The shelf is built. Here is what is on it.
Insights is where we publish what we learn running AI visibility work for medical, legal, wealth, and mental health practices: what moved a number, what did not, and the methods behind both. Three pieces are up. Every figure in them carries the organization that published it and a link, and where a survey came from a company with something to sell, we say so on the page.
Three pieces, published.
Read them in any order. The first one is the argument the other two sit inside.
People bring the biggest questions of their lives to a machine they say they cannot verify. The engine cannot settle that on its own, so it hands over a name. What it takes to be that name, and what it does not.
The free audit, the alarming score, the jargon you cannot check, the monthly retainer. One question that separates a vendor selling the work from a vendor selling the invoice, applied to us as well.
Most of what improves how AI sees your practice needs doing once. Sort the list by how often each item actually needs doing and the question of what to pay for monthly answers itself.
Four rules, decided before the first piece.
This field produces an enormous amount of writing that exists to fill a content calendar. We would rather publish less and be quotable, so the rules are on the page before the articles are.
Every piece comes from work we actually did: a client engagement with permission and the identifying details removed, a change we made to our own site, or a probe we ran across engines. No roundups of other people's takes.
A probe that moved nothing is more useful than one that worked, because everyone publishes the second kind. When a piece reports what we tried, it carries the attempts that went flat alongside the ones that moved. The flat ones are the whole reason to trust the rest.
If a piece contains a rate, it carries how many runs stand behind it, from which market, on which engines, and over what window. A number without that is decoration, and we would have to argue with it ourselves.
A physician or an attorney reading this at the end of a long day should be able to work out what it means for their practice. No jargon that a definition would not fix, and no advice we would not give a client we liked.
How the writing is organized.
Pieces appear under their category, newest first, with the date on them. Two shelves are still empty, and they stay labeled that way until something real goes on them.
How AI visibility gets measured properly: run counts, frozen question sets, mentioned versus cited, confidence and its limits, and the places our own method is weaker than we would like. The technical spine of everything else here.
What we see inside specific verticals: how a dermatology patient and a bankruptcy client ask AI differently, which surfaces the engines lean on per field, and what actually moved for practices in each. Identifying details removed, always.
What the engines appear to read and reward, based on what we can observe from the outside, with the uncertainty stated plainly. Nobody outside a model lab knows the internals, and pieces here will say so rather than imply otherwise.
Where board and bar advertising rules meet AI answers: testimonials, superlatives, outcome claims, and the awkward fact that the most compliant version of your marketing is usually the version engines trust most.
The commercial layer around all of this: what vendors sell, how the pricing is built, which parts of the work genuinely recur and which only recur on the invoice. We are a vendor too, so these pieces apply the same test to us in writing.
On the desk, not yet on the shelf.
These are in progress rather than published, so there is nothing to click yet. They are listed because saying what is coming is cheap, and it keeps us honest about what is actually finished.
The arithmetic behind a sound polling cadence, worked through step by step: how wide the confidence band is at five runs, at twenty, at a hundred, and where the extra runs stop being worth the money for a local practice.
Why a single negative review from years ago keeps surfacing when someone asks an engine about a named practice, what that looks like across engines, and the correction paths that are actually available to an owner.
Our working list per vertical, how we assembled it, and how often it changes. The useful version of this is short, which is the opposite of what the fifty-directory submission services would like you to believe.
The same question asked from three cities, with the answers side by side. The argument for geo-varied probing, made with the results rather than with adjectives.
No dates promised. A piece publishes when the underlying work is finished and checkable, not when a calendar says it should. If you want to be told when the next ones land, write to human@aeoptim.com and we will put you on a short list that does nothing else.
Three more things worth your time.
The writing is behind the work, not ahead of it, and three pieces is not a library. These are the pages that show you how we think while it catches up.
A live engine asked the question your clients ask, with the answer returned word for word. No card, no call. It is the fastest way to see whether any of this matters for your practice.
Run it →Our measurement method, published in full: how we probe, how many runs stand behind a rate, what counts as mentioned versus cited, and where the method has limits.
Read the method →The questions your clients ask an engine, answered on your own pages in the format the engines quote, with paste-ready schema for each. The fastest paid step from knowing to fixed.
See the kit →The most useful thing on this site is still free.
Nothing published here will tell you as much about your own practice as one live answer will.
Run your free check →Want a heads up when the next pieces land? human@aeoptim.com