How to Track Competitors Inside AI Assistants

When a prospect asks ChatGPT, Perplexity, or Google's AI Overviews for "the best commercial HVAC company in Phoenix," the model returns a short list of names. If yours is not on that list and three competitors are, you have a problem you cannot see in your analytics. Nobody clicks through. There is no referral string. The recommendation happens inside a conversation you never observe.

So the first job is observation. Before you can change how you appear inside AI assistants, you need a reliable, repeatable way to watch what they already say — about you and about the companies you compete with. This is not a dashboard you buy once. It is a process you run on a cadence, the same way you would check your Google Business Profile or your ranking positions.

Why you cannot skip the manual check

There are paid tools that claim to track "brand visibility in AI." Some are useful. But AI assistants are non-deterministic — ask the same question twice and you can get two different answers, different company names, and a different order. A tool that samples once a week gives you a snapshot, not a pattern. You want the pattern.

The other issue is that the answer depends heavily on how the question is phrased. "Best dentist near me" and "affordable family dentist in Tucson that takes new patients" pull different businesses into the response. If you only track one phrasing, you are watching one narrow slice of demand. The practitioners who get real signal build a small set of prompts that mirror how their actual customers ask, then run them repeatedly.

Build your prompt set first

Start by writing down the questions a real buyer would type. Not keywords — full, natural questions. For a plumbing company, that looks like:

• "Who are the best plumbers in [city] for emergency repairs?"
• "I need a plumber for a water heater replacement in [city]. Who should I call?"
• "What's a reliable plumbing company near [neighborhood]?"
• "Compare the top-rated plumbers in [city]."
• "Which plumbers in [city] offer financing?"

Aim for 8 to 12 prompts. Cover your core service, your highest-margin service, a comparison phrasing, and at least one that includes a differentiator you actually have (financing, 24/7, commercial, a specialty). The comparison and differentiator prompts matter most, because that is where AI assistants name specific companies side by side.

Pick your assistants

Run each prompt through the assistants your customers actually use. In practice that means ChatGPT, Google's AI Overviews and AI Mode, Perplexity, and Gemini. If you serve a B2B audience, add Microsoft Copilot, since it rides on a lot of corporate desktops. You do not need all of them. Three is enough to see a pattern.

How do I actually monitor what AI assistants say about competitors?

Run your fixed prompt set through each assistant on a set schedule and log every business name that appears, so you can see which competitors show up most often and in what context. The mechanics matter, so here is a step sequence you can follow:

1. Use a clean session. Log out or open a private/incognito window and turn off personalization where you can. You want the generic answer a stranger would get, not one shaped by your own history.
2. Run each prompt three times. Because answers vary, a single run is noise. Three runs per prompt per assistant shows you which names are stable and which are occasional.
3. Record the output in a spreadsheet. One row per run. Columns: date, assistant, prompt, every business named, the order they appeared, and any reason the model gave ("known for fast response times," "highly rated on Google").
4. Capture the citations. Perplexity and AI Overviews show source links. Note which sites the model pulled from — directories, review platforms, the competitor's own pages, local news. Those are the sources feeding the recommendation.
5. Tally frequency. At the end, count how many of your runs named each competitor. A company that appears in 11 of 12 runs is entrenched. One that appears in 2 of 12 is incidental.

This takes about 60 to 90 minutes the first time and less once you have the template. Monthly is a reasonable cadence for most local businesses; quarterly is fine for slower B2B categories.

Reading the results without fooling yourself

The raw log is not the insight. The insight comes from the patterns. Here is a hypothetical tally for a commercial electrician running 12 prompts, three runs each, across three assistants (108 total runs):

A few things fall out of this immediately. Competitor A is being pulled from its own pages, which means its site is structured in a way the models can read and quote — clear service descriptions, probably good FAQ content, specific locations. Competitor B rides on directories and reviews. You appear, but low and infrequently, and only your Google profile is feeding the answer. Your own website is contributing almost nothing to how AI describes you.

That diagnosis points to a specific fix: your site is not giving these models anything quotable. The gap is not reputation. It is content structure.

What should I do when competitors appear and I don't?

Figure out which sources the assistants are citing for those competitors, then make sure your own site and profiles carry the same information in a format a model can extract. The work usually breaks into three layers:

1. Fix what the model can read on your own site

AI assistants quote content that directly answers the question. If a prompt asks "which electricians offer emergency service," the model needs a page that plainly says you offer 24/7 emergency service, in what areas, with what response window. Vague hero copy like "excellence in electrical solutions" gives it nothing. Write pages that answer the actual questions in plain sentences, with the specific facts — service area, specialties, certifications, pricing structure, who you serve.

2. Strengthen the third-party sources

If competitors are surfacing through directories and review platforms, your absence there matters. Make sure your Google Business Profile is complete and your category is correct, and that you are present and accurate on the directories specific to your trade. The models blend these sources; you want to be in the same pool.

3. Close the differentiator gap

Look at the prompts where you were named least. If you vanish on "offers financing" but you do offer financing, that fact is not published anywhere the model can find it. Each missing differentiator is a page or section waiting to be written.

Re-run the same prompt set the following month and watch the tally move. Because you are measuring the same questions the same way, the comparison is honest. This is slower than a paid dashboard but far more trustworthy, and it tells you exactly which content to build. If you would rather not run it in-house, this kind of question-mapping and content work is what managed programs like ClearPath Content are built to handle.

The practical takeaway

You cannot manage what you never see. Spend ninety minutes building a fixed prompt set, run it across three assistants with three runs each, and log every business named plus the sources behind each answer. Do it again in thirty days. The gap between your tally and your competitors' tally is your real to-do list — and unlike most marketing metrics, it maps directly to the exact pages and profiles you need to fix.