A dentist in Scottsdale asked ChatGPT which practice in her area she should send her mother to. She wasn't testing anything. She was curious. The assistant named three practices, described each in a warm sentence or two, and hers wasn't one of them. One of the three was a practice that opened four years after hers, half a mile away, run by a former associate. She screenshotted it and forwarded it to her marketing person with one line: "How is this happening?"
That question is the entire market for AI visibility services right now. And the answer to it, delivered before you've pitched anything, is the best prospecting tool an agency has ever had.
The screenshot that closes the deal
Most agency prospecting leads with what the agency does. Capabilities, process, a case study or two, a price. The prospect nods along and thinks about it and doesn't call back. The problem with that approach is that it asks the prospect to imagine a problem they can't see and then trust you to fix it.
A competitive gap analysis inverts that completely. You don't describe a problem. You show one. You put the actual assistant response in front of them — the one where someone asked for the best orthodontist in their zip code and got a competitor's name in clean, confident prose — and you let the silence do the work. There's a particular expression people make when they see it for the first time. It's not curiosity. It's the sensation of ground shifting under something they thought was solid.
That's the whole play. A side-by-side look at how AI assistants answer the questions their future customers are already asking, with their name and their competitors' names placed exactly where the assistants put them. No projection. No maybe. A report on the present.
Why this isn't theoretical anymore
The people making decisions about which dentist, which attorney, which agent to call have quietly changed how they decide. They're not opening ten browser tabs. They're asking an assistant to narrow it to two or three and reading what comes back as if a knowledgeable friend said it. When an assistant names a practice, the practice inherits that trust. When it doesn't, the practice doesn't exist for that person. There's no second-page-of-Google equivalent where the diligent searcher eventually finds you. There is a list of three, and you are on it or you are not.
RankCommander's own AI Visibility Index — built on thousands of real assistant answers evaluated across categories and cities — makes the pattern hard to wave away. Within a single vertical in a single metro, the businesses assistants name and the ones they skip aren't sorted the way the local market would sort them. The busiest practice in town is routinely absent. A quieter competitor gets named again and again. You can see the shape of it in the Index data, and once an agency internalizes that the assistant's picture of a market and the market's own picture of itself have come apart, the prospecting motion writes itself.
Why the assistant names one and skips the other
Here's the part that turns a screenshot from a scare tactic into a diagnosis you can actually stand behind.
An AI assistant doesn't know anything about a dental practice the way a patient does. It has never sat in the chair. What it has is a vast amount of text about the world, and a strong bias toward information that shows up the same way in more than one independent place. Agreement across sources is what the model reads as trustworthy. If a practice describes itself one way on its own site, and nothing else in the model's view confirms it — or worse, other sources describe it differently, with a different name format, a different address, a different specialty emphasis — the model doesn't have a claim it can lean on. It has an assertion. And when a model is asked to recommend and finds only assertions, it hedges. A hedging model reaches for the name it's more confident about. That's the competitor.
Confidence, for a model, is really just consistency it can corroborate. The practice that gets named tends to be the one whose story lines up wherever the model looks.
Take a legal example to make it concrete. An attorney in Tampa wonders why an assistant keeps recommending a firm across town for exactly the kind of case she handles. Part of what's happening lives in a place like Martindale-Hubbell — a directory that predates the internet, that peer-rates lawyers, and that assistants have absorbed as a long-trusted, structured record of who practices what and where. When a firm's presence there squares cleanly with what its own site says, with how it appears in a bar listing, with how a legal directory describes its practice areas, the model has multiple independent voices telling the same story. It grows confident. When the attorney across town has that alignment and our Tampa lawyer has a stale profile that lists a practice area she abandoned years ago, the model reads two different stories about her and one coherent story about him. It recommends the coherent one. Not because he's the better lawyer. Because he's the more legible one.
That's the mechanism, and it's genuinely worth understanding, because it explains why the busiest practice in town can be invisible: being busy is a fact about the physical world, and the model can't see the physical world. It can only see whether the record agrees with itself.
Why a screenshot beats a spreadsheet
Agencies love a dashboard. Prospects don't feel a dashboard.
A visibility score compressed to a number gives the prospect something to negotiate with. "Is that good? What's average? How'd you calculate it?" You've handed them an argument. The raw assistant response gives them nothing to argue with, because it isn't your claim. It's the machine's answer, verbatim, and their competitor is in it. There's a reason the reaction to the number is a question and the reaction to the screenshot is a physical flinch.
The most effective version pairs the response with a competitor the prospect already has feelings about. Every practice has one. The rival who undercuts on price, the associate who left and opened up nearby, the newer firm with the aggressive marketing. When that specific name is the one the assistant recommends, the analysis stops being about technology and becomes about them. That's not a manipulation. It's just showing the truth at the resolution where it actually registers.
Where agencies get the gap wrong
The common failure is treating the analysis as a one-time reveal on one platform. An agency runs a single prompt through ChatGPT, gets a good scare-screenshot, and stops. Then the prospect, mid-call, opens Perplexity on their own phone and gets a completely different answer where they actually appear, and the whole pitch wobbles.
A prospect can look strong on Gemini and vanish on Claude. They can be named in a Google AI Overview and skipped by Grok and Copilot in the same afternoon. The assistants pull from different sources, weight them differently, and update on different clocks. A gap analysis that only checks one is a gap analysis you can be caught out on. The one that holds up under a skeptical prospect's own live testing is the one that already accounts for all seven, and that keeps checking, because today's screenshot is a photograph of something that moves.
This is exactly the difference between a scare tactic and a service. Anyone can generate a single alarming screenshot. What retains a client is showing the gap at signing, then showing it narrowing month over month, across every assistant, with the same rigor each time. (If you want the reporting cadence that makes that renewal conversation easy, we've written about reporting for agencies and the ROI case-study framework separately.)
Running the analysis without doing it by hand seven times
You could open seven assistants, type the same prompts into each, capture each response, note who's named, and repeat that for the prospect and three competitors. It works. It also takes an afternoon per prospect, captures one moment in time, and gives you nothing to show the client next month.
RankCommander's scan exists to collapse that into something you can run before a cold email and show on a call — the prospect's category, their city, their competitors, checked across ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews, Grok, and Copilot, with the actual response text captured, not just a verdict. And because it keeps running, the gap you open a relationship with becomes the progress you renew a relationship on. For the pitch itself, the walkthrough on pitching a skeptical client pairs naturally with the raw scan output.
What separates the agencies winning this category isn't a cleverer script. It's that they show up to the conversation already holding the picture the prospect hasn't dared to look at yet.
The window is closing while your prospect waits
Right now, in your prospect's city, an assistant is answering "who's the best [their category] near me" for a real person deciding who to call. That person will never see the practice that isn't named. They'll book the one that is. And the competitor being named today is compounding — every answer that includes them makes the record around them a little more coherent, which makes the next answer more likely to include them too.
Your prospect built their practice over a decade. It can lose ground in a market that most people haven't even realized exists yet, to a rival who did nothing but become more legible to a machine. That's the fear you can name for them, honestly, because it's already happening. And it's the fear you can stand between them and. Show up to /agencies and see how the scan turns that quiet erosion into the first thing you put on the table — before the associate who left, or the firm across town, becomes the name the assistant gives instead of your client's, for good.