How To Get More Enterprise Leads from AI Search

Here are the basic steps for how to get more enterprise customers with GEO:

1. Listen to recordings of your sales calls with your top 5 enterprise customers
   1. Note exactly what those 5 customers say and exactly how they describe their needs
   2. Create a simulated prompt for ChatGPT using their own language
2. Run those 5 prompts in ChatGPT/Claude/Gemini about 50 times each (this is enough to get directionally reliable results even though LLMs vary from answer to answer)
3. Make a list of which sources are cited the most
   1. If the sources are your competitors’ own sites, build equivalent pages on your site
   2. If the sources are third-party websites, reach out and ask to be featured
4. Make a list of which competitors are recommended most often
   1. Take your top 2-3 competitors and build a list of every site that mentions them (tools to do this recommended below)
   2. Build a list of every site that mentions you
   3. Find the gap between your competitors and your own site
   4. Do outreach to the same types of sites that feature your competitors to help close the gap
5. To measure performance, track three signals or KPIs:
   1. Visibility in the prompts you’re measuring (e.g. “before we were recommended 55% of the time, and now we’re recommended 65% of the time”)
   2. Self-attribution from post-purchase quizzes (ask new leads “where’d you hear about us?” and save the answer to your CRM)
   3. Conversions in GA4 or demos with ideal customer tracked in your CRM
6. Repeat the process with your next 5 best customers or use one of the advanced scaling strategies below to get better results over time

Every step in this process is simple to understand, but can be hard to execute. For example, listening to five sales calls tells you how your customers talk to salespeople, but doesn’t specifically tell you how they talk to AI. However, using your customers’ exact words - not a marketing persona you created four years ago - will get you initial results. Then, once you have momentum, you can use more advanced strategies to get even better results over time.

This article is meant to walk you through that process from beginner to advanced. You’ve already read the basic version of how to get started above in the intro. In the next section I’ll expand on that outline enough that you can actually do the work, including recommendations for tools, frameworks, and processes. At the end, I’ll explain more advanced strategies you can use to get even better results and show you how to scale this process over time. And, I’ll explain why AEO has given you an opportunity that SEO never could: the opportunity to get in front of all the perfect customers you do want, and none of the low-quality leads you don’t.

NOTE: This blog is under construction, but I’m surfacing anyway in case someone can find value from it while I refine the ideas. I’ll refine over time.

• The key to this entire process - build around your ideal customer
  • The key to this entire process is starting with your ideal customer in mind. Your best customers don’t talk to ChatGPT the same way everyone else does. To get more of your best customers, you need to show up for their specific questions, not general questions like “what is the best crm”. Of course, exact accuracy here is hard, but starting with your customers’ exact words will get you moving in the right direction.
• Startup Phase: How to Get Your AEO Practice off the Ground
  • Figure Out What Prompts To Track
    • Use their exact language bc keywords no longer exist
    • They’ll either talk stream of consciousness using Wispr or whatever, or type stream of consciousness without editing
    • This is the behavior of an AI native, and even if your peeps aren’t doing this now, they will be.
    • Create one prompt for each of your 5 best customers. You’ll end with 5 prompts total.
      • Ideally we’d have 5-10 interviews per ideal customer persona, so scale up as you want. But I’d recommend starting with 5.
    • Mimic your customers’ language the best you can
      • Examples: if they say “software solution” say “software solution” not “software”. If they say “something to help me fix this problem” make sure you say “something to help me fix this problem”, not “software”
      • Only caveat: make sure your prompt is at least semi-product intent. You know these folks want a product kinda like yours otherwise they wouldn’t be on a sales call. So run your final prompt through ChatGPT or Claude once or twice manually and just see if it recommends a product in your general category at all.
        • E.g. vocal video prompts
          • Three examples
      • If ChatGPT recommends a diy spreadsheet or something completely basic and unrelated to your product, it’s going to be much harder to influence them to recommend a product like yours. If they recommend a crm, but it’s just not your crm, that’s fine. If they recommend a SaaS, but it’s not a CRM, that’s probably fine too.
    • Advanced: the best way to improve results here is to do research on your customers’ AI behavior, and to do more than 5 interviews.
  • Track the Prompts
    • Okay, about 50 times is the baseline for getting directional measurements. 100 times is better still, 350 times is best. Above that is probably not worth doing. Exact details explained later.
    • Pull results for each prompt you created over API. You can have Claude create a simple runner for you with almost no work. Just feed it this article and it’ll create a runner that connects to your ChatGPT API account and stores the results in a local SQLite database.
    • Setup is very simple. You just need to set up a ChatGPT/Claude/Gemini API account (I’d start with ChatGPT because they have the most users by far . The API account is different than your standard account), and tell Claude or Chat what you want it to do.
    • Make sure to set search to “auto” and ask ChatGPT to pull all search results and searches back.
      • Reasoning: Auto is the default setting and what most users use. Explicitly telling the AI to pull the searches it performs and the sources it uses is what enables you to build a usable database.
    • Cost for each prompt will be about $5 for 50 runs.
      • It’ll be a bit more expensive with the searches and search results pulled back, but still will only be about $5 per run, so $25 for all 5.
    • If you want to use a rank tracker go for it. Basically a rank tracker is someone else’s opinionated take on advanced prompt analysis. But they often come with annoying limitations and are not necessary if you’re handy with Claude Code. If you trust the person running the tracker, go for it, otherwise read the advanced section and make your own decision.
  • Find the Gaps
    • Once you have all your results in a database (5 prompts x 50 runs = 250 total runs), start by asking ChatGPT one question:
      • Over many runs did ChatGPT search? Tell me simply in number/total and a percentage.
      • It’ll tell you, “ChatGPT searched in 247/250 runs, or 94% overall”.
    • Then ask: “of those search results, how many times did ChatGPT include a specific brand name in the search, or a “site:” qualifier? Tell me a number/total and percentage.” it’ll tell you ChatGPT included a brand or site qualifier in 247/250 (94%) overall
      • The point of this is to tell you how much of your strategy should focus on showing up in search results vs training data. If ChatGPT uses “site:” or brand searches a lot, you need to focus on getting into training data. If they use those searches very little, you can focus more on showing up when they search (in citations/sources).
      • Reasoning: If ChatGPT is searching for specific sites (using “site:”) or performing searches that mention specific brands (insert example), that tells you they already know who they want to consider or recommend before they search, and they’re just searching to confirm their original answer.
        • Example of each again
      • In this case, you need to study the overall internet profile of the brands ChatGPT is already searching for and try to build a similar profile for your brand over time.
      • This also tells you how long term your strategy will have to be. Showing up in training data = longer term, budget for a year to 18 months and hope for less. Showing up in citations or sources = shorter term, budget 6 months and hope for less
    • Ask ChatGPT to build you a Google Sheet with two tabs:
      • 1. Citations & Sources, where each row contains:
        • Website
        • Whether the site was cited or sourced [insert image showing example]
        • How many times it showed up
      • 2. Brand visibility, where each row contains
        • Competitor
        • Whether they were recommended or just mentioned (mentions can be negative, like “I wouldn’t recommend X for you”)
        • How many times they were recommended
        • How many times they were included in a brand search
      • These two reports show you:
        • Your visibility
        • Your competitors visibility
        • Which competitors are winning
        • What websites ChatGPT is using as resources
    • To start, always at least match your competitors page for page. Meaning, if your top competitor has 7 of their own pages listed as sources for ChatGPT, you should build pages just like those 7 on your site. That’s the bare minimum. Make sure to work these pages into your overall site structure and SEO strategy so you’re not cannibalizing your other SEO efforts.
      • Then, look at the third-party sites ChatGPT lists as sources. And yes, look at the individual sites, sure, but more importantly, look at the types of sites. If ChatGPT uses G2 as a resource, but you cannot get into G2 for some reason, Capterra is probably a pretty good substitute, as is SoftwareAdvice.com. But a Forbes article is NOT. Your priorities should be to:
        • A. Get into all the exact sites that ChatGPT uses as resources
        • B. Get into similar sites and build a competitive profile compared to your most-recommended competitors
    • Build training data profile
      • Backlink checker pro is good for backlinks
      • Dataforseo is good for brand mentions
      • Pull in other for YT
      • You can't get an exact profile but this will show you directionally. It's not like you're coming up with every site. It's more like, if you know “man athletic greens is going hard on yt sponsorships” at least you have an understanding at the right order of magnitude.
      • To do this better invest in more and better tools. I'm in the process of trying more and testing, otherwise you can do it yourself.
      • Do this for top two competitors. Only one might show you aberrations from one brand. Two at least shows you a bit of pattern data.
      • Build the same model for your brand.
      • Compare.
      • You'll probably have to sort out the garbage sites and group some sites into categories. I'll create a guide on this at some point. Vote if you want it.
  • Fill the Gaps
    • Gaps fall into one of four categories:
      • Pages you build on your own site
      • Sites you have to build (a la PBNs)
      • Sites you have to do simple outreach to (a la standard SEO backlink style outreach)
      • Sites you have to build relationships and/or pay big bucks to get featured in (affiliate style relationship mgmt)
    • Prioritize by lift and impact
  • Measure the Result
    • Track conversions in ga4 or demos in your crm
    • Self reported from quizzes but follow up in demos
    • Visibility in the 5 prompts
    • Next step is to track 10+ prompts, and to expand to all enterprise personas, up to 25. Also try building headless and tracking in the browser, not via just via API.