A new row has appeared on a lot of marketing dashboards this year: some measure of how often AI assistants mention your brand when somebody asks about your category. Four tools in particular — Profound, Otterly.AI, Peec AI and Scrunch AI — have grown quickly enough to make it a category rather than a curiosity.

The problem they are pointed at is real. If buyers increasingly ask an assistant instead of typing into a search box, then a company can lose visibility without any of its rankings moving. The question is whether these tools measure that, or measure something adjacent to it.

What changed, in one paragraph

For twenty-odd years, finding out where you stood in search meant checking a ranking. Your page was at position four for a query, you could see the query volume, and you could see how many people clicked. Every SEO tool ever built rests on that chain: query, position, click.

When somebody asks an assistant instead, that chain breaks in three places at once. There is no stable ranking, because the answer is generated fresh. There is often no click, because the answer is the destination. And there is no query volume you can see, because the question was asked inside a private conversation. Whatever is happening in there, the old instruments do not reach it.

What the new tools actually do

The mechanism is simpler than the marketing suggests. These tools maintain a list of prompts relevant to your category — the kind of thing a buyer might ask — and run them against several assistants on a schedule. Then they parse the answers for mentions of your brand, your competitors, and the sources cited.

From that they derive the numbers you see on the dashboard: how often you appear, where you appear relative to competitors, which sources the assistants lean on when answering about your category, and whether the description of you is accurate.

Put plainly, they are running a survey. A large, automated, regularly repeated survey, but a survey — with all the sampling questions that implies.

Four limits worth understanding before you trust a number

The prompt set is the study design. Everything downstream depends on which questions the tool decided to ask. Two vendors monitoring the same brand will report different visibility because they chose different prompts. Ask to see the list. If you cannot, the number is not comparable to anything.

Answers vary between runs. Assistants are not deterministic. The same prompt on the same day can produce a mention once and not the next time. A single-digit change week to week is usually sampling noise, not a movement you caused.

There is no clickstream. These tools cannot see real user conversations, so they cannot tell you how many people actually asked, or what happened after. Visibility here means visibility in a simulation of demand, which is a genuinely useful proxy and is not the same thing as traffic.

Personalisation and memory are invisible. An assistant that knows a user's history may answer them quite differently from how it answers a clean monitoring account. The measurement is of the average stranger's experience.

What they are genuinely good for

Having said all that, the category earns its place, and the reason is not the visibility score.

The most valuable output is usually the source list: which pages the assistants keep citing when answering about your category. That is actionable in a way a ranking never was, because it tells you which third-party pages are shaping how you get described. Often it is a comparison article you have never heard of, or a forum thread from three years ago.

The second most valuable is accuracy monitoring. Assistants confidently state wrong pricing, discontinued features and outdated company facts. Finding out that several assistants describe a product you retired in 2024 is worth the subscription on its own, and it is a problem you can usually fix by correcting the source they are drawing from.

The visibility percentage is the number that gets put in the board deck and the one that deserves the least weight.

How to use one without fooling yourself

Start by writing your own prompt list before you look at anyone's default. Twenty questions you actually believe a buyer asks, in their words, beats two hundred generated ones.

Then set a baseline and leave it alone for a month. The temptation is to check weekly and react, which mostly means reacting to noise. Change one thing at a time — a page, a comparison table, a correction to a source — and give it long enough that a shift is distinguishable from variance.

And treat the accuracy findings as the urgent queue and the visibility trend as the slow one. A wrong price being repeated to buyers costs you something today. Being mentioned 4% less often this month probably does not.

Four of these tools are listed in our SEO monitoring directory, alongside the more traditional rank trackers. If you are new to the category, read the free tiers' methodology pages before the pricing pages — the methodology is where the differences actually are.