By improving AI unit economics, even as we increase the accuracy and value we deliver to clinicians. On a GAAP basis, stock-based compensation, or SBC, was $41 million in the quarter, or 23% of revenue. This is consistent with the low 20% SBC guidance provided last quarter. As a reminder, this year's SBC increase is primarily related to the FY26 grant made to our AI-focused R&D team. Excluding the grant, SBC would have been approximately 19% of revenue in Q2. We expect the impact of the AI grant to ease in FY28. Our gap effective tax rate was approximately 31% in the second quarter compared to 15% in the prior year, driven by the tax treatment of equity compensation. Our non-gap effective tax rate remained at 21%. In Q2, gap diluted EPS was $0.18 per share and non-gap diluted EPS was $0.36 per share. Fully diluted shares declined by 13 million, down 6% year-over-year to 188 million shares outstanding. Our balance sheet and cash flow generation remained strong. We ended Q2 27 with $713 million in cash, cash equivalents, and marketable securities, and we remain debt-free. We generated $91 million of free cash flow, outpacing adjusted EBITDA and up from $40 million in Q1. As expected from last quarter, collections activity normalized. During the quarter, we repurchased $54 million of shares. We believe share repurchases remain an attractive, opportunistic use of capital. As of September 30th, we had approximately $347 million remaining in our existing repurchase program. Let's turn to our outlook. For revenue in Q3, we expect a range of $195 to $196 million, representing 6% year-over-year growth at the midpoint. For FY27, we've raised our revenue guidance range to $678 to $683 million, also representing 6% growth at the midpoint. We expect slight acceleration in second-half revenue growth versus first-half on a year-over-year basis. For FY27, we are increasing the guidance midpoint by $4.5 million. This reflects the flow-through of our Q2 outperformance, plus a modest incremental raise driven by growing visibility into the commercial AI pipeline. AI Search is now an eight-figure bookings product, and we started recognizing revenue from the product when the first program launched in July. As a reminder, Q3 is our upfront season. This is when most pharma clients commit the majority of their marketing budgets for next year's programs and what are called upfront contracts. It is the largest bookings quarter of the year, and most contracts are signed by the end of December. We then typically recognize those commitments as revenue over the following 12 months, depending on the timing of program launches. One key difference this year is the availability of our new AI search product. To date, we've seen early broad-based adoption with 16 of the top 20 pharma companies having purchased programs, and three more are in our pipeline. We will provide an update about the upfront and AI search progress in our February call. In Q4, we expect year-over-year revenue growth to be similar to the Q3 guided growth rate. As with prior years, we have lower visibility into Q4 until we finish the upfront. In addition, new products like AI Search have a steeper ramp in the first calendar quarter of a new year as we help new customers through the medical legal review process. As a result, we expect a gradual ramp in AI Search revenue throughout calendar 27. For adjusted EBITDA, in Q3, we expect a range of $102.5 to $103.5 million, representing a 53% adjusted EBITDA margin at the midpoint. For FY27, we now expect a revised range of $317 to $325 million, representing a 47% adjusted EBITDA margin at the midpoint. This outlook incorporates the continued investment required to support the stronger than expected AI engagement we've seen this year. In the second half of FY27, we'll continue to invest in growing clinician AI usage and expanding our moats. Key initiatives include increased safety and accuracy through programs like DoseCheck, greater clinical content through partners like UpToDate, and growing distribution leadership with U.S. health systems. In closing, our long-term approach remains unchanged. We will continue to deliver one platform for AI clinical decision support, documentation, and patient communication. Improving efficiency in our AI investments allows us to continue playing offense in the midst of this generational technology shift and help our clinicians be more productive so they can provide better care for their patients. Lastly, as a reminder, you can find our modeling considerations appendix in our prepared remarks where you can see more detailed financial commentary. And with that, I will turn it over to the operator for questions.
Operator
We will now begin the question and answer session. Please limit yourself to one question and one follow-up. If you would like to ask a question, please press star one to raise your hand. To withdraw your question, press star one again. We ask that you pick up your handset when asking a question to allow for optimum sound quality. If you are muted locally, please remember to unmute your device. Please stand by while we compile the Q&A roster. Your first question comes from the line of Brian Peterson with Raymond James. Brian, please go ahead.
Hi, gentlemen. Thanks for taking the question. So, Jeff, I was encouraged by your comments on AI Search, the eight-figure bookings, you know, 16 of the top 20 with more on the way. Maybe you could talk about how that kind of built over the course of this year versus your expectations, and is there anything qualitative that you could share about what that could do for your positioning into 2027?
Yeah, thanks, Brian. This is Jeff. I'll say, qualitatively, search is... is exciting to our clients and to us. We've had more what they call top-to-top meetings with pharma companies, chief marketing officers, CEOs in the top 20, and they're going really well. It's an exciting time. We're getting real insight into the real problems that real patients and real doctors have without having to fly around and do lots of focus groups and rep rides and And things that, frankly, aren't as efficient. So we're excited about that. We think we're entering a whole new market here, which has historically been paid search. And I'll give a hat tip to Bank of America. They did a survey of 35 pharma marketers. Absolutely independent survey. Nothing to do with it. And they asked, if you had more incremental budget, more to spend on marketing, who would you spend more with? And we hit an all-time high with 81% of the respondents saying they'd spend more with us. And I think that's because of the traction we're seeing with our 190 health system partners and AI search growth. But interestingly, the group that fell the most there was paid search or Google, which fell to only 6% of the respondents saying that they would spend incrementally there. So I think that's proof that with AI search, we are entering search budgets, which is about $14 billion a year in the United States, according to eMarketer, just for pharma. So $14 billion a year spent on paid search. That is a much larger market than we have been in historically, and we're excited to enter it. So as I think of the build over the course of this year, I think we'll continue to see a big upfront there. But starting early next year or late this year with a couple pilot clients, we'll also offer a biddable or auction-based purchasing there, Which does make things a lot easier for our clients and for us. And again, brings us into an exciting whole new world that is more like paid search and less like endemic marketing.
I appreciate the caller. And maybe just a follow-up. I wanted to hit on some of the AI compute things that you mentioned in terms of efficiency measures, in terms of multimodal usage or some of the performance gains from your harness. Could you guys unpack that a bit? I know that's been a key topic of conversation with investors. Thanks, guys.
Sure. Matt can speak to this a bit as well. I just spoke at the Fireworks conference a couple days ago. There's a lot going on in AI inference right now, and most of it's good for companies like us in that there are just more options, and the platforms are getting more flexible to be able to handle, again, the security and medical needs that we have. We have a great team here. I will say I continue to be very pleased with the pathway acquisition and the new hires. We've kept almost all of our AI team, so no turnover. So from our end, we're really excited about, again, all the new software tools that are allowing us to do our jobs better at helping doctors make.
Yeah, and I can just pile on a little bit. I think, honestly, the most important AI investment metric we have is just accelerated engagement, right? It speaks to us kind of rapidly engaging with clinicians through Ask, through Scribe, through what is now the most accurate. The most accurate clinical AI product in the market. So I think that's a good place for us to be. The accelerated engagement is the proof point. I think we're also getting better at learning how to develop with the infrastructure. And so all the things that we talked about in the prepared remarks. Better performance on our proprietary harness, more multi-model deployments, buying more capacity and compute up front. All of those things help. I think maybe my favorite financial metric from the quarter was our cost of revenue grew up less than half of our overall prompt volume in Q2. And in Q1, those two things were pretty similar, right? Our cost of revenue grew at a much higher rate at a similar rate of prompt volume. So I think we've shown that we can be efficient as we deliver great results. Look forward to keeping that up.
Operator
Your next question comes from the line of Michael Czerny with Lee Rink Partners. Michael, please go ahead.
Great. Thanks so much. Maybe just one, Jeff, here, if I can. I would love to get a little bit more detail on the up-to-date partnership. First time we've heard from you since that was announced, Walters Kluwer, clearly a well-regarded partner in the market, but how did you settle on that being the right partner at this point in time as you build out your AI capabilities? And as you think about what your other opportunities were, what else was the range of outcomes that landed here and landed at this being the right and the best partner at this point in time? Thank you.
Thanks, Michael. Yes, we're excited, super excited to work with UpToDate, and, you know, they're picky. They're the number one player in the market, the most used, most trusted. I think Evercore ran a survey on this a little earlier this year. I mean, they're far and away, head and shoulders above anyone else. They're, I like to joke, IQIA like market share when it comes to being the clinical decision support point of care tool for doctors, most used, most trusted by far. And like I said, they're choosy. So they've chosen to work with Epic, Microsoft, and us, and frankly, not many others. And I think what we're bringing Greg and his team is a lot of AI savvy when it comes to being able to pull in broader, faster-moving evidence segments. But what they bring us, and uniquely in the market, is keep in mind, everyone has access to all of the journals that are paid for largely by government grants and are all in the National Library of Medicine. And everyone has access to all that, all the frontier models. But very, very few. I mean, none of the frontier models have access to up-to-date. So we're excited about that. We're just getting out in market with it now. It's been just a couple weeks since our announcement, but the client reaction to it has been very positive. And I would just say the partnership and the people also has been very synergistic. Up to date is a mission driven company, similar values to us and our team. And again, it seems to be just a strong synergy of. You know, our AI evidence tools and their unique expert opinion and physician expertise in that medical layer of knowledge that, frankly, isn't in a randomly controlled trial or published area. A final point note, we have significant customer overlap. They have 2,500 health systems as clients. We have, you know, 190. So we see a lot of overlap between our client base, and that also creates additional synergies for us. So anyway, very excited about UpToDate, and you'll see more from us with them over the coming quarters. We're just getting started.
Operator
Your next question comes from the line of Craig Hettenbach with Morgan Stanley. Craig, please go ahead.
Thank you. Jeff, on the point of around half of prescribers now using the workflow tool, could you just touch on what's driving the acceleration there, and then what type of signal may that provide for growth in the coming quarters?
Thanks, Frank. Yeah, that's a real milestone for us as a company to have over half of all U.S. prescribers, and there are about 1.8 million U.S. prescribers, so about over 900,000 U.S. prescribers, and that's MD, DO, NP, or PAs, so folks who can legally write prescriptions in the U.S. To have over half of them in the last quarter use our workload tools is a big milestone. We've never been that. Big before. And again, as we said, that's up 34% year on year. So it's a good question. What's the biggest drivers? The two biggest areas there are telehealth and scheduling. So telehealth. It continues to grow in this country. It's very popular, especially in disease categories like diabetes. And, you know, let's face it, doctors are finding it's more efficient. And patients, of course, appreciate the no commute of being able to see your doctor virtually during the day. We're seeing a lot of doctors now have a half day or full day a week that is just dedicated to work-from-home telehealth, and we think that's better for doctors and their overall work-life balance and sanity. It's better for patients. We're really proud to help power it. And again, we've never done as many telehealth calls as we did this last quarter. We're, again, seeing continued growth there kind of quietly. During COVID, we talked about telehealth a lot and that it sort of receded, but we're well above the call volumes that we were doing during COVID now. And so we're just seeing continued growth here in 23, 24, 25, 26. Scheduling also is a growth area for us. Doctors, of course, need to have their, you know, schedule all their shifts. We make it easy to call each other, secure message to them. So that's a continued growth area for us. Those are the two I'd point to. The third, of course, is AI, which we've already been talking about a lot. But the reality is that's a whole new category of usage and growth for us. And I am particularly excited about the thousands of doctors we have now using the full patient context. So doctors don't want to go type the full patient history into a search bar. They want a suite, not a search bar. And that's what we offer them. We offer them a suite, not a search bar. That suite allows them to do the tell-out visit, to have notes taken on that visit, what we call scribe, and then to ask questions about that visit with the patient context. Again, that's the most efficient way to get the best care in this country, and we're seeing that with that full patient context. There are already cases of, you know, tough diagnoses that probably would have taken a lot longer, right? Specialist visits and lots of follow-ups. And AI is really very good at, you know, finding that rare fast sonoma on page 89 and triangulating that into, you know, a patient that has a 1 in 100,000 disease. So It's just so fulfilling for me that my eyes get wide, and frankly, it's right to the core of our mission to help doctors, again, provide better care. That patient context in that suite and that AI growth, which is over half of our workflow users, those are the three big growth areas, telehealth, scheduling, and AI search.
And Craig, I think you asked. A little bit as well about how that leads into a monetization opportunity. You know, I think I draw upon some of my experience in consumer web, in particular at LinkedIn, and like, I think in those models and the same business model that we have. Your engagement growth leads your monetization opportunity. I like that we're building our engagement growth the right way, based on workflow, based on the value that we're adding for our clinician users. And I'd be remiss if I didn't point out that. A year ago, our workflow active user growth was in the single digits. Today, it's 34%. So, it's accelerated dramatically. I think that just gives us a pretty incredible opportunity over the next couple of years to grow our business as well. So, we're really excited about that.
A couple contexts, thanks. And then just as a follow-up, Matt, nice to see the upside to margins in the quarter and the guide. You pointed to some things in terms of efficiencies. As you kind of play this out, like, is there anything else you would point to in terms of perhaps building confidence of operating leverage in the business?
Yeah, I think we talked a little bit about this in the last border earnings call, where This is a matter of time sequencing, right? In order to drive that network engagement, the value for our users, that happens before you drive the monetization that follows it. And there's nothing about any of the AI investments and the early monetization we're seeing with AI search that would suggest that our margin profile would be dramatically different. Potentially, it could even be better over time, although it's not necessarily what we're over-optimizing for. But I think more than anything else, we're really encouraged about the incremental margins that we'll see in the business as we really start to scale AI search in FY28 and beyond. Nothing gives us pause there, and revenue is going to be the ultimate thing that gives us the most leverage. I think some of the efficiency on the cost side just gives us more encouragement in that direction from this past quarter.
Operator
Your next question comes from the line of Ryan Daniels with William Blair. Ryan, please go ahead.
Yeah, thanks for all the color, and thanks for taking the question. Jeff, one for you, and you talked a little bit about this, but I want to dive a little bit deeper into it. With 16 of the top 20 pharma clients buying AI search solutions, are you able to tell if that's at all cannibalizing any of your other sales with them, or is it really all proving to be incremental at this point?
Yeah, it's a good question, Ryan. I'd say it's mostly incremental. Of course, it's impossible to know that exactly, but I think we will know that exactly when we offer it as a biddable solution where they're purchasing it, usually out of a separate team that is more of a search team, because it's more of a keyword-driven approach to marketing, and that typically is a separate marketing team within our clients. So, it's- It's mostly incremental. I can't promise you it's 100% incremental, but I will say, going back to that survey of pharma marketers that was done, we're proud to be at an all-time high, 81% incremental spend, and I think that's mostly that paid search bucket that we are now entering, $14 billion a year spent on Google by pharma per year. And it is interesting in that same survey that Google is at an all-time low at 6%. So, again, mostly incremental. And the excitement to be able to really be in those highly relevant contextual places is something that we're excited about.
Ryan, the only other thing I'll add, there was some of the language in the prepared remarks that you'll see around Q2. We saw positive growth in Q2 across all of our major product categories. And that's really encouraging. And while I think the doc, I'm sorry, why the AI search pipeline is continuing to grow, you know, it's still It's still early days and pretty nascent. Most of what we're kind of seeing in the upfront cycle so far, again, early in the cycle, is demand for our core products, whether it's in the news feed or it's video around Tyler and telehealth. I think we're in a good place with a solid foundation to build AI search on top of. So I think that's encouraging for us, too, as we go throughout the year.
Yeah, super helpful caller. And then, Jeff, as my follow-up, it seems like we're seeing a lot tighter governance among health systems, especially bigger ones or academics, about how doctors use AI. So, you know, early on, maybe just download a free tool, but now they're getting a lot more concerned about liability and patient care. And I'm curious how that plays into your advantage, given the outcomes you've been able to present, like you talked about earlier, plus the massive channel distribution you have with your dialer asset into those enterprise accounts.
Thanks. Yeah, thanks, Ryan. Yeah, I think on last quarter's call, you know, we had made some comments about, you know, AI accountability. That's the next age or next phase for our health system clients, which we did grow from 165 to 190. So 25 new clients in the last quarter. And we are going to keep growing there. We have over 300 enterprise clients overall on our telehealth and other products. Uh, and, and we have, you know, privacy agreements in place with all of them, but we don't have the AI review with all of them. Um, but that 300 enterprise clients that we have are nearly half of all US physicians. So we cover all the largest, uh, health systems. And again, that's where we really see this, this AI growth and TAM. Um, one specific case that came out this last week was, uh, Johns Hopkins. Uh, they had 3000 patients that they had to send an email out. out to, which isn't fun for anyone to do, telling them that an employee had uploaded some of their data to a unapproved AI. So when we see things like that happen, and we've seen a bunch of other health systems, you know, put in IP level blocks. I think we're just going to see more of that happen, because the reality is people should be concerned about the security of their patient data being put into AI that may leak out in various places. So I think we're moving beyond the Wild West phase of AI and health systems. We're proud to announce NYP this quarter as a new AI logo. We're proud to announce UCSD, both top 20 hospitals, as a new logo. But I do think you're right. I think we're entering this new phase here where hospitals, I think, will rightfully start to say, okay, what kind of data might we be leaking? And what sites, frankly, do we need to block? They've done that for years and years. It's not a new thing for them. I think they've just been... Catching up on how much that AI leakage might be happening.
Operator
Super insightful. Thank you. Your next question comes from the line of Elizabeth Anderson with Evercore ISI. Elizabeth, please go ahead.
Hi, guys. Good afternoon, and thanks so much for the question. Obviously, you talked about the 16 of the top 20 customers or pharma customers in terms of three of the pipeline. Can you talk to us about the smaller customers? Are they also kind of in the same, you know, are they a little bit more reticent or they're just smaller? And sort of how do we think about the typical ramp of some of these buying? Are they kind of trying it out first and then sort of with future expansion plans in 27, if that goes well? Are they jumping right in? Just a little bit more color on how you see those two segments progressing over the next few quarters to year would be very helpful.
Great, Liz, this is Jeff. I'll take a first crack with a hand to Matt here. Short answer is, you know, we know that the medical legal review processes of the top 20 pharma are intensive, and we wanted to start there. So we started at the top, if you will, at the most difficult place. And that's been going well. It does take time. Obviously, we have to, you know, go through and understand, you know, suppression keywords and keywords and where things appear. And, you know, there's a lot of testing and all that, but it's all part of the process. And, again, we proudly work with, you know, this will be 19 of the top 20 pharma. With the three we have in contracting. In terms of smaller accounts, actually, things move faster there. But generally speaking, we haven't focused as much on selling to that audience because we think they're going to be the perfect clients for our new auction or biddable tool. And that will be, you know, less. A more efficient go-to-market motion for those smaller accounts. We started with our more strategic, bigger accounts, which is what we typically do. Again, our best new products go to our best or biggest clients first.
I actually don't know if I have that much more to add. I think probably one of the things that signals well is one of the data points from the preparedness marks, which is that Based on customer demand, we've made available kind of more than double what we started with in terms of available conditions, and that really just speaks to, I think, in a broad set of places that pharma customers are interested in, conditions, drug areas, different types of treatments, you'll see that there's just a lot of interest here. And Jeff's point. AI search is going to become a lot more accessible as we move into FY28. So we're excited. We're encouraged. I think the demand is there as we just continue to ramp it up.
Yep, that makes a ton of sense. Thanks for all the additional color.
Operator
Your next question comes from the line of Ryan McDonald with Needham and Company. Ryan, please go ahead.
Hi, thanks for taking my questions. Congrats on the quarter. Jeff, as we think about the example you provided with New York Presbyterian now providing access to the AI solution directly through the EHR, once that occurs within your health system clients, what impact do you see or what kind of uplift do you see on usage? And then how important has that sort of uplift or that EHR integration been as you've been going to market with the AI search product?
Great, Ryan. Yeah, thanks. I will say that with our patient context doctors or clients, we've seen roughly a doubling in the amount of usage. Even though they're able to type less themselves because they have more patient context coming in, there's a big time savings there. And again, I think that's really the future is Your doctor will be sitting there with you in a visit, and the AI will be there with him or her, and it will have, again, your history there so that nothing gets missed. And that's really a very exciting future. With regard to just, you know, turning on within an EHR, we've done that in lots of places. I got to be honest, the truth is just putting a button inside an EHR does not in and of itself get immediate usage. You actually have to do a bit of training and webinars. Maybe this is a sad statement on electronic health records, but, you know, a lot of doctors are trained not to click a button if they haven't been trained on it, and there's just a lot of that sort of heptid behavior around trying out new features or new things inside some of these systems which have lots of cul-de-sacs and So, we're good at that. Again, we've been working with them for years, usually, as a partner on telehealth and scheduling, and it's actually been fun for our team to have, you know, our monthly Zoom call be about a whole new series of features and products. And I think we've got many months of good Zoom calls there with our clients to help them see all the things they can do with it. But short answer is patient context does lead to an immediate lift, but just being inside an EHR, less than you might think.
That's helpful, Keller. I appreciate that. Maybe just as a follow-up, as you think about sort of expanding that top-down penetration within the health systems, how valuable do you think the up-to-date partnership can be, and is there any sort of go-to-market sort of relationship there where they can help bring you into, obviously, that very large customer base that they have? Thanks.
Thanks, Ron. You're absolutely right. UpToDate is virtually everywhere in the U.S. health system. And as I said, our teams have been getting along great. We think that, again, they're just a very high integrity, real boom for the overall health system. We're proud to work with them and we hope and expect to keep expanding our relationships. I don't have anything to announce today on go to market with them, but that's some of the things that have been discussed. But let's just say that I think it's the beginning of what we think will be a very strategic, a very, it's a long-term partnership, multi-year. We're excited to keep going with them.
Operator
Your next question comes from the line of Scott Schoenhaus with KeyBank. Scott, please go ahead.
Hey, guys. Thanks for taking my questions. I just wanted to follow up on Elizabeth's question. Just wanted to make sure I understood how long does a typical medical legal review process take, and how should we think about that ramp in AI search revenues after you clear that process? Is it very back half-weighted in calendar 27?
This is Jeff. I'll chime in first. I mean, for new products, MLR review can be months and months. And we learned this candidly the hard way back with our vertical video product, as you may recall, a few years ago. And, you know, we had a lot of signed contracts, but we didn't have the revenue coming in for a many month gap because Video, honestly, is difficult for lawyers to review well, and there's lots of tools and processes that need to be put in place to be able to comment on frame 68 or 25 seconds in. I think we're being appropriately cautious here in our timeframes, and we're assuming that it will also take many months. To get live with these contracts. So that does lead to, again, I think, again, a cautiously optimistic approach to the rest of this fiscal year.
I'll add a couple of points. I think one is that I think you see that assumption on that gradual ramp, especially kind of from a calendar year 27 standpoint. In our guidance, right, our growth rate in Q4 is very similar to our growth rate in Q3. And part of that is because we know that especially for customers that are buying AI search for the first time, like there will be that ramp up that just described in terms of the MLR process. I would say, you know, in terms of some of our early upfront activity, a lot of the AI search. Contracts kind of assume that too, right? It will be a ramp up throughout the next calendar year. So that's a dynamic that we're seeing as well. I don't think any of that is surprising based on what Jeff described about some of our past product cycles, but that's kind of what the assumptions are that we're making for our business.
Perfect. Thanks, Matt. And just as a follow-up, so that 6% growth rate for 3Q and then for 4Q, in which we start into the next calendar year, Typically, people read that as a reflection of your view of what maybe the market can grow at or what you can grow at. How should we view that in context of what you're viewing the market growth at as you've outlined 6% growth into the calendar new year? Thanks.
Yeah, I think it reflects, you know, the beginnings of reacceleration in our business, right? I provided the comment that our second half, our fiscal second half outlook has slight acceleration versus our first half. I think that's where you see the beginnings of optimism for us as we look into next year. At the same time, you know, it's early. We're not through our upfront cycle yet. And so I think we want to be thoughtful, you know, especially as we think about kind of fiscal Q4 and calendar Q1. See how the upfront lands. Also know that some of the AI surge demand in particular will ramp throughout the year. But I think we're, you know, we're pretty encouraged and optimistic, you know, as we sit here today in November for future growth.
And Scott, I'll just add on the market itself, I think there's some cautious optimism that hopefully next year is better than this year. This year was a perfect storm of a lot of different things not going the market's way. And so I think there's some reasons to be a little more optimistic for next year.
Operator
Thanks, guys. Your next question comes from the line of Richard Close with Canaccord Genuity. Richard, please go ahead.
Yeah, thanks. Congratulations on the report and the AI progress. So, Jeff, for Matt, just given the comments that AI may be additive and a different buyer, I'm curious, do you have separate sales force going after this, or? You know, is it the same account teams as the other products? And then with respect to the upfront contracts that you're working on now, is AI bundled in or is that, you know, completely separate since it's a different budget possibly?
Hey, Richard, this is Jeff. Right now we have an overlay, we call it, but basically a team approach where we have some new experts we have in the AI search who join our sales team in, you know, meeting with some of these search budget holders. And right now, the contracting, I mean, we typically have, you know, these master service agreements, and then we have these statements of work they're called. Sometimes they're together on the same statement of work. Usually, I think it's a separate statement of work. But again, this is the advantage of being a A reference client, someone who's already vertically working with all of the top 20 pharma, it's been relatively low overhead for us to add on this search AI business to existing clients.
Operator
Your next question comes from the line of Jessica Tassen with Piper Sandler. Jessica, please go ahead.
Hi, guys. Thank you very much for the question, and congrats on New York Presby and the 25-plus new AI hospital lens. So, as you guys have started to deploy AI search, including patient contacts in these enterprise contracts, and commercialize the AI search offering via pharma bookings, can you help us understand how you're monetizing this product, basically, you know, what the eight-digit bookings consist of? Are those enterprise subscriptions or pharma bookings or both? You know, what I'm trying to get at is just in these enterprise contracts, is AI search exclusively monetized via subscription revenue? So, ostensibly for those 190 clients, you know, it might hit at day one. And then is pharma revenue kind of exclusively generated from independent physicians or physician practices using Doximity Ask? Any color there on just how exactly these AI products are being monetized would be super helpful.
Hey, Jess, this is Matt. Thank you for the question. It's a good one. I would say the primary way we're monetizing, and when we talk about eight-figure bookings, it's primarily with our pharma customers and selling AI search as an advertising product. We do sell AI. As a separate enterprise agreement to health systems customers, we charge a nominal amount as part of our broader workflow suite. And so I think as we think about this going forward, we see the primary monetization engine being on the pharma side of our business, specifically through the ad products that we sell.
Okay, that's great. That's extremely helpful. And then just any early thoughts on ROI for the pharma educational content displayed in AI search relative to Doximity's core business? I think, Matt, you mentioned most of the early upfront bookings are core Doximity, so does that imply kind of parity ROI? Just interested in any early thoughts on ROI for the educational content in AI search. Thanks again, guys.
Hey, Jess, I think the answer is it's just too early. A lot of these programs just launched in the last few months. That being said, this is the most contextually relevant product we have. And so I think the working thesis is give us a few more months, we'll have more kind of results in. But this should be just as, I would say, ROI strong as the other products, if not better. And the nice thing is it's also making all of our product portfolio better. There's AI signal now in other products. You know, when you buy AI search, you're not just buying media. You're buying, you know, insights. You're buying retargeting to other parts of the platform. And so everything gets better. Everything gets better education at the right time to the right docs. So I think the long and short of it is it's a little too early right now, but the thought is it'll be just as good, if not better.
Operator
Your next question comes from the line of Alexei Gogolev with J.P. Morgan. Please go ahead, Alexei.
Hi, this is Destiny Jackson on for Alexei Gogolev. I wanted to touch on NRR. I saw that it moved down for overall to 103% and to 106% for top 20. Can you touch on what's driving that trend down, and what are the levers to re-accelerate expansion?
Hi, Destiny. Thanks for the question. The- I would say those metrics tend to follow our revenue growth, and I think we see that as being at a trough level. Like we said, we're expecting the beginnings of reacceleration in the second half. Usually, those metrics follow the rate of revenue growth. Thank you. Also, I would say, you know, as we move forward and something I kind of am thinking about, you know, as a newer person in my seat, like, I think these metrics may not be the right set of metrics to measure our growth going forward. I'll take the customer count one as an example. It's very focused on our largest customers that we've landed and had in place for a long time. It misses our small and medium growth segment, which has been one of the faster growing segments for several quarters now within our pharma business. So I think we'll see and evolve the right metrics for our next chapter of growth, but it's not something I'm particularly worried about as we consider the opportunities that are sitting in front of us over the next several quarters.
Destiny, I'd just add on the reacceleration point. I think there's three things you can look at. One of them is just AI search, new incremental. We're getting deeper, better breath at all of our customers, even the existing ones in particular. The core, the other products have higher value because of the search signal. There's more excitement around just the whole portfolio than we've seen in a while. And then, you know, my earlier comment, cost-optimism in the market itself will be a little bit better, at least a little bit better next year than this year. So I think all three of those could contribute to, you know, improving NRR as we go forward.
Thank you. And then as a follow-up, the customer count for the customer was over 500,000. That came down for 2Q. Can we touch on what drove that deceleration?
And, Destiny, I'm going to have you repeat the question again. It just sounded a little garbled on our end.
Sorry. It was on the customer count for 500K customers. It came down this quarter, so I just wanted any color there.
Oh, right. So... Q2 is actually usually one of the weaker quarters for that metric, just from an expand standpoint. We just see less kind of incremental upsell activity. That's basically what I would point to if you look at past quarters and past Q2s and past years, you see a kind of similar pattern of relative seasonal weakness. That's the primary explanation for that.
Operator
Your next question comes from the line of Alan Lutz with Bank of America. Alan, please go ahead.
Good afternoon, and thanks for taking the question. First, Jeff, thanks for the shout out on the survey. Really appreciate that. Can you talk a little bit about the evolution of the competitive landscape here in AI search? You're looking at really two very different markets in your traditional pharma customer and in the health system customer. We'd love to get a sense of how the competitive landscape in each is evolving over time. Thank you.
Thanks, Alan, and I appreciate you listening to the shout-out. I'd say I've reluctantly come to like your survey because it is 35 pharmaceutical marketers, which is good coverage, given the top 20 is most of the market, and you do do a good job of adding in all the various companies or partners that they work with. You know, your chart there pointed it out, I think, pretty clearly. We're at the highest level there of interest in incremental dollars we've ever had. Paid search, or Google, is at the lowest level it's ever been, and some of the new search players there have also dipped, as I think. You know, again, there are ups and downs in the market, and some folks are pleased with their partnerships and some are not. So I guess my short answer to your question is that I think you are pointing us to the fact that these really are two different markets. One is search, and one is media. And the good news is we're really expanding. As a year ago, we were, you know, a pure media company, and now we are a media and a search company. And as Perry pointed out a few moments ago, there's a lot of synergy between the two because The reality is Google, knowing that you search X here, means that they're better able to target you on YouTube there. And those two things go together very well, and that's what we see with our business, our telehealth, which, again, is at all-time highs, our scheduling, our other workflow tools in addition to AI search.
I appreciate that color. And then one for Matt, the margins in the quarter were really strong, and you talked a little bit in the Q&A around some of the operating levers you could see in the business. We'd love to get a sense of, are there opportunities to invest where it would be a high ROI investment? And how do you think about investing more in things like AI search versus maintaining these margins north of 50%? Thank you.
Yeah, and I think you see us making that concerted investment in our EBITDA margin for the full year. I think we have what we feel like is the right firepower to be able to make some of these incremental investments. We literally look at Things that we can do differently on a week in and week out basis, especially right now, and given it such a dynamic time, and that's led us to kind of, I think, continue to iterate on the infrastructure that we're using, begin to make investments like in the auction biddable product that Jeff was describing earlier, kind of play offense through some of these partnerships. We'll continue to do that. I think what you also see is that we have a really great business model as a company. It generates really incredible incremental margins, especially when we outperform from a revenue standpoint. And we have a really great team who excels at executing. And I think you saw that this quarter again. The metric I shared earlier in the Q&A, which is this quarter, our cost of revenue, which is highly AI-focused right now, grew at half of the rate of our prompt volume in our AI products, whereas last quarter those things were one-to-one. So we saw from that measure kind of a 50% efficiency on our incremental spend, and that is just a huge, huge kudos and credit to our team who continues to innovate on AI infrastructure. So we'll continue to play offense, very much so, but I'm also pretty confident in our ability to generate great incremental margins as we're doing it. Great, thank you.
Operator
Your last question comes from the line of Ryan Halsted with RBC Capital Markets. Ryan, please go ahead.
Thanks for squeezing me in and taking my question. Any perspectives you can offer on just the general health of HCP marketing budgets going into the upfront season? You know, just seeing a lot of biopharma activity, you know, lots of new drug approvals. Just curious if, you know, how you're seeing HCP marketing budgets so far.
I can take a first crack at it. This is kind of my first go around in this seat here at Doximity and kind of thinking through the upfront. Am I... You know, as I go through it and I compare it to a lot of the team's experience last year, I think it feels like a more stable environment. I wouldn't say it's world beating or having a massive growth uplift. I think there are some dynamics in terms of pharma approval pipelines that are improving, and whereas last year there was a lot of policy uncertainty and friction, I think that is stabilizing, although I wouldn't say, you know, it's And it's up and to the right. And so I think it feels like a more stable, a more comfortable environment to do business in, but not necessarily one that's dramatically re-accelerating from a macro standpoint.
Yeah, this is Jeff. I'll just add, you know, if you look at the iShares Pharma ETF, it's up about 40% year on year. And you think about where we were a year ago compared to today. Again, there were a lot of There's negotiations around price and tariffs, and there's just a lot of things in the mix that I think we look at this year as a more stable environment for us to grow and do our jobs.
Operator
There are no further questions at this time. I will now turn the call back to CEO Jeff Tagney for closing remarks.
I'd like to end by thanking the entire Doximity team for the hard work serving more doctors every day than ever before. So thank you, everyone, for joining.
Operator
This concludes today's call. Thank you for attending. You may now disconnect.
