How to track when ChatGPT or Perplexity recommends your competitor

Last month I lost a deal to a competitor I'd never heard of. The prospect told me, almost apologetically, that they'd asked ChatGPT for "the best tool for managing customer onboarding" and my company didn't come up. You can't see that in your CRM. You can't see it in your analytics. But you can simulate the buyer's question, and that's exactly what I started doing.

Here is the uncomfortable truth: AI assistants are now a distribution channel, and most founders have zero visibility into it. You can check your Google rankings in five seconds. You can see who mentioned you on X. But when a buyer asks Perplexity "what's the best alternative to [competitor]," you have no idea if your name appears, what it says about you, or which source convinced the model to say it.

The prompts that actually matter

Stop thinking about "generative engine optimization" as a concept. Start thinking about the ten questions your best customers actually type. Not the ones you hope they type. The ones they type when they're tired, slightly confused, and looking for a shortcut.

Here is the prompt set I run every Monday morning, roughly 15 minutes total:

• "best [category] software for [use case]"
• "alternatives to [top competitor]"
• "[competitor] vs [my company]"
• "what do people use for [problem]"
• "is [my company] any good"
• "cheapest way to [job to be done]"
• "tools like [competitor] but for [niche]"

Run each one in ChatGPT, Perplexity, Gemini, and Google AI. Yes, all four. They give different answers because they cite different sources. Perplexity leans hard on Reddit and recent forum threads. ChatGPT tends to synthesize from a broader web corpus. Gemini pulls from Google's index in ways that feel almost like old-school SEO. Google AI Overviews, when they trigger, are a whole different animal.

The first time I did this, I found that Perplexity recommended a competitor I'd never tracked because someone on Reddit had written a detailed comparison six months ago. That single thread was doing more pipeline damage than a bad review on G2.

How often the answer changes

More than you think. In my experience, recommendations shift weekly, sometimes daily, for categories with active communities. A new Reddit thread with strong opinions can flip a Perplexity answer within days. A well-argued Hacker News comment can show up in ChatGPT's synthesis a week later.

I started tracking this manually in a spreadsheet. Column A: the prompt. Column B: the date. Column C: which tools were recommended. Column D: which sources were cited. Column E: whether my company appeared at all.

After three weeks, the pattern was obvious. The same three or four sources kept appearing in the citations. A Reddit thread from 2024. A comparison blog post from a competitor. A LinkedIn post from a practitioner. The models weren't making independent judgments. They were aggregating a small number of high-signal conversations and treating them as ground truth.

This is what I now call share of model: the percentage of AI-generated answers in your category that mention you, recommend you, or cite a source that does. It's not a metric anyone asked for. It's a metric the distribution shift forced on us.

Which cited threads moved the answer

Here is the part most "GEO" content skips. It's not enough to know that ChatGPT recommends your competitor. You need to know why. And the why is almost always a specific thread, post, or article that the model has decided is authoritative.

When I dug into the citations for my category, I found three sources doing the heavy lifting:

• A Reddit thread titled "What do you all use for [problem]?" with 200+ comments. The top comment recommended my competitor with specific, credible detail. That thread was cited by Perplexity in 4 out of 5 test runs.
• A blog post from a competitor comparing themselves to three alternatives. They'd written the comparison in a way that made them look like the obvious choice. ChatGPT cited it repeatedly.
• A LinkedIn post from a founder who'd switched tools and explained why. One post. Six months old. Still showing up in Gemini answers.

None of these were things my keyword alerts would have caught. I wasn't tracking "what do you all use for" threads. I wasn't monitoring LinkedIn for competitor mentions. I was doing keyword alerts that kept failing me because I was thinking in terms of brand mentions, not buying conversations.

The anatomy of these threads matters. They're not reviews. They're not complaints. They're people asking for recommendations in plain English, and other people answering with the confidence of someone who's actually used the tool. That's the exact shape of a "what do you use for" thread, and it's the highest-converting content on the internet right now.

What to do when you're not showing up

You have three levers, and they're all unglamorous.

First, get into the conversations that are already being cited. If a Reddit thread is driving Perplexity recommendations, you need to be in that thread. Not spamming. Not dropping a link. Actually answering the question with enough specificity that a model would quote you. This is the difference between finding customers on Reddit and becoming the guy everyone downvotes.

Second, write the comparison you wish someone else would write. If your competitor has a blog post comparing themselves to you, and you don't have one comparing yourself to them, you're letting them define the frame. Write the honest version. Include the tradeoffs. Models reward specificity, not cheerleading.

Third, track it like a channel. Share of model is not a one-time audit. It's a weekly check, same as your pipeline review. The answers change. The sources change. The threads that mattered last month might be replaced by a new one next month.

I built ThreadSonar because I got tired of doing this manually in a spreadsheet. It watches for the conversations that AI models cite, scores them for buying intent, and tells me when my share of model moves. But the manual version works. It's tedious. It's imperfect. It's also the only way I know to see what a buyer sees when they ask an AI for a recommendation.

The founders who win the next two years won't be the ones with the best SEO. They'll be the ones who know exactly what ChatGPT says about them, and which thread made it say it.