How to check what AI assistants say about your business

To find out what AI assistants say about your business, ask each engine the exact questions your buyers ask, while logged out, and write down which brands get named and which sources the answer cites. Do it in ChatGPT, Claude, Gemini, Perplexity, Grok, DeepSeek, and Mistral, and ask each question several times, because the answer changes from run to run. That is the honest, free method. Below is how to do it well, what to look for, and the point where doing it by hand stops being enough.

The manual way: ask the engines yourself

Start here before you pay for anything. The manual method is real and it teaches you how the answers behave, which no dashboard can do for you. The steps are simple.

• Write down your buyers' real questions. Not "is [your brand] good". Buyers do not search for you by name when they are still deciding. They ask category questions like "best tool for X", "what should I use to do Y", or "alternatives to [the incumbent]". Those are the questions where you either get named or you do not.
• Ask logged out, or in a fresh session. If you are signed in, the engine may lean on your history and your own site visits, which flatters the result. A clean session is closer to what a stranger sees. Turn off memory and personalization where you can.
• Record who gets named. For each question, note the brands the answer lists and the order. Your name, your competitors, anyone you did not expect.
• Record which sources are cited. On engines that show links, note the domains behind the answer. These are the pages doing the work. You will start to see the same handful appear again and again.
• Ask each question three to five times. One answer is a coin flip. Run it a few times and you begin to see which brands show up reliably and which were a one-off.

An hour of this on your top ten buyer questions will tell you more than a week of guessing. You will see, in plain terms, whether the machine knows you exist.

The limits of the manual check

The manual method is honest and free, and it is also where most people quietly give up, because it does not scale and it does not stay true. Here is where it breaks.

• It is a one-off. A check you did on Tuesday is stale by the next model update. The answers move, and a snapshot does not tell you which way.
• It is noisy. Because answers vary run to run, a few manual tries can leave you convinced of a pattern that is really just sampling noise. You saw a competitor twice and now you think they dominate. Maybe they do. Maybe you got unlucky. By hand, you cannot tell the difference.
• It misses engines. Checking two engines because they are the ones you have open is how you miss the engine where you are actually losing. Coverage by convenience is not coverage.
• It is not repeatable. Next quarter you will phrase the questions slightly differently, count a little differently, and remember the last result through rose-tinted glasses. You cannot compare this month to last month if the method drifts every time.

None of this makes the manual method worthless. It makes it a first look, not a measurement. Treat it as the way you learn what is happening, not the way you track it over time.

What to actually look for

Whether you check by hand or with a tool, four questions matter. Everything else is detail.

• Are you named at all? The first cut is binary. When a buyer asks the category question, does your brand appear in the answer or not. If you are absent from most runs, nothing else matters yet.
• Who wins instead? Note which competitors get named ahead of you and how consistently. A rival that shows up in nearly every run owns that question in the model's mind. That is your real benchmark, not the market-share chart.
• Which sources power the answer? Look at the domains cited across the runs. If the same review site, Reddit thread, or roundup keeps appearing, those pages are deciding your category. Being present in them is how brands get named.
• Does the engine cite your site? Separate question from being named. Perplexity and the search modes cite sources. If your own pages never appear as a citation, the engine is not reading you as an authority, even on questions about your own space. That is a fixable gap, and you cannot see it without looking at the source list.

Doing it across engines, and why one engine misleads

There is no single AI answer. ChatGPT, Claude, Gemini, Perplexity, Grok, DeepSeek, and Mistral are different models reading different sources, and they disagree. Perplexity leans hard on live web citations. Gemini pulls on Google's index. ChatGPT blends training memory with retrieval in its search mode. Claude answers a good deal from what it learned in training. Grok pulls on its own platform's discussion. The result is that you can be the confident top pick in one engine and completely missing from another, for the same buyer question.

This is why checking one engine is worse than checking none, because it hands you a false conclusion with confidence. If you only ever open Gemini and it likes you, you will believe you are winning while your buyers in Perplexity never see your name. Coverage across the engines your buyers actually use is the whole point. One engine is an anecdote. Seven is a picture.

The automated way: measure instead of spot-check

When you need the answer to be reliable and repeatable rather than a Tuesday-afternoon impression, you stop asking by hand and start measuring. The mechanics are the same as the manual method, run properly and at scale. A tool takes your buyers' questions, runs each one across every engine many times, samples enough to see past the noise, and reports a share of voice for each brand plus the sources cited behind the answers.

That is what Saymetry does. To show why single manual checks mislead, look at our own category scan. We ran dozens of buyer questions across 7 engines, which produced hundreds of AI answers, and those answers drew on roughly 20 recurring cited sources, with Reddit the single most-cited. No one who asked one engine once could have seen that. The pattern only appears when you run the questions enough times, across enough engines, to separate signal from a lucky run. That is the difference between checking and measuring.

In practice the Saymetry scan is a $249 one-time report. It runs your questions across the 7 engines, shows every question and the exact sources cited, includes a website audit, and hands back an action plan for the gaps. If you want it tracked over time rather than as a single snapshot, that is $149/mo, or $119/mo billed yearly. The scan is the honest version of the manual method: the same idea, run enough times and across enough engines to be trustworthy, with the source list pulled out for you.

Manual check versus automated measurement

Neither column is a trick. The manual check is the right first move and it costs nothing. The automated version earns its price when you need coverage, trustworthy numbers, and the sources handed to you rather than reconstructed by hand.

How to act on what you find

Whichever way you looked, the point is to change the result, not to admire the problem. What you find maps straight onto what to do.

• If you are not named, you are absent from the sources the engines read. Get into the review sites, Reddit threads, and roundups that keep showing up in the citations for your category. Presence there is how brands get picked up.
• If a competitor keeps winning, read the pages the engine is quoting them from. They usually have a page that answers the exact buyer question in plain language. Write your own, clearer version.
• If your own site is never cited, your pages are hard to read and quote. Clear headings, a real pricing page, plain descriptions, and schema markup make you easy to parse and easy to lift a sentence from.
• Then check again. This is not a one-time fix. The answers move, so you re-measure and watch the share of voice shift as your presence in the sources grows.

You cannot pay an engine to recommend you and you cannot prompt-inject your way in. This is earned presence in the places the machine reads. The first step is simply to look, honestly, across the engines. Do that by hand today. Measure it when the spot-checks stop being enough.

FAQ