The Hidden Algorithm That Decides Which Software AI Will Recommend | Tim Sanders, G2

0:00For discovery, I think most people, I certainly have, have switched to asking one of the AI models before they go any further. And you do that instead of going to Google and using keywords. You represent at least half the market according to our latest research. Like half of all B2B software buyers do it exactly how you do it, Craig. The plot is like the Oscars. Everybody talks about the movie. Not many people see the movie. Certainly there is an industry growing up around filling the training data with positive information about your company so that when you're asked, when the model is asked, it'll come up with your how big of a problem is that for you guys?

0:38When it comes to the purchase recommendation, ChatGPT and Gemini enter into a process called validation layer work at inference. What does that mean? They are validating that the entity that they're about to recommend is not going to lead to a regrettable purchase. I look in the future, Craig, three years from now. You express business goals and business problems, and the agent then will write the prompts and locate the software and perhaps buy it on your behalf with maybe a couple of checkpoints or guardrails. I see that coming. Hi, I'm Tim Sanders. I came to G2 in the fall of 2024. Today, I serve as the company's chief innovation officer.

1:21And you have a pretty interesting background. You are also part of the AI Institute at Harvard. and then you also have written a number of books. One, Love is the Killer App, which I've found fascinating. Can you just talk about those two aspects of your background before we get into G2? Yeah, glad to. Last three years, I've served as an executive fellow at the AI Institute at Harvard. And the charter of the institute is to democratize artificial intelligence for everyone, especially business leaders. And they do that with case level research. They do that with publications. They run executive outreach programs for a lot of companies you've heard of to really help people understand the simplicity of AI and more important, how it can be leveraged for business outcomes.

2:23Love is the Killer App. So that was my first book. I published it years ago when I was an executive at Yahoo. And the thesis of the book that is that in a very high tech world, it's never been more important to be high touch. And, you know, Craig, I wrote that with the advent of the internet and digitization, digital transformation. And here we are in 2026, high tech, AI is as high tech as tech gets, and high tech leadership, and judgment, and taste, and all those other great human attributes around emotional intelligence have never been more important. Yeah, yeah. And how did those two, your work at Harvard, and the philosophy behind this book, which is very generally that business should be conducted with empathy and compassion rather than sort of cutthroat, zero-sum point of view.

3:26How do those help what you're doing at, relate to what you're doing at G2? Love is the Killer app had this idea that when we talk about love in our business context, what it really means is the selfless promotion of the growth of the other. So when leaders show love to their teammates, they're growing them. When they show customer love, they're promoting the success of the customer. And I also like that the way that we do this promotion to scale is by sharing the three intangibles every human, every organization has that actually grow as you give them away. Your knowledge, your network of relationships, and your compassion.

4:06What appealed to me with Harvard's AI Institute is their charter. They wanted to share knowledge to promote the success of businesses in light of this opaque thing called artificial intelligence, which is keeping too many people on the sidelines waiting, not trusting it yet. I study trust a lot now. I felt like there was just a real synergy between my longstanding belief around knowledge sharing is love and the AI Institute's belief that we're going to share knowledge to democratize the most powerful technology in the history of civilization. okay uh and g2 i mean its mission is to be the uh the arbiter is that a way to describe it of uh software uh particularly uh software as a service yeah so that you have an impartial a view of all of the software in a particular domain or for particular tasks.

5:13Can you talk about that, how G2 evolved, what it is today? You know, G2 was founded around the idea of creating the highest form of intelligence for buyers of software. G2 stands for the highest form of military intelligence, for example. The original name of the company was G2 Crowd. The belief was the crowd, those people writing reviews, the community, the peers, are the best way to generate the highest form of intelligence. At the time, our founders were struck by the, if you will, pay-to-play model of analyst firms like Gartner and how difficult it was for a software company to establish trust with the buyer and actually become successful as they added more and more capabilities.

6:02So that was the driving motivation. And today, when you look at how the agents at Frontier Labs, like ChatGPT, OpenAI, ChatGPT, and Google's Gemini, when you think about their logic today, they absolutely love human natural language as expressed in reviews. They love the idea that a company like G2 has reviews to scale. And as a result, we've become the dominant player in driving AI search results when you're trying to buy software. We're often going to be the most cited third party when recommending the answer, which just illustrates over more than a decade of work how far G2's come. I would say though, Craig, that we had the founding mission to provide this highest level of intelligence for the buyer.

6:51And let me tell you where we are today. I hear a lot about this in our meetings. G2's new mission is to become the trust layer for the age of artificial intelligence. And that's where we are right now. We want to help buyers and owners and users establish trust in artificial intelligence, especially as agents and autonomous computing come online and offer people a breakthrough in business results. Not 30%, not 40%, but what we're hearing now, 2x, 5x, 10x in certain situations. Only if they trust it, though. And trust is a hard thing. Yeah. And the reviews, are these reviews that are written exclusively on G2's platform, or do you collate them from different sites around the world?

7:47They're written exclusively on G2's platform. Some are typed increasingly. Others are just spoken. We have a voice product now that actually leads to longer reviews. I would say over half of what is submitted on the platform doesn't make it to the website because our trust and safety approach is to not only verify the identity of who's writing the review, by the way, the agents love that for establishing trust. We also verify that you own the software and that creates a lot of friction, but at the same time, it differentiates us from others, from communities like Reddit or other review sites that don't go to that rigor in the spirit of trying to aggregate the highest quantity of reviews.

8:33We could have twice as many reviews as we have, but we feel like we have enough. Yeah. Although it would be interesting to cross reference your your reviews. I mean, in aggregate with reviews on on Reddit or another platform to see if they try. I love that idea. It's something that I've been thinking about for a few years. I believe that especially for buzzy, hot new things like a launch of a new language model. I truly believe there are thousands, if not tens of thousands of unstructured reviews living in communities like Reddit that really could be cross-indexed, AI-sourced. So certainly I wouldn't be surprised to see if something like that would come online to even give more signals.

9:21Because what I would say is that, yes, Reddit is anonymous, but they have a wonderful Redditor system, if you will, of upvoting and other such things that can separate the quality from the spam. So I do believe there are high quality signals. By the way, Reddit is top three B2B software citations when you're looking to buy software. It's G2, YouTube, Reddit. That's top three right now. Wow. In G2, you have over 3 million verified reviews now. That's right. That's right. 90 million annual visitors and 200 million software buyers globally. That's right. We've added a little bit with the acquisition of Gartner Digital Markets, which involved the brands Capterra, Software Advice, and GitHub.

10:10Yeah, what are those three? And what is your relationship with Gartner? I mean, do they use your reviews? Do they refer to your reviews? Nope. We just made an acquisition of three of their properties, which were rolled up in a property called Gartner Digital Markets. and we completed the acquisition February of 2026. Gartner continues to be an analyst firm. They still own the review site Peer Insights, which their analysts use, but we bought the other elements of that business. Capterra's leading software review site, probably top three or top four prior to the acquisition, depending on how you want to count it.

10:53Software advice, fascinating service. If you're buying software, you can call a 1-800 number and have a conversation with an expert that will help configure for you the exact solution you're looking for. We think that's a great layer of intelligence to feed trust. And then GitApp really focuses on startups, very small businesses, in terms of how they buy software to fuel their companies. And it also promotes startups and small businesses that are just coming online, trying to compete with the incumbents. Yeah. And what's the business model? I mean, are you guys a subscription model? I mean, for example, GetUp, that's, I'm a small business and I'm always trying to figure out, you know, what software I should be looking at.

11:38Is it pay for, what's the term, you know, usage payment or subscription? It's not yet. It's subscription. And we can talk later on, I think the world is going to move to pay for consumption. But let's just talk about G2's business model, marketing services. So G2's business model is we provide a variety of marketing services. We provide platform access. You subscribe by claiming your profiles on G2 for your products. That gives you tools to generate higher quality reviews, have a higher quality presence on G2, which is often what cited your profile on G2. We offer review as a managed service to help companies augment what reviews they can naturally collect so they can scale that voice.

12:25We offer buyer intent data, not only like who's looking at your products and services, but more interesting like which competitors are looking at, you know, their customers are looking at your products and services. So that data intelligence now available via MCP. And then finally, advertising. So we sell media opportunities for software brands to extend their reach in this evaluation stage of the buyer journey. And I would say that when you think about Google Ads, which is like the massive, massive business in advertising, that's more discovery. So you start on Google, but you end up at a review site, analyst firm reports, publishers, experts like yourself.

13:09That's called the evaluation stage. So we offer advertising solutions for companies that say, when that customer has that shortlist in hand, they're trying to evaluate, we want to show up there either to be picked off the shortlist or to disrupt the process and enter the conversation even if we didn't make a chat gpt shortlist these days you said that g2 is g2 youtube and reddit are the the top three places that people go for uh to to analyze software or understand uh users they're the most cited i think the way to think about that is if you are uh putting out a commercial intent prompt and if you like i'll share something with you.

13:52If you're putting out a commercial intent prompt and you're saying, you know, I'm looking for a CRM software where I want it to be medium-sized hospital, month-to-month payments, and it works great on mobile phones, then whatever the shortlist is it gives you, it will cite various sources that it relied on to make that recommendation for those three CRMs. That's what we track at G2. So we're currently number one by a margin. Number two, depending on the month is going to be Reddit or YouTube as being the cited source behind the recommendation. Right. So we track our influence since our business is marketing services.

14:32For discovery, I think most people I certainly have have switched to asking one of the AI models before they go any further. What are the top software for this particular use case? You do that instead of going to Google and using keywords. So you're writing a natural language prompt. You represent at least half the market according to our latest research. Like half of all B2B software buyers do it exactly how you do it, Craig. This time last year, it was only 29%. If I'm looking into the tea leaves next year, probably two thirds of the people are going to behave like you do starting in chat, maybe even generating the shortlist out of chat and then doing the evaluation work.

15:21yeah and something that i've wondered uh and certainly there is a an industry growing up around filling the training data with positive uh information about your company so that when you're asked when when the model is asked it'll come up with your how how big of a problem is that for you guys? Because I don't know. And what is the mechanism for doing that? So, yeah, let's talk a little bit about that. So it's a huge AEO, answer engine optimization, huge industry. But I want to make a distinction here. The purchase recommendation use case is the only use case G2 is concerned with because we sell marketing services to help software companies drive purchase, right?

16:10There's many use cases on ChatGPT. You could do fact lookup. You could do document summary. You could be tool use. But purchase recommendation represents a unique risk to the user. OpenAI researchers call this your money, your life. So when it comes to things like medical advice or purchase recommendations, OpenAI, and for that matter, Google Gemini, they don't just rely on pattern recognition, matching the natural language and the prompt to natural language on a website. That's what many companies are retooling to do, is to have all that language on their website to win the pattern match. You may win in a variety of use cases, but when it comes to the purchase recommendation, both of the models I mentioned, ChatGPT and Gemini, enter into a process called validation layer work at inference.

16:55What does that mean? They are validating that the entity that they're about to recommend is not going to lead to a regrettable purchase. How do they do that? They look for trust signals from high authority sources, usually with strong underlying data. One thing I would share is there's a fantastic study out there by a company called First Page Sage. And First Page Sage, they broke down that algorithm for how the agent at ChatGPT is thinking at inference time about validation. And so as you can see here, 41 % of the weights that drive a software company being recommended is their appearance. And quite frankly, they're ranking on authoritative lists.

17:39lists. Another 18 % is driven by awards, accreditation, 16 % by online reviews. All of those are third party. They can't be generated by a vendor. They must be acquired. Now, what are the authoritative lists? They could be review site lists like G2's Best of Software Awards. Last year, we did an audit for where our citations came from. 60 % of our citations came from the best of software awards and all the best of category pages across G2. Okay. Other sources could be authoritative lists published by analyst firms, Gartner Forrester, right? If they're released to AI for crawling. And then publishers, say TechCrunch, top 10 CRMs for hospitals.

18:22But again, the models are distinguishing verified identity of the user with sentiment expression. They're also looking for just like the number of people involved in the generation of that authority list. This is different than listicles. So let me come back to what you asked. Is there an industry helping software vendors create content to show up and have AI visibility? Absolutely. But AI visibility is not the same thing as winning the answer. Being cited means a link to your content is showing up in an AI return. In the case of that CRM prompt I gave you, the medical month-to-month works great on iPhone one, that return from AI isn't just the three CRMs.

19:08It's several pages, really breaking down CRM in hospital, the rise of it, mobile endpoints, different vendors that offer Android versus iOS, different pricing matrix, the advantage of month-to-month versus a flat fee, for instance. And then at the end of that three or four page return, it makes three recommendations. You can use one of these services to generate all of your content to show up in the first three pages, get a little traffic from it. But that doesn't mean you're going to win the answer. So I always tell people being cited doesn't mean you were recommended. Those are different things.

19:43Being cited is how you generate traffic to your website, which hopefully you can convert. Winning the answer generates pipeline. And that's the business we're in at G2. Yeah. And you can't do that with first party content. You can't win the answer with vendor claims. Right. And you're talking about this evaluation layer. Is that how you refer? Yeah, they call it the validation layer. Yeah, it's fascinating. And it's driven around like the regrettability index. Like I'm trying to look for signals that predict, because that's what these machines do. They make predictions. I'm trying to predict that if Craig buys this CRM solution, he's going to renew in a year.

20:19Because if you buy it and you hate it, that leads to platform health decline. And that regrettability index has been around since social media and regrettable minutes years ago. Oh, and so that is something that the model builders are deliberately building into their models. Yeah, well, they're building it into fine-tuning for various use cases based on prompt categories. So I said purchase recommendation. What triggers a model then to look for fine-tuning related to purchase recommendations called a commercial intent prompt. It's where the natural language expresses the desire to purchase something.

21:01And that commercial intent then triggers the model to say, oops, we need to do validation layer work, your money, your life. Craig, it's the same if I ask medical advice. If I say, I want you to look at this picture and tell me if it's skin cancer it's so it's uh-oh your money your life we need to go start to really identify high authority sources to gain confidence to tell you and generally speaking in a situation like that it's going to be very iffy anyway i mean it's not going to be very middle but anyway yes that that's that's that's what we've learned over the last few years that's just fascinating and and you mentioned open ai and google or gemini those are the ones What about Claude, which increasingly is the following?

21:44Well, we could talk about this. Claude is being adopted by a lot of companies. That doesn't mean their employees use it to research software. G2, we studied this. We studied it every year. What do you use the most when it comes to work and doing research for software? The answer is Claude is like the Oscars. Everybody talks about the movie. Not many people see the movie. Claude is being used for coding. It's being used for summarizing documents, Slack conversation. It's used to automate various tasks that are usually weekly. Those are called cron jobs. And most popular these days, Claude is being used to create presentation decks.

22:19But it's not being used to do primary research. The primary research, and if you look here, the market shares 81 % is ChatGPT and Gemini. I would also say Claude has the lowest net promoter score for the research use case of any of the ones I just showed you, including perplexity, because it's just not a priority. Labs devote tokens at inference to the things it thinks it's going to win by, and for CLOD, that's coding, and increasingly, it's design. Their reasoning is mid. Their reasoning, to me, is not much better than DeepSea. Right. So they don't have this validation layer or validation model.

22:57They may, but they don't call upon the breadth of data that, say, OpenAI and Google do. So, for example, a profound, a leading AEO researcher, Josh Blyskall, revealed that less than 40 % of the time when you're on cloud doing a research call, it does live retrieval, like less than 40 % of the time. When I'm on OpenAI, 100 % of the time, it does live retrieval. It changes the results. So even if the algorithm's the same, I just think that the users are seeing more performance for research. The other thing I'll just quickly say here too, Greg, is that a shocking number, more than one out of four b2b software buyers are using their own chat gpt their own phone because it's got their memory they don't want to use the corporate one with all the guardrails they're very good at using it etc etc um and and then it's really the chat gpt gemini show especially chat gpt when you're talking about that yeah and so if you ask uh chat gpt 5.6 or whatever the top model is for a recommendation on a particular software.

24:11And you asked, Claude, are you likely to get different answers? Yes. Well, you're likely to get different answers from OpenAI if you ask it today and ask it tomorrow. That's why that slide I showed you that tracks G2 citations, it's not based on one prompt. We're Profound's biggest customer. They run hundreds of thousands, if not millions, of synthetic prompts every month to get that statistical body that generates real directionality. Like when you hire an AEO firm to build a dashboard for AI visibility, what they do, Craig, is they work with you to set up something called a prompt panel. Think of it as questions you're tracking.

24:51And then what they do with their software is take your questions and what they call fan them out to versions of your questions so they can run thousands and thousands or hundreds of thousands if you can afford it. Synthetic prompts to generate synthetic returns. And by averaging them all, they can give you pretty good analysis of AI visibility. On the other hand, if you just went to chat GPT and put the prompt in to see if your company shows up, it's going to vary every time you do it. But if you did it enough, you would reach a statistical average and have directionality. Yeah. And again, just because I want to get beyond this, but what are they doing?

25:31I mean, that's a good explanation of how they're seeing what the average is across multiple sessions. But what are they doing to get a company, their client, into the fine-tuned data so that they show up? Most of the companies are hiring AEO firms to recover the traffic they lose because of the zero-click tendency of users like yourself. So, again, G2 is laser-focused on pipeline, like showing up on more short lists, winning more shootouts. That's where we play. A lot of the AEO firms, they have more of a monolithic view of AEO, like companies need to get back their traffic. So if you're cited frequently, then theoretically you get clicked on.

26:22Now, the disappointing part here, PromptWatch has done some research around click-through rates, Craig, and the average click-through rate for the average citation in a return is somewhere between one-tenth of one percent, one-third of one percent. Now, if you're cited in relationship to a recommendation, PromptWatch found it's 1 % to 7%. It's significantly higher. That's why G2 got a million human visitors from OpenAI UserBot just last year alone. my point is the AEO folks can help you create the type of content and the schema structure of your website that makes it very easy for the machines to consume where you have visibility you're being mentioned your content is being cited and there's ever the chance that you'll see some traffic come back now what we do know is the traffic you do get from chat bots is higher value traffic because usually by the time they click through to you, they have more conviction.

27:19They were coming based to do evaluation from a short list or they're just further along the buyer journey. So the traffic that you do get is much higher quality. Some might say as high as two, three X. And so I work with a lot of companies now that are beginning to partition off like here's the traffic we're getting from the machines and it's very valuable. Here's the traffic we're getting from humans. But the reality is most companies are realizing organic traffic is going to start getting close to zero as time goes by because we get our answers in the chat and there's no read to poke around.

27:54The one thing that I would say quickly though, the reason that the click-through rate is highest, the deeper you get into a report where the recommendations live and it's the lowest at the beginning of a chat report where all the explanations live is because in 2026, we have this like, this trust, if you will in chatbot results. So now we're at a point where we only click on what is risky to believe on its face, your money, your life. So that's why when you see a citation next to a recommended purchase, you are the most likely to check on that and evaluate. Yeah. And just again, these AEO firms that are trying to boost your visibility in chatbot responses, Are they, and I understand they're creating relevant content, but then are they like building a million websites and spreading it around the internet?

28:54I mean, so that the crawler will or the internal search of the chatbot finds them. I mean, you know, just optimizing the content on your website may not do much if your website doesn't get a lot of trouble. Yeah, it's foundational, though. If your website is not optimized to be easy to crawl in training or easy to retrieve at inference, it's kind of game over. So there's companies, Scrunch is a great example. They were just acquired by Sitecore. Their agents can create website remediation, meaning literally spinning up new code for you to improve the machine readability across your site. Different AEL vendors have different specialties.

29:43Aerops, they're great at content engineering. They can have agents identify content gaps across your site where there's a market demand for a certain type of content you're not providing. You provide that content. They call it frontier content if you're first on the scene. And you're going to see an outsized number of citations against that content. I mentioned Profound. They have general purpose marketing agents. Profound can do everything from doing constant reconscience and daily reporting to you on how your visibility may or may not be moving. They can generate either suggestions or actual content artifacts for you.

30:16They can also provide remediation of schema. And I'm seeing a lot of the AEO companies really moving from just dashboards two years ago to always-on agents today to help companies kind of catch up in this world of AEO, at least from a traffic generation standpoint. Yeah, it's fascinating. And that really is really - I haven't seen anything like it since I was CSO at Yahoo 25 years ago. Yogi Berra said it best, it's deja vu all over again. Yeah. You recently introduced MCP capabilities, a model context protocol, which allows models to interact with your software that let AI systems reference G2's verified reviews.