One AI answer is not a ranking

Somewhere this week, a buyer you never met asked ChatGPT who they should go with. It answered in a paragraph, named two or three companies, and the deal was decided before your funnel ever knew it existed. No click, no search page, no trace in your analytics.

That is the new front door. A growing share of buying questions never see ten blue links, and either you are in the paragraph or you are not. Most companies have never once checked which one they are. I find that both terrifying and, if you move now, a genuine opening.

The screenshot lie

The tempting move is to ask the question once, screenshot the answer, and call it a ranking. Half the market is already selling exactly that. But these engines are probabilistic. Ask the identical question ten times and you get different paragraphs, different names, different citations. One answer is one sample from a distribution, and a distribution is the thing you actually live in.

Anyone selling you a ranking made from a single answer is selling you a coin flip with a logo on it.

Measure it like an operator

Start with the questions, not the tools. Write down the twenty or thirty questions your buyers actually ask when money is on the line. Not your keywords. Their words: who is the best option for a team like mine, what should a small operation use, who do people switch to when they outgrow X. Then run every question repeatedly, across the engines, on a schedule.

Count three things. How often you appear in the answer. How often the engines cite you as a source, which is a different and harder win. And who gets named instead of you, question by question. That third list is the one that stings, and it is the one worth paying for, because it is a map of exactly where the next dollar of effort goes.

Run a baseline like that once and expect two surprises. The brands winning the AI answers are not always the brands winning the old search page, which means the leaderboard genuinely reset. And the engines disagree with each other, sometimes badly, so a win on one is not a win on the others. Both of those are measurable facts about your market that almost none of your competitors have looked at yet.

The loop, not the report

A visibility number by itself is a souvenir. The loop is what makes it worth running: measure, judge, find the gaps. Which questions do you deserve to win and do not? What evidence is missing that an engine would need before it could cite you?

Because closing a gap is rarely a trick. These engines cite evidence, so the work is making the evidence exist: the comparison page that answers the question head-on, the documentation that says plainly what you do and who it is for, the third-party mention that corroborates it. Those gaps become work briefs, a human reviews and executes them, and then the loop re-measures to see if the rate moved. The re-measure is the whole point. A report you cannot re-run tells you where you were, not whether anything you did worked.

Lines we do not cross

This space is young enough that the shortcuts are still legal and already ugly. So the rules are written down. No astroturfing. No fake reviews or planted mentions. No scraping walled consumer apps, and no bypassing the gates they put up. Where an engine has no compliant way to read it automatically, a human captures the answer by hand, and every datapoint carries a tag for how it was collected, so the lanes never blend silently. Refusals and no-answers get recorded too, because a refusal is data. And every measurement cycle runs under an explicit budget cap, so the meter never runs unattended.

The honest status: this runs in the field today with a first partner. Baselines are measured, the first gap briefs are cut, and the first re-measure cycles are underway. Early, and improving every cycle, which is the only claim I trust anyway.

This is the Visibility canvas in One Ghost, many canvases. Same spine as everything else here: rates, grades, human gates, and a loop that closes against what actually changed.

The leaderboard reset. Measure it as a rate, close the loop, and never buy a ranking made from one screenshot.