How to Measure Whether AI Assistants Recommend Your Product

How do you know if ChatGPT recommends your product when buyers ask?

You test it on purpose, the same way every time, and track the results over weeks. Build a fixed set of buyer prompts, run them across the main AI assistants on a schedule, and record whether your product is mentioned, recommended, described accurately, and cited. One answer means little. A trend across dozens of answers means a lot.

Founders often check this once. They ask ChatGPT "what is the best tool for X," see a competitor listed, and panic. Or they see their own name and relax. Both reactions skip a basic fact: AI answers vary from run to run, from assistant to assistant, and from one phrasing to the next.

So a single check is an anecdote, not a measurement. This framework turns AI recommendations into something you can track, compare, and improve, using a spreadsheet and a little discipline.

Why is measuring AI recommendations harder than measuring rankings?

Because there is no fixed results page. A search ranking is fairly stable for a given query and location. An AI answer is generated fresh each time and can change with wording, follow-up context, and the sources the assistant retrieves. You are measuring a probability, not a position.

That changes how you design the measurement. You need more prompts, more runs, and more patience. You also need to track several things at once, because being mentioned is not the same as being recommended, and being recommended is not the same as being described correctly.

The upside is that this kind of tracking reveals far more than a ranking report. You learn how assistants frame your category, which competitors they pair you with, and which sources they lean on. That is strategic information, not just a vanity number.

Which prompts should you track?

Track the questions real buyers ask at each stage, written the way they would type them. Cover four types: category questions, use-case questions, comparison questions, and problem questions. Aim for twenty to forty prompts to start. Keep the wording fixed so changes in results come from the assistants, not from your prompts.

Category prompts sound like "What are the best tools for revenue forecasting?" Use-case prompts add context: "What is the best forecasting tool for a 20-person SaaS sales team?" Comparison prompts name names: "Compare Tool A and Tool B for forecasting." Problem prompts skip the category: "How do I stop my sales forecast from being wrong every quarter?"

Problem prompts are the ones teams forget, and they often matter most. Many buyers do not know your category name yet. They describe their pain. If assistants recommend your product in answer to the pain, you are in the conversation before the buyer even starts comparing vendors.

What should you record for each answer?

Record five things: whether your product is mentioned, whether it is recommended or just listed, where it appears relative to competitors, whether the description is accurate, and which sources the assistant cites. Add the date, the assistant, and the exact prompt. That gives you a dataset you can sort and trend.

The difference between mentioned and recommended matters. "Tools in this space include A, B, and you" is a mention. "For a team your size, I would start with you" is a recommendation. Track both, because the second one drives action and the first one only builds awareness.

Accuracy deserves its own column. An assistant that recommends you but gets your pricing or main feature wrong can do harm. If you see repeated inaccuracies, the cited sources usually show where they come from.

Which metrics turn answers into a trend?

Use four simple rates: mention rate, recommendation rate, accuracy rate, and citation rate. Mention rate is the share of answers that name you. Recommendation rate is the share that actively suggest you. Accuracy rate is the share with a correct description. Citation rate is the share that link to your own site.

Calculate each rate per assistant and per prompt type, not just overall. You might be strong in ChatGPT and weak in Perplexity, or strong on comparison prompts and absent on problem prompts. Those splits tell you where to focus.

Some website platforms and specialist vendors now package this kind of tracking as AI visibility or AEO analytics. Check each vendor's documentation for exactly which assistants it covers and how it defines its metrics, because definitions differ. A homemade spreadsheet works fine to start.

How often should you run the prompts?

Run the full set at least monthly, and run each prompt more than once per session to account for variation. Weekly is better if you are actively changing content. Because answers vary, three runs of the same prompt give you a more honest picture than one. Average the results rather than picking the best run.

Consistency matters more than frequency. Use the same prompts, the same assistants, and the same settings each time. If you use logged-in accounts, note that personalization may affect results. Clean sessions give you a more neutral baseline.

Keep a change log next to your results. When you publish a comparison page, update your pricing page, or earn a mention on a major review site, note the date. Then you can see whether the AI answers moved afterward.

Why do assistants recommend competitors instead of you?

Usually because competitors have more clear, consistent, and widely cited information about the exact use case being asked. Assistants lean on sources they can retrieve and trust. If your site is vague about who you serve and third-party sites rarely mention you, the assistant has little to work with.

Look at the sources cited when a competitor gets recommended. Often it is a comparison article, a review platform, or the competitor's own use-case page. Each one is a gap you can close: a clearer use-case page on your site, a better review profile, or outreach to the publication.

I covered the most common reasons in detail in why AI assistants recommend competitors for your own category. The short version is that the fix is usually clarity and presence, not tricks.

How do you connect AI recommendations to pipeline?

Add AI assistants as an option in your "how did you hear about us" field, track referral traffic from assistants in your analytics, and ask about it on discovery calls. Recommendation rates show influence. Self-reported attribution and referral data show whether that influence turns into leads and deals.

Self-reported attribution is imperfect, but it captures something analytics misses. A buyer might ask ChatGPT for options, then search your brand name directly. Analytics records that as direct or branded search. Only the buyer can tell you the assistant started it. I wrote about handling that in how to attribute a deal that started in an AI assistant.

Referral traffic is the other half. Some assistants pass referral information when a user clicks a cited link. Set up a channel group or filter for those sources so you can see visits and conversions. I walked through one setup in how to track ChatGPT and Perplexity traffic in Webflow Analyze.

What does a good result look like?

A good result is steady improvement on the prompts that matter most to your pipeline, with accurate descriptions. Do not chase a perfect mention rate across every prompt. Focus on the use cases you sell best and the problems you solve best. Being recommended there beats being mentioned everywhere.

Set a realistic baseline first. Run the full prompt set for a month without changing anything. That shows your starting point and how much natural variation to expect. Then make changes and compare against the baseline.

Expect slow movement. Assistants update their sources and models on their own schedules. Content changes can take weeks to show up in answers. That is why the trend matters more than any single week.

What should you do next?

Write twenty buyer prompts across the four types, run each three times in ChatGPT, Perplexity, Gemini, and Claude, and record the five fields for every answer. That first baseline will show where you are recommended, where you are missing, and which sources shape the answers.

Then pick the weakest prompt type that matters to your pipeline and fix the biggest gap the citations reveal. Rerun the set next month and compare.

If you want help setting up AI visibility tracking, or turning the results into content and pages that change what assistants say, reach out. I build websites and the automations behind them for lead generation, and this is where SEO, AEO, and GTM meet. Let's chat.