Ask most marketing teams where they rank for their main keyword and they will tell you within a few seconds. Ask them whether ChatGPT names them when someone asks which tool to buy, and the room goes quiet.

That gap matters more every quarter. A generated answer names three or four companies and links three or four sources, and for a large share of questions the session ends there. Nobody scrolls to your result because there is nothing left to look up.

You do not need to buy anything to find out where you stand. Here is the manual version. It takes about an hour, it costs nothing, and it is the same method any tool automates.

First, two different questions

Before you collect anything, separate the two things you are looking for. They are not the same and they fail for different reasons.

A mention is the engine naming your brand in the answer. A citation is the engine linking your domain as a source for it.

You can have either without the other, and the combination tells you what to fix:

• Mentioned, not cited. The engine knows who you are, but it learned it from somebody else's page. Your reputation is doing the work and a review site is getting the traffic. This is the most common result and the most fixable.
• Cited, not mentioned. Your content is good enough to quote, but your brand is not registering as an answer to the question. Usually this means your pages explain the topic without ever positioning you as the option.
• Neither. The honest starting point for most sites.

Record them in separate columns. Collapsing them into one "are we visible" score throws away the only part of the result that tells you what to do next.

Step one: write ten questions you did not choose to flatter yourself

This is where the exercise is usually ruined.

The instinct is to type your company name and see what comes back. Do not. An engine asked about your company will describe your company. You learn that it has heard of you, which you knew, and nothing about whether you win.

Track the questions people ask before they know your name:

• The category, asked plainly. "What is the best rank tracking tool."
• The category with your qualifier. "Best SEO tool for a small agency." "Rank tracker for Polish sites."
• The comparison people actually make. "Ahrefs vs Semrush for a solo consultant."
• The problem, asked before the category exists in their head. "How do I tell if AI answers are taking my traffic."
• The objection. "Is an SEO tool worth it if I only have one site."

Ten is enough to see shape. Write them the way a person types them, in full sentences if that is how they would ask, because that is how they are asked.

One test for whether a question belongs on the list: if the answer changed tomorrow, would you do anything differently? If not, cut it.

Step two: ask every engine, and do it in one sitting

Five surfaces are worth checking, and they behave differently enough that you cannot use one as a proxy for the others:

Run all ten questions through each one on the same day. Use a logged-out session or a fresh profile. Personalization and chat history will quietly feed you a friendlier answer than a stranger gets, and a friendly answer is the one result you cannot learn anything from.

Step three: log more than yes and no

For each question and engine, write down four things:

1. Were you mentioned, yes or no.
2. Were you cited, and if so, which URL.
3. Which competitors were named, in order.
4. Which other domains were cited.

Column four is the one people skip, and it is where the work comes from. The sites cited instead of yours are a list of pages that answer the question better than anything you have published, ranked by an engine that has read all of them. That is a content brief you did not have to write.

Column three is the honest competitive picture. Not who you think your competitors are, but who the engine offers when a buyer asks.

Step four: treat one run as a sample, not a measurement

Generated answers are not deterministic. The same question, to the same engine, on the same day, can name different brands. This is the single most common way an AI visibility audit produces a confident wrong conclusion.

Two rules keep you honest:

• Run each question at least three times and record how often you appear, not whether you appeared.
• Never make a decision from one run. A brand that shows up in one of three attempts is in a genuinely different position from one that shows up in three of three, and a single check cannot tell them apart.

If you only have time for one pass, that is fine. Just write "sample of one" at the top of the sheet so that nobody quotes the number in a board deck six weeks later.

Step five: act on the pattern, not the score

By now you have a grid. Read it for patterns rather than a total.

If you are mentioned but rarely cited, your problem is source-worthiness, not awareness. Find which of your pages do get cited anywhere, look at what they have in common, and make the rest of your library look like them. Usually it is specificity: a page with a real number, a real comparison, or a real method gets quoted, and a page of positioning does not.

If you are cited but not mentioned, your problem is positioning. Your pages answer the question and never say who you are or what you would do about it. Adding one honest sentence about what you offer, in the paragraph that answers the question, changes this more often than you would expect.

If you are absent entirely, start from column four. Publish against the questions where somebody else is cited and you are not, one at a time, and re-run the check a month later.

What breaks when you do this by hand

The method works. Keeping it up is the part that fails.

Ten questions across five engines, three runs each, is 150 prompts in a sitting. Done once, it is an afternoon and a genuinely useful picture. Done monthly, so you can see whether anything you shipped moved anything, it is a day you will not keep giving it, and the sheet stops being updated in month three. Meanwhile the answers keep changing, because the engines keep changing.

That is the whole reason automated AI visibility tracking exists, Ranksmile included. We run the same surfaces on a schedule, four of them on Growth and all five on the larger plans, keep the history so the trend is readable rather than a sequence of one-off impressions, and show which of your pages get cited rather than only whether any of them did. Our documentation on choosing prompts is the longer version of step one, and it is worth reading even if you never track anything with us.

But run the manual version first. An hour with a spreadsheet will tell you whether this is a problem you have, and that is a better reason to buy a tool than anybody's landing page.

Three questions people ask after running this

How often should I repeat it? Monthly is enough for most sites. Sooner if you publish something significant against one of the tracked questions, or if organic traffic drops without a ranking change, which is the signature of an answer surface absorbing the click.

Does ranking first still matter if the answer sits above my result? Yes, and less than it used to. Classic positions still produce clicks, and generated answers still draw on pages that rank. The two measurements disagree often enough that you need both. Our rank tracking and AI visibility FAQ covers where they diverge.

Should I check Claude too? Check it manually if your buyers use it. Coverage varies between tools, and Ranksmile does not track Claude yet. The five surfaces above are where the commercial questions are being asked today.