You cannot improve what you have not measured.
Most business owners have no idea what AI assistants say about them. Some assume they are being recommended. Others assume they are invisible. Usually neither assumption is fully accurate.
The actual picture is almost always a mix: partially present on some platforms, absent on others, accurately described in some ways and incorrectly described in others. Running this audit gives you the facts — and the facts tell you where to spend your effort.
There is a secondary reason to check regularly. AI models update their knowledge as they crawl new content, and competitive landscapes shift. A business that was invisible six months ago might start appearing after a competitor's GEO work; a business that was named regularly might drop off after a competitor fills gaps the first one left open. Quarterly checks are enough for most businesses. Monthly is better if you are actively working on visibility.
Run these prompts on each platform.
Do this fresh each time — do not rely on a saved chat. AI assistants can give different answers across sessions, especially on platforms with live retrieval. Take screenshots.
Category-and-city prompt.
Ask: "Who are the best [your service type] in [your city]?" — for example, "Who are the best physiotherapy clinics in Scarborough?" Look for whether your business name appears. Note which competitors are named. If you do not appear, read the businesses that do: they have done something you have not yet.
Problem-framing prompt.
Ask as a buyer would: "I need [specific problem your service solves] in [city]. Who should I contact?" — for example, "I need a plumber for an emergency leak in Etobicoke. Who should I call?" This surfaces problem-based recommendations, which sometimes differ from brand-name searches.
Direct name prompt.
Ask: "What do you know about [your exact business name]?" Read the answer carefully. Is it accurate? Is the service described correctly? Is the city right? Does it describe services you no longer offer, or miss services you do? Inaccuracies here indicate what your own web presence is communicating to models.
Repeat on Perplexity and Gemini.
Paste the same three prompts into Perplexity and into Gemini. The answers will differ — each platform has different retrieval and weighting. Some businesses appear in one and not the others. The gaps per platform point to different fixes: Perplexity is heavier on web content freshness; Gemini leans on Google entity data and your GBP.
Log everything in one place.
Create a simple document with each platform, each prompt, the date, and a screenshot or copy of the answer. This becomes your baseline. When you fix gaps and re-run in 60-90 days, you want to compare against something specific — not a vague memory of what the model said.
What the gaps tell you to fix.
Each type of absence or inaccuracy maps to a specific fix. Here are the most common patterns.
Entity and consistency work first.
Total absence usually means the model has too little reliable data to cite you. Start with your entity definition — a clear, structured description on your homepage backed by Organization and LocalBusiness schema. Then check that your name, address and phone match exactly across your site, your Google Business Profile, and the major directories. See which structured data types matter most for generative AI.
Correct the source, not the model.
If the model describes your business incorrectly — wrong city, wrong services, outdated information — the fix is updating the sources it reads. Check your website copy, your GBP description, and your listings on Yelp, Facebook and any industry directories. Update them all to say the same correct thing. The model will catch up as it re-crawls. See how reviews and off-site signals shape what AI reports.
Close the gaps they have already filled.
Ask the model directly: "Why would you recommend [competitor name] over [your name]?" The answer is often specific. More reviews, a more established web presence, content that answers the question directly — these are signal differences, not quality differences. Closing them is a process, and it takes a few months, but it is not mysterious.
Related, and what to do next.
Questions, answered.
How do I check what ChatGPT says about my business?
Open ChatGPT and run two types of prompts: a category-and-city search ("best [service] in [city]") and a direct name search ("what do you know about [business name]"). Screenshot the answers. Then repeat the same prompts on Perplexity and Gemini. Each platform may say something different.
How often should I check what AI says about my business?
Quarterly is reasonable for most businesses. Monthly is better if you are actively working on AI visibility. Run the full audit after any major change — a rebrand, a new service, or a significant update to your website or Google Business Profile.
What if ChatGPT says something wrong about my business?
You cannot directly edit what a model says. But you can correct the sources it reads. Update your website, fix your Google Business Profile description, and make sure every directory listing reflects the accurate information. As the model re-crawls, accurate sources tend to displace inaccurate ones.
Why does ChatGPT say different things in different sessions?
AI assistants have some variability in their outputs — the same prompt can return somewhat different answers across sessions. This is normal. Run the prompts at least twice and compare. If you get wildly different answers, the model has low confidence about your business, which is itself a signal worth acting on.
Does checking what Gemini says differ from ChatGPT?
Yes. Gemini draws heavily on Google's entity data and your Google Business Profile. ChatGPT and Perplexity lean more on web content quality and freshness. A business might appear in Gemini answers because of a well-optimized GBP while remaining absent from Perplexity because its web content is thin.