Your client's retainer renewal is in three weeks, and the meeting is going to hinge on a question you may not be ready for: what did I actually get? Rank tracking used to answer that for you. A screenshot of position two, a green arrow, done. But more and more of your client's prospective patients and clients never see a ranked page anymore. They ask ChatGPT for a good family dentist in Tempe, or Perplexity for a divorce attorney in Sacramento, and they act on the name the model hands back. If your work moved that answer and you can't prove it, the renewal conversation gets harder than it needs to be. If a competitor's agency can prove it, you're the line item that gets cut.
This is the gap most agencies are walking into right now. The work is real and the results are real, but the documentation habits are still built for a search results page that fewer people look at. A case study for AI visibility has to prove a different kind of win, and it has to prove it to a client who can't see the AI answers themselves the way they used to eyeball their own Google ranking.
Why the old proof stopped working
The clean thing about rank tracking was that the client could verify it. They'd search their own name, see the position, and trust your report because it matched their reality. AI recommendations break that loop. Your client can't easily audit whether Gemini names them, because the answer shifts by phrasing, by location, by which of the seven assistants they happen to ask — ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews, Grok, and Copilot don't return identical answers to identical questions, and none of them return the same answer twice with perfect reliability.
So the burden shifts to you. You have to bring the evidence, and it has to be the kind of evidence that holds up when a skeptical partner at a law firm pokes it. That means controlled conditions. The same questions, asked the same way, against the same platforms, at two points in time. Change the query between the before and the after and you've handed the client a reason to doubt everything. Keep it fixed and the delta becomes something you can stand behind in a room.
Start with a before-state you can't fake later
The most valuable thing you will ever capture in an AI visibility engagement is the state of the world on day one, before you've done anything. You cannot go back and get it. Once you've improved a client's presence, the original absence is gone, and with it goes the most persuasive contrast you had.
Picture an orthodontist in Boise. Before you started, you asked all seven assistants some version of "best orthodontist in Boise for adult braces" and screenshotted every answer. The practice appeared in none of them. Two competitors appeared in most. That set of screenshots, dated and boring-looking, is the emotional core of the case study you'll present in ninety days, because it's the thing the client felt in their gut but had never seen laid out. The absence made concrete. Agencies skip this constantly, out at the starting line, and then spend the rest of the engagement unable to prove they changed anything.
The middle nobody documents
Between the before and the after is the part that turns a coincidence into a case. If visibility improves and you can't say what you did, a sharp client will ask whether it would have happened anyway. Sometimes it would have. A dated log of your interventions is what lets you answer honestly and confidently.
This is also where the real mechanism of AI visibility does its quiet work, and it's worth understanding rather than treating as a black box. AI models don't take a business's own website at its word. When a model decides who to name for "best pediatric dentist in Naperville," it's leaning on information that shows up consistently across independent sources it already trusts. Agreement is what reads as true to a language model. When the same practice name, the same specialty, the same address turn up the same way in several places that don't depend on each other, the model's confidence climbs and it's willing to say the name out loud. When those sources contradict each other — a different address here, a different practice name there, a specialty listed one way in one place and another way somewhere else — the model hedges. And a hedging model reaches for a safer, more consistently-described competitor instead.
Take Healthgrades as a worked example of why that matters for a medical practice. It's not that Healthgrades is a magic ranking lever. It's that it functions as a corroborating witness. When a model is assembling its picture of a physician, a Healthgrades profile that agrees with the practice's own site — same name spelling, same subspecialty, same location, same credentials — is a second independent voice confirming the first. That agreement is worth more to the model than anything the practice says about itself in isolation, because a business will always claim to be good; the value is in an outside source echoing the specifics. When a physician's Healthgrades entry has stale information, a maiden name, a former clinic address, that's not a neutral blank. It's an active contradiction the model has to reconcile, and the reconciliation often ends with the model quietly favoring a doctor whose story is told the same way everywhere. Documenting the day you resolved that kind of conflict, and dating it, is what lets you connect it to the visibility that followed.
The after-state, run against the same questions
Ninety days in, you run the exact same query set against the same assistants. Same phrasing. Same seven platforms. Now the Boise orthodontist appears in four of the seven answers where before they appeared in none, and in two of those they're named first. You put the before screenshots and the after screenshots in the same frame, and the client sees the shift with their own eyes instead of taking your word for a number.
Context helps the client understand whether that shift is a big deal, and this is where industry benchmark data earns its place in the presentation. RankCommander's AI Visibility Index evaluates thousands of real assistant answers across verticals — the published figures disclose their sample sizes, which matters, because a benchmark you can't source is just a number you made up with a straight face. Dropping a client's before-and-after against a vertical benchmark tells them not just that they moved, but that they moved from behind the pack to inside it. That framing changes how a renewal feels.
The bridge that actually closes the renewal
A visibility improvement is a proxy. Clients renew for revenue, not proxies, and the case study that wins is the one that connects the two. This is the part most agencies leave on the table, and it's the part that's usually sitting in the client's own records waiting to be read.
The connection doesn't require a fancy attribution stack. It requires you to ask the client what they already collect. A dental front desk almost always logs how new patients heard about them; you just have to get them adding an "AI assistant / ChatGPT" option and start counting. A law firm tracks new matter inquiries by source as a matter of habit. A real estate team knows exactly how many buyer consultations they booked last quarter. Call tracking on the number the assistants surface gives you a cleaner line still. The point is to sit the inquiry trend next to the visibility trend across the same window and let the client draw the line themselves. When an orthodontist can see that adult-braces consultations climbed in the same quarter their name started appearing in AI answers, the renewal stops being a debate about your fee.
It helps to have done the arithmetic on what one new client is worth in that vertical before you walk in. An incremental patient case at an orthodontic practice is worth thousands over its life; a single new legal matter can be worth far more; a closed real estate transaction dwarfs both. When the value of one client dwarfs a monthly retainer, you don't have to prove a flood. You have to prove the mechanism is real and pointed in the right direction, and let the client's own math finish the sentence. (If you're presenting to a client who's still skeptical the channel matters at all, that's a different conversation — we've written about pitching the skeptical client here.)
Keep it to two slides
Nobody in a renewal meeting wants a forty-page report. The version that works is short enough to absorb in the room: one view showing the before and after query results side by side against the benchmark, and one view showing the inquiry data across the same period. That's the whole story a client needs to say yes. The reporting cadence that feeds those two slides is worth building once and reusing across every account.
What separates the agencies keeping their retainers from the ones losing them isn't the size of the visibility gain. It's whether they captured the starting line before they crossed it, and whether they can hand a client a straight line from that starting line to the phone ringing. Everything upstream of that is just diligence you either did or didn't do.
Right now a client of yours is being described by AI assistants in a way you haven't measured, and a competitor's agency may already be walking into their renewal with the exact before-and-after you didn't capture. The prospect that would have called your client this month called the name the model gave instead. You built that account over years; you don't get to lose it in a meeting because the proof was sitting in the client's front-desk log and nobody connected it. Run a scan on your client accounts through RankCommander for agencies and get the baseline on record before the renewal calendar makes the decision for you.