This article was originally published on Law.com here
It takes fifteen seconds. A legal ops director asks ChatGPT a question and walks away with a shortlist: three or four vendors, a line on each, maybe a citation. The week of demos and Google searches now fits in a single prompt. For the buyer, that’s progress. For any vendor left off the list, it’s a lost deal they’ll never even see.
This is already happening—and faster in legal tech than most of the industry realizes. General counsel ask Perplexity about contract lifecycle management. Paralegals ask Gemini what peers use for trial prep. Litigation support managers prompt ChatGPT to compare document review platforms. The buying journey that once started with a search now starts with a conversation, and the AI’s answer shapes the shortlist before a vendor ever gets a call.
The discipline taking shape around this is Answer Engine Optimization, or AEO. SEO is about ranking in a list; AEO is about earning a spot inside the answer itself. Different problem, different playbook—and so far, most of the advice is guesswork.
We wanted data instead. So we conducted original research: 110 buyer-intent prompts tested across ChatGPT, Perplexity, and Gemini. Every response manually reviewed. The dataset: 329 AI-generated responses, 3,266 brand mentions, 1,375 cited URLs across 697 unique domains. The prompts were designed to sound like a buyer who’s past browsing and into shortlisting, across multiple verticals.
The results upend a few assumptions most marketing teams have never thought to question. Here are the five things they point to—and what to do about each.
Build the comparison page your competitors haven’t.
Comparison and “alternatives” pages accounted for 29.2% of all citations in the study. When the prompt was a comparison question (“What are the best alternatives to [Vendor X]?” or “How does [Product A] compare to [Product B]?”), that jumped to 56.3%.
The reason is simple. A comparison page is already formatted as an answer: it names multiple products, weighs trade-offs, and matches fit to use case—the exact work an AI would otherwise have to assemble from scattered sources. A product page doesn’t. A case study doesn’t do that. A comparison page does it by design, and AI assistants reward the convenience.
Now think about how buyers in this market ask questions. “Which document management systems work best for midsize firms?” “What are the best alternatives to our current e-billing platform?” “How do CLM tools compare for legal departments with a small ops team?” These are real prompts, and the vendors whose websites directly address them in a structured, substantive way are the ones getting surfaced. The vendors with a features grid and a demo request button are not.
The opportunity is wide open precisely because so few companies have built these pages. In a market where the competitive set is well-defined, a thorough, honest comparison page is one of the fastest things to ship and one of the hardest for competitors to ignore once it exists.
Open up your documentation—AI trusts it more than your blog.
This one took us by surprise. Help centers, implementation guides, API references, product documentation, and knowledge bases were cited more frequently than blog posts, executive bylines, or thought leadership content. Consistently. Across categories.
The logic, once you see it, is hard to argue with. AI systems need content they can cite with confidence: factual, specific, clearly structured, unambiguous. A blog post about the future of legal operations is opinion-forward. An AI can’t use it to justify recommending one platform over another. A documentation page that explains exactly how a platform handles matter budgeting, invoice review, or integration with an existing DMS is something an AI can point to and say, “this is what this product does.” That’s the kind of source these systems reach for.
The problem is how most companies treat documentation. It’s a support function. It sits behind a login. It’s written for existing customers. Marketing never touches it. In the context of AEO, every one of those decisions is working against visibility. The vendors whose product documentation is comprehensive, publicly accessible, and well-organized are building citation-generating assets without necessarily knowing it. The vendors whose documentation is locked behind a customer portal are invisible to the systems that matter most.
Make sure AI can read your site at all.
73% of the sites in the study had technical configurations that block AI crawlers from accessing their content. Not partially. Completely. Robots.txt rules, JavaScript rendering requirements, authentication walls. The content behind those barriers might be excellent. It doesn’t matter if AI can’t get to it.
This is the kind of problem that makes everything downstream irrelevant. A company can invest in comparison pages, documentation, community strategy, and cross-platform monitoring, and none of it will work if the retrieval bots AI assistants use to gather source material can’t read the site. It’s the equivalent of publishing a magazine and forgetting to distribute it.
Enterprise software websites are particularly vulnerable. Gated resources, client portals, single-page application frameworks, dynamic rendering: these are standard architecture choices that happen to be exactly the barriers AI retrieval systems struggle with. And the bots these systems use aren’t Googlebot. They have different user agents and different rendering capabilities. A site fully indexed by Google can be completely invisible to ChatGPT, Perplexity, or Gemini.
A crawlability audit against AI-specific user agents takes days, not months. It’s the most fixable problem in AEO, and it should come before any content investment.
Pay attention to the channels AI actually listens to.
Reddit was the single most-cited domain in the entire dataset. Ahead of G2. Ahead of Gartner. Ahead of every vendor blog tracked in the study. LinkedIn, where most B2B marketing teams spend the bulk of their publishing effort, accounted for less than 1% of cited URLs.
The channel vendors pour the most energy into is nearly invisible to AI. The one most ignore is a primary source these systems draw from.
The reason seems to be about signal quality. A Reddit thread where practitioners compare tools they’ve used, with follow-up questions, disagreements, and real specifics, reads to an AI as authentic peer evaluation. A LinkedIn post from a vendor’s marketing team, however well-crafted, reads as promotional. The AI weights them accordingly.
This doesn’t call for a Reddit marketing strategy. Forced community participation backfires fast, and anyone who’s tried it knows that. But it does mean that community sentiment has become part of the citation layer, and most marketing teams haven’t accounted for it. When attorneys discuss litigation support software in a professional forum, or when legal operations teams compare CLM platforms in a practitioner Slack group, those conversations have downstream effects on which vendors AI assistants recommend. The polished content published on owned channels is barely reaching these systems. The conversations that happen without any marketing involvement are shaping the answers.
Track all three engines—the same question gets three different answers.
Same prompt. Three AI assistants. Three nearly different sets of vendors. Only 3% of cited domains appeared across all three. On roughly one in six prompts, the three engines named zero brands in common. None.
This is not a minor variation in ranking. It’s fundamental disagreement about who belongs in the answer. Each assistant has its own retrieval logic, its own source preferences, its own way of deciding which content is trustworthy enough to cite. A vendor that shows up reliably in ChatGPT’s response to “best e-discovery platforms for midsize firms” may not appear at all in Gemini’s answer to the same question. A source Perplexity trusts might be absent from ChatGPT’s citations entirely.
The implication is that monitoring one AI assistant and building strategy around what you see there will miss most of the picture. Testing buyer-relevant prompts across all three platforms, tracking which competitors appear, noting which sources get cited, and finding the gaps needs to become routine. In a market where a relatively small number of vendors compete for the same buyers, knowing where you’re visible and where you’re not across all three platforms is the difference between a strategy and a guess.
The Bigger Picture
One more finding from the study puts everything above into context. 7.8% of the pages AI assistants cited don’t rank in Google’s top 10 for the equivalent query. AI citation and search engine ranking have measurably come apart. They are now two distinct problems requiring two distinct approaches.
Marketing teams that have spent years building strong SEO programs will want to treat AEO as an extension of that work. The data says that won’t be enough. The content types that earn AI citations are different. The platforms that carry influence are different. The technical requirements for access are different. And the competitive landscape inside an AI-generated answer looks nothing like a search results page.
The legal tech vendors that build for citation, not just ranking, are the ones buyers will find when they ask AI for a recommendation. In a market this small and this well-defined—where everyone is chasing the same general counsel and legal ops teams—being early here isn’t an edge. It’s the whole game.
About the Research
This article draws on original research conducted by The Proxy Agency analyzing AI-generated purchase recommendations across multiple B2B verticals. The study tested 110 buyer-intent prompts across ChatGPT, Perplexity, and Gemini, producing a dataset of 329 responses, 3,266 brand mentions, and 1,375 cited URLs across 697 domains. Full methodology is available upon request.
Reach out at hello@proxyagency.com.
About the Author
Haani Kapasi is a B2B marketing specialist at The Proxy Agency with nine years of experience in SEO and content strategy, now focused on the emerging discipline of answer engine optimization (AEO) and generative engine optimization (GEO) for legal technology vendors.
