, so the list of pages can be pulled and reviewed with less manual setup.
That would make the workflow much closer to automatic.
I think this will be useful for GEO, product, and legal. GEO can see what external pages are saying about Remote. Product can catch wrong claims about features or capabilities. Legal can see which issues need a stronger response.
The internet will keep creating content about Remote. My goal is to make sure we are not waiting for someone to stumble across the wrong info by accident.
If you want to build something similar for your own work, sign up for Remote’s free AI for Actual Work course at aiforactualwork.com.
