AI for marketing

Is AI Recommending Your Company? How to Check, and What to Fix

A company website on screen

A lot of buyers now start with a question instead of a search. “Who makes good training videos in Raleigh?” goes into ChatGPT, and the answer comes back as a short list of names. If your company isn’t on that list, you may never know you were in the running.

Google is doing the same thing at the top of its results. When Pew Research tracked the browsing of 900 US adults in 2025, people clicked a regular search result on 8% of visits when Google showed an AI summary, and on 15% when it didn’t. The summary is doing more of the choosing.

So this fall I built a system to watch this for BlueForest. Every week it asks ChatGPT, Claude, Perplexity, Gemini and Google’s AI Overviews the same 30 questions a buyer of video production might ask. It records which companies each answer names and which web pages it cites. I built it with Claude Code, and the AI answers cost about $5.50 a month.

You don’t need to build anything to get most of the value. Here is what I’d do first.

1) Write down the questions your buyers ask

Use the words from your sales calls and contact forms, not your keyword list. Buyers ask in three ways, so include some of each:

  1. Category: “best commercial cleaning companies in Charlotte”
  2. Problem: “how do I make our safety training less boring”
  3. Comparison: “is it better to hire an agency or freelancers for video”

Ten questions is plenty to start. We track 30.

2) Ask them in three or four assistants

Ask in ChatGPT, Gemini, Perplexity and Claude, logged out or in a private window so your history doesn’t steer the answer. For each one, note three things: are you named, who is named instead, and which pages are cited as sources.

That last column is the useful one. The cited pages tell you where these assistants get their information about your industry.

3) Check again every month, with the same questions

AI answers change from one run to the next, so a single check means very little. What matters is the pattern over several weeks. With 30 questions we still treat small swings as noise.

4) Make sure the AI search bots can read your site

This one is easy to miss, and it is the fix with the strongest evidence. OpenAI, Anthropic and Perplexity each run separate bots for search and for training:

  • Search bots (OAI-SearchBot, Claude-SearchBot, PerplexityBot) are how your pages show up in their answers. Blocking them hides you.
  • Training bots (GPTBot, ClaudeBot) collect material to train future models. Blocking those is a separate decision.

All of them follow your site’s robots.txt file. If someone on your team blocked “AI bots” across the board last year, you may have taken yourself out of the answers without meaning to. Ask whoever runs your website to check it.

5) Get listed where the answers come from

When the answers cite a directory, review site, trade publication, Reddit thread or YouTube video, that is where the work is. Make sure your listings on those sites exist, are accurate and say what you actually do. We now check every week whether we appear on 12 sources: 10 directories plus Reddit and YouTube.

One caution from Google’s own guidance: chasing fake or inauthentic mentions across the web doesn’t help as much as people hope. Real listings and real customer reviews are the point.

6) Skip the fixes that aren’t backed up

A lot of AI search advice is guesswork. Before spending time on these, know what the companies themselves say:

  • llms.txt files. Google says it ignores them. OpenAI, Anthropic and Perplexity don’t say they read them either.
  • FAQ markup. Google stopped showing FAQ rich results in 2026, so the markup has no documented benefit in Google Search.
  • Writing “for AI.” Google says you don’t need to write in a special way for its AI features, or break your content into tiny pieces.

What still works is the boring part: clear pages that answer the real questions your buyers ask, in plain language, with specifics.

What this costs you

Steps 1 through 3 take about an hour the first time and less after that. Step 4 is one email to your web person. Step 5 is a few afternoons. None of it needs new software.

If you’d like help setting up a check like this for your company, it’s one of the AI systems we now build and train teams to run as part of our monthly retainers. I’m happy to schedule a 15-minute call to talk it through: book a time here.

And if you’ve found something that actually moved the needle for you, I’d like to hear about it.