We had just wrapped up an AI demo when a customer told us something I did not expect.
They were amazed by the range of AI work we had already delivered at Abilytics. We had walked them through use cases, case studies, and some of the engineering decisions behind the work. Then they told us they had used ChatGPT to look for an AI partner before meeting us. Abilytics had not come up.
“My first instinct was to ask what they’d typed. The question I’ve kept coming back to is what they were asking before they knew they needed a partner.”
That is the part of Answer Engine Optimization (AEO) I am most interested in. If a buyer asks for help with a problem your team knows how to solve, could the answer point them to something useful on your website? You could earn their attention before they have started making a shortlist.
- Pre-vendor discovery: Enterprise buyers consult ChatGPT and LLMs to define engineering architecture, evaluate rollout constraints, and diagnose problems weeks before issuing an RFP or compiling a vendor shortlist.
- Beyond keyword spending: LLM answers synthesize crawlable engineering depth, technical guides, and concrete post-launch trade-offs rather than high-level marketing claims or paid keywords.
- Dual-stage measurement: Effective AI visibility tracks both stages: advice-stage citation (answering how to solve the problem) and partner-stage shortlist recommendation (accurately describing what your team delivers).
The same question can produce a different shortlist
I do not know exactly why Abilytics did not show up for that customer. There are plenty of possibilities. The exact wording matters. So does whether it was the first question in a fresh chat or a follow-up after they had spent twenty minutes describing their project. The model, plan, workspace settings, and available sources can affect what ChatGPT can use. OpenAI also explains that ChatGPT can turn a prompt into several web searches and use relevant memories to help formulate them.
That makes the familiar paid-search question, “How much did they spend on that keyword?”, a lot less useful here. I want to know what the buyer was trying to decide, what sources the answer drew on, and whether there was any public evidence that gave it a reason to include us.
One of my KPIs over the next few months is to get Abilytics onto more ChatGPT shortlists for agentic AI delivery and AI systems integration. On a broad question about the top AI partners, we are competing with the Big Four, global integrators, and thousands of specialist firms. It is a hard target. The work gets more concrete when we look at the questions our ideal customers actually ask.
1. Start with the problem your buyer is trying to solve
A buyer might eventually ask for the “top agentic AI companies.” Earlier in the conversation, their questions might sound more like these:
“Our AI agent works in a demo. How do we get it into production if it needs access to internal systems and some actions require human approval?”
“Our support agent answers most questions well. How do we test whether it knows when to hand off to a person?”
“We can build the pilot ourselves. What would an embedded AI engineering team help us get done, and what would our team still own?”
The first buyer needs a way to plan a rollout. The second needs examples of what to test. The third is evaluating a delivery model. A generic page about agentic AI can mention all three and still leave each buyer with the same question they started with.
I would start with sales calls, architecture reviews, and RFPs. Listen for the moment someone says, “Here is where we are stuck.” Write down the question in their words, including the constraints. That is a question worth testing in ChatGPT and answering on your site.
2. Give ChatGPT a reason to recommend you when the buyer asks for advice
Imagine a buyer asks how to test when a support agent should hand off to a person. ChatGPT points them to an Abilytics guide that walks through a few scenarios. What should the agent do when it finds conflicting policy documents? What if the customer needs an exception the agent cannot approve? What context should the agent pass to the human taking over?
The buyer gets something they can use that afternoon. They also get a glimpse of how our engineers think. If they later need a team to work alongside theirs, they have a reason to look at how our forward-deployed engineering (FDE) pods work.
Anthropic’s engineering team publishes the kind of material I mean. In its guide to agent evaluations, it uses a flight-booking example. An agent might say it booked a flight. The evaluation can check whether the reservation actually exists. That one distinction could change how a team tests its own agent.
Before publishing, I would ask whether an engineer would send the page to a colleague who was stuck. If the answer is yes, you have given an AI answer a reason to point someone there, too.
3. Put the detail that impressed the customer into the case study
A lot of AI case studies end right where I want them to get specific. The pilot worked. Great. What happened when the first live workflow failed? Who decided which actions needed approval? Who owned the evaluations after launch?
OpenAI shared an example of its forward-deployed engineers embedding with a chip verification team. They worked with the customer’s engineers to build evaluations for whether the AI could generate useful tests and diagnose chip failures. You can picture what the embedded team did and why working alongside the customer mattered.
That is the level of detail I would like us to share in more Abilytics case studies, with customer approval. What did the pod take on? What decisions stayed with the customer’s team? What could that team run on its own after handover? We have a few examples on the Abilytics website that show what I mean.
Make that evidence easy to find and crawl, then see what the answers actually cite. Reddit and Quora may come up for some questions. For others, an engineering guide, a technology partner directory, or a customer story might be more relevant. Looking at the sources gives you a practical place to start. Maybe a competitor explains a trade-off your site barely mentions. Maybe your best example is still in a private sales deck.
4. Measure the advice and the shortlist
I want to track two moments in the buying journey.
When someone asks how to solve the problem. Does ChatGPT cite or recommend a page we have published?
When someone asks who can potentially help. Does it recommend Abilytics, and does it accurately explain what we do?
I am not chasing to be the best or fastest or cheapest in the searches.
Practitioners are already measuring pieces of this. Ahrefs compared how often AI platforms mentioned its brand with how often they linked to its website and found considerable differences across platforms. It also asks new customers where they heard about the company, which can surface discovery that a referral report misses. Bing’s Topics Portfolio beta groups AI citations by topic, so publishers can see where their content is being used across Bing’s supported AI experiences.
You can start with a handful of real buyer questions. Ask for advice in a fresh chat. Then add the detail the buyer usually shares ten minutes into a call and ask which firms could help. Record the exact prompts, prior messages, model, memory settings, and whether web search was used. Repeat the exercise over time. A citation, a recommendation, and an accurate description of your capabilities each tell you something different.
At Abilytics, we bring together prompt audits, technical SEO, content strategy, and AI visibility monitoring. The details of how we run and prioritize that work are part of our IP. We are still learning, and applying it to our own agentic AI and FDE practice gives us a way to test what we can take back to customers.
By the end of it, your customer should be able to see what you bring to the table even before the first demo.
